Patent Application
20150066533
The present disclosure relates systems and methods for detecting and monitoring patient conditions in clinical medicine settings.
Patient care in a hospital setting involves a complex management process because healthcare workers address multiple patient issues simultaneously. Decisions about patient priority and care made by the healthcare workers are subjective to some degree and may vary depending on the level of expertise and experience of each person involved in patient care. In addition, patients' complaints and symptoms are often complex, because a disease process may have its own associated complications, and a disease may also affect other concurrent patient conditions. Patients may also bring with them a degree of subjectivity in describing their symptoms, which may generate both variable indications of clinical conditions.
Uncontrolled complexity is a cause of large numbers unnecessary death in hospitals. Unfortunately hundreds of common but subtle modes of failure which lead to complications and death can potentially occur with every patient in the hospital. However, present hospital patient monitoring devices are entirely insufficient relative this level of complexity. There is an acute need for a quantum advance in patient data processing which is capable of managing the actual pathophysiologic complexity present. Without such technology vast numbers of unnecessary deaths can be expected to continue in hospitals, unabated and worldwide.
One common example of the challenges involved in detecting complex patient conditions, is undetected septic shock. Whether or not a given patient with an infection progresses to shock often depends on a complex relationship of patient-specific physiologic responses to immunologic and inflammatory perturbation as well as the physiologic state of the patient at the onset and during the perturbation and the timeliness and adequacy of intervention (e.g. antibiotics and/or fluid). These factors interact to define the dynamic state of the patient. This level of complexity, evolving as a “silent” mechanism of death on a busy hospital ward, represents a major threat to patients worldwide.
Advantages of the present disclosure may become apparent upon reading the following detailed description and upon reference to the drawings in which:
One or more specific embodiments of the present disclosure will be described below. In an effort to provide a concise description of these embodiments, not all features of an actual implementation are described in the specification. It should be appreciated that in the development of any such actual implementation, as in any engineering or design project, numerous implementation-specific decisions must be made to achieve the developers' specific goals, such as compliance with system-related and business-related constraints, which may vary from one implementation to another. Moreover, it should be appreciated that such a development effort might be complex and time consuming, but would nevertheless be a routine undertaking of design, fabrication, and manufacture for those of ordinary skill having the benefit of this disclosure.
The present disclosure provides systems and methods for diagnosis, monitoring, and treatment of certain clinical conditions. A processor-based system may characterize and quantify patient physiological conditions by analyzing data relating the patient. In one embodiment, a database of is converted into a format which is favorable for searching and/or analysis of complex relational patterns of trends and variations. One such format is a time series matrix which itself may be formatted to generate an image or moving image of the abnormal components of the time-series matrix that may be further processed into operator-interpretable data.
In one embodiment the time-series matrix constructed by the processor may include a plurality of (e.g., hundreds or thousands) of individual time-series each including different chemical, electrical, mechanical, and/or state related parametric and/or non-parametric values of an individual patient. In one embodiment, data derived from the patient is organized into a comprehensive set of time series each aligned along a single time axis. Each individual time series is two-dimensional with one dimension being time but there are no other limits on second dimensions so that the matrix may have thousands of other dimensions in addition its unifying time dimension. For example the dimensions may include derivatives and frequency measures or calculations of values, calculated or measured relationship between a pluralities of values, time relationships between a plurality of values to name a few. In a few more examples, the dimensions may be further defined by a instantaneous magnitude value, a moving window derived averaged magnitude value, an instantaneous slope of any of the measured or calculated values, and moving window average of the slope of any of the measured values, an acceleration value or area related value, a peak or nadir value, a difference value, a recovery value, a threshold breach, a threshold approximation, a statistical parameter or value, a derivative value, a trend value, frequency domain derived value, to name a few.
One embodiment provides visibility into the complete matrix of patient data by direct search of a hospital's global and/or centralized medical records which may be automatic on a scheduled regular basis, continuous, triggered, or manual basis. Direct search allows rapid access to information relevant to the time and relationships involved complex failure cascade images or other patterns searched for, including outputting an identification of the patient exhibiting the image. The output can comprise a display located at the nurses station or another location and can include direct notification of the patient (as through a pager or phone attached to the patient, nurse or physician (which may be is capable of presenting the image for review). The detection therefore can occur at the central database repository at a central location for the hospital, hospital system, a region, or the entire world, with the provision of direct notification to the hospital, the caregiver, or the patient him or herself.
In one embodiment, for manual searching and/or for preparing the automatic search process a health worker or researcher is presented with a single search box in which the name of a condition, pattern or other occurrence can be entered. The health worker or researcher can then direct the patient safety processor to execute the search. Search results are presented in a paged display allowing for rapid scan. With each “hit” is presented a link to navigate into a screen of data time-relevant to the “hit” and data relevant to the occurrences that constitute the hit. This may include time series displays, occurrence stream displays, patient data, data with regard to the definition of the occurrence and/or constituent occurrences to name a few. A few of the characteristics of the search which can be used to narrow the view of the condition, pattern or other occurrence are listed below. The processor may be programmed to navigate the search using these characteristics. These will be discussed in more detail in the subsequent disclosure.
Time relevance—the understanding of the relationships within the specified occurrence provides the information to display a series of time spans based on the overall evolution of the occurrence within the organism. One embodiment of the patient safety processor search provides rapid switch between these time horizons to provide both a macro and micro-view of the evolving pattern
Data relevance—using the definition of the specified occurrence one embodiment of the patient safety processor search system may display the channels of data relevant to the occurrence. These channels of data include channels in which related occurrence appear and also channels in which occurrences do not appear that would be expected. The relevant data set is closely related to the time relevance and therefore is coordinated with the time relevant horizons.
Statistical relevance—One embodiment, using the statistical information, gathered, for example, from local sources and optionally from centralized image data, the statistical relevance of the occurrence may be provided. If the data is real-time data then probable paths of evolution may be presented as well as alternative channels of data that can be executed to further identify the occurrence in question.
Meta-relevance—combining the statistical information with the meta-system of all possible patterns/images the patient safety processor may indicate alternative interpretations of the data, candidate occurrences as well as “what-if” scenarios in which the health worker or researcher indicates certain data points (lab results, etc.) as suspect or provides “best guesses” on missing data.
The health worker or researcher may include filters in the search—such as time horizons, patient populations, geographic locations or occurrence categories to name a few. Occurrence topologies may be created to assist in the filtering. These filters may be pre-defined to provide typical focus areas or may be defined by the health care worker. Advanced Boolean combinations may be provided as well as simpler mechanisms.
Search results may be aggregated by specific levels of granularity (e.g. patient, patient stay, geographical location to name a few). If data is aggregated then the health-care, once a “hit” has been selected, can choose a single occurrence from the aggregation or can choose to see a multi-occurrence display (e.g. within a patient stay).
Further, an aggregation may be selected and then searched within. In this way the health-care worker can iteratively narrow her search until a specific occurrence or occurrence group is selected for display.
The health worker may navigate out of an occurrence or occurrence group display back into the search results. Search or simple time/data selection may provide entry, into the matrix. Once the matrix has been entered the processor display may center (if desired) on a particular occurrence. The relationships maintained within the processor for this occurrence may provide a wide range of navigation. The health care worker may choose to navigate down into the constituent elements of the occurrence. At each level, the display may be configured to the desired relevance. The health care worker is thereby provided with one alternative mechanism of essentially “zooming in” on a particular aspect of the image. The health care worker may also choose to “zoom out” by navigating up to an occurrence (or occurrences) to which the current occurrence is a constituent part. In an example detected fall event in bicarbonate may comprise the occurrence from which manual or automatic navigation zooms out to view an entire sepsis cascade of which it is a part. The higher view may show a high-level image evolution. Lateral movement may occur not only by simple panning (e.g. sliding forward/backward in time) but also by navigating to the next similar occurrence or the next stage of image evolution within the current occurrence. Navigation may also be through the meta-data. This allows the health worker to navigate to similar images to the ones that actually have been identified. These images can be super-imposed for comparison. For example, the health care worker may request to see images that are statistically known to be in the evolution path of the current image. By navigating to expected (but not yet realized) evolution paths the health care worker can anticipate the development of perturbation and see expected reactions to treatment.
Further, within an occurrence or occurrence group display the health care worker or researcher will be presented with opportunities to search based on elements within the display. For example, if an image contains a specific event, the health-care worker may search for other occurrences of that event within a timeframe or other search horizon (e.g. related to a specific physician).
The spin-off search may be modified in one or more ways from the element from which the search was constructed. For example, the health care worker may select a threshold violation and indicate search, but allow a slightly higher or lower threshold. Or, as another example, a binary may be selected but the requirement of the time between occurrences may be relaxed.
In effect, search permeates the patient safety processor system allowing for any element or meta-element to become the launch-point for a patient safety processor search. In this context the health worker or researcher can rapidly follow his or her intuition regarding the nature and evolution of a condition of interest allowing the patient safety processor to provide substantially immediate aggregations of data in views relevant to the relationships within the focus data, the overall statistical relationships in the overall available body of research and the intuition of the current health worker. The processor can be programmed to automatically search using one or more (or another type) of narrowing characteristics. For example the processor could be programmed to, upon the detection of a trend rise in ventilation of a given patient, search for other trends, breaches, events, or images in time relevance to the detected rise on and to determine if an image (such as a cascade) or an event (such as a drug infusion or a fall in bicarbonate) is present and where in combination with the rise in ventilation a new image is detected.
The individual time series of the matrix may extend the entire length of the matrix for example as the weight of the patient or ejection fraction of the left ventricle of a patient or may be transient, as for example with a single injection of an IV narcotic or may be intermittent, as with a time series of the peak (recovery) SPO2 values within sleep apnea clusters. In one embodiment the individual time series are linked to each other by processing to convert the raw time series into time series of events (such as objects by the process of objectification) but another method may be used. Linkage by event processing and/or objectification produces a comprehensive (potentially omni-dimensional) objectified matrix of times series of data relating to a patient, extending along a central common time axis with extensive cross-linkages, series that are not time series (for example a series of frequencies of a parameter detected at a single point in time) may also be objectified, but in this case this type of series extends perpendicular to the time axis of the matrix at the time (or window of time) of the occurrence of that series, series which are not time series have the same time designation for the entire series and, like time series, they may be linked with any other time series or other series in the matrix.
In one embodiment events, relational events, and aggregated events are defined to construct images of physiologic failure and care. Identifications of modes of physiologic failure and care by the analysis of the images provides for earlier recognition and intervention and improved protocolization of testing and treatment. A processing network provides for the development of extensive archives of images of physiologic failure and care to provide for processed integration of the international experience into the image recognition protocols to enhance and accelerate real-time physiologic image recognition and to improve the cost effectiveness of testing and treatment including those hospitals in remote and/or less well served communities.
One embodiment includes a real-time processing method for searching for and detecting physiologic occurrences, physiologic failure and/or care having steps of: (1) converting medical records into at least one time series matrix of a particular configuration suitable for imaging (for example the configuration may be a 2 or more dimensional configuration, and/or a 2 or more dimensional temporal configuration, and/or 2 or more dimensional frequency configuration, and/or another configuration suitable for imaging.); (2) imaging the matrix to detect at least one image indicative of physiologic occurrences, physiologic failure, and/or care (which can for example include a sepsis cascade pattern, a sepsis shock pattern, a drug reaction pattern, to name a few); and (3) taking action based on the detection of the image which can for example, include outputting an indication (which can be an alarm) of the image and/or the likely cause of the image.
In one embodiment the time series matrix is processed (and this processing may be provided as part of the construction of the time series matrix to generate and time series matrix of events. In an embodiment these events are objects (such as objects having a relational hierarchy by the process of time-series objectification) thereby rendering an objectified time series matrix. The raw time series matrix and/or the time series matrix of events and/or the objectified time series matrix may be rendered in a particular configuration such as a particular default digital spatial and/or temporal and/or frequency and/or statistical configuration (to name a few), and then digitally imaged and the image characterized by known digital image recognition methods to detect a pattern and/or a plurality of patterns. These image patterns may be defined along the configuration of the matrix by selections of the user, by methods of digital image recognition, by statistical processing, as by neural net processing or by other methods as for example the methods described and/or listed in U.S. patent application Ser. No. 11/351,449 of the present inventors, the entire disclosure of which is hereby incorporated by reference for all purposes as if completely disclosed herein.
One embodiment includes a real-time processing method for detecting physiologic occurrences, physiologic failure and/or care having steps of: (1) converting medical records into at least one time series matrix of a particular configuration suitable for imaging; processing the matrix to render an matrix of time series of events (such as an objectified time series matrix (for example the configuration of the objectified time series matrix may be a 2 or more dimensional spatial configuration, and/or a 2 or more dimensional temporal configuration, and/or 2 or more dimensional frequency configuration, and/or another configuration suitable for imaging.); (2) imaging the time series matrix of events to detect at least one image indicative of physiologic occurrences, physiologic failure, and/or care (which can for example include a sepsis cascade pattern a sepsis shock pattern, a drug reaction pattern, to name a few); and (3) taking action based on the detection of the image which can for example, include outputting an indication (which can be an alarm) of the image and/or the likely cause of the image. Steps one, two and three, can be combined so that the time series matrix is built and objectified simultaneously or the time series are objectified and/or imaged and then the matrix is built and the matrix is then objectified and/or imaged. A default configuration of the raw time series matrix may be displayed directly with the images detected in raw form along the default configuration (as may be over-read by physicians or nurses). Alternatively, or in combination, another default configuration may be displayed for the user with the matrix presented in a processed (such as objectified) form with the detected images being highlighted or represented or replaced as icons, motion pictures, or other visual images. In another embodiment the detected images are reprocessed to simplify them and these reprocessed images may be digitally imaged to detect larger and more complex images. In this way the images themselves (much like the objects in the matrix) may have an inheritance hierarchy which reduces the complexity of the digital image recognition of the larger, complex and prolonged images.
One embodiment includes a real-time processing method having steps of: (1) converting medical records into a predetermined format for searching, as for example at least one time series matrix; (2) defining events such as objects (which can be relational events) along the time series matrix; (3) defining patterns including combinations of events, (4) using a processing search engine, searching for the events and/or the patterns and detecting at least one complex pattern or image (which can for example include a sepsis cascade pattern, a sepsis shock pattern, a drug reaction pattern, to name a few); and (4) taking action based on the detection which can for example, include outputting an indication (which can be an alarm) of the pattern(s) and/or the likely cause of the patterns.
While an embodiment event identification and processing using objectification of time series matrices presently contemplated there are many alternative processing methods which can be incorporated to define events or occurrences, such as perturbations, trends, variations, or threshold violations, to name a few along the time series and along the time series matrix to be identified by a waveform search engine (such as the imaging processor). For example the time series matrix or the “event, trend, or perturbation processed or formatted” (such as objectified) time series matrix may be rendered in a particular configuration such as a spatial configuration, and then imaged and the image characterized by image recognition methods to detect a pattern. These image patterns may be defined along the configuration of the matrix by statistical processing, as by neural net processing or other methods. Also the events, objects, or and/or images may be processed using a wide range of search engine types and may be incorporated into larger computer search engines.
Events may be defined by a wide range of methods, for example, by objectification, by probabilistic discovery, by neural net processing, by peak and trough detection, adaptively, by objectification, by a specific rules set, or by a single or combination of signal processing and/or characterization methods in the time domain and/or the frequency domain and/or by other methods.
According to one embodiment, provided herein is a processing method and system for characterizing and quantifying complex layered non-linear systems such as physiologic systems, using a matching layered processing architecture, termed an “objectified time series matrix.” Although the objectified time series matrix is applicable to a wide range of signal processing environments, in an example, the matrix is applied to organize, control, characterize and quantify substantially all aspects the electronic medical records (EMR) of a patient and is employed as a real-time patient safety processor. Here, the matrix is including hundreds or thousands of parallel and, in some cases, perpendicular series (and of segments of parallel time and perpendicular series) derived from the EMIR which are programmatically bonded together in a manner to produce an embodiment which may provide real time programmatic images which organize the complexity of the evolving processes indicative of the global and regional state of health and disease.
In one embodiment such a matrix is including hundreds or thousands of individual time-series each including different chemical electrical, mechanical, and/or state related parametric and/or non-parametric values of an individual patient. In one embodiment all data derived from the patient is organized into a comprehensive set of time series each aligned along a single time axis. Each individual time series is two-dimensional with one dimension being time but there are no other limits on second dimensions so that the matrix may have thousands of other dimensions in addition its unifying time dimension. For example the dimensions may include derivatives and frequency measures or calculations of values, calculated or measured relationship between a pluralities of values, time relationships between a plurality of values to name a few. In a few more examples, the dimensions may be further defined by a instantaneous magnitude value, a moving window derived averaged magnitude value, an instantaneous slope of any of the measured or calculated values, and moving window average of the slope of any of the measured values, an acceleration value or area related value, a peak or nadir value, a difference value, a recovery value, to name a few. The individual time series of the matrix may extend the entire length of the matrix for example as the weight of the patient or ejection, fraction of the left ventricle of a patient or may be transient, as for example with a single injection of a IV narcotic or may be intermittent, as with a time series of the peak (recovery) SPO2 values within sleep apnea clusters. In one embodiment the individual time series are linked to each other by objectification but other methods may also be used. Linkage by objectification produces a comprehensive (potentially omni-dimensional) matrix of times series of data relating to a patient, extending along a central common time axis with extensive cross-linkages, series that are not time series (for example a series of frequencies of a parameter detected at a single point in time) may also be objectified but in this case this type of series extends perpendicular to the time axis of the matrix at the time (or window of time) of the occurrence of that series, series which are not time series have the same time designation for the entire series and, like time series, they may be linked with any other time series or other series in the matrix.
Also provided herein is an objectified time series matrix of patient related signals. A comprehensive matrix derived from a complex patient would appear relatively opaque when viewed with hundreds or even thousands of parallel, objectified time series with various objectified linkages. Provided herein is a processing system and method for imaging the objectified matrix of patient signals to render motion pictures of physiologic failure. According to one embodiment, the construction of a matrix of time series for the organization and analysis of non-linear systems is accomplished by: (1) generating a large set of time-series of data relating to the nonlinear system; (2) converting the datasets into parallel time series; (3) identifying occurrences along and between the time series; (4) aggregating the occurrences into a real-time hierarchal data matrix; and (5) analyzing specific nonlinear processes using the data matrix.
According to one embodiment, the construction of an objectified matrix of time series for the organization and analysis of non-linear systems is accomplished by: (1) generating a large set of time-series of data relating to the nonlinear system; (2) converting the datasets into parallel time series; (3) objectifying the time series; (4) identifying occurrences along and between the objects along the objectified time series; (5) aggregating the occurrences into an image including a real-time hierarchal data matrix of linked objects; and (6) comparing the image to other images and/or to other values or outcomes, or images to determine the significance of the image.
One set of steps applied in building and using a time series matrix of patient related signals to detect physiologic failure includes: (1) conversion of the data set of the electronic medical records into time series sets, (2) linking at least a portion of the series sets; (3) defining controllable micro-domains along and between the series sets; (4) analyzing these micro-domains for occurrences, which may for example be a primary or relational perturbation having a specific primary or relational slope, amplitude, polarity, state, acceleration, frequency, pattern, value, or other characteristic to name a few; (5) aggregating the occurrences rendering at least a programmatic image of the aggregated occurrences; (6) comparing the image to stored images indicative of disease; (7) taking action based on the detected image; and (8) outputting an indication based on the detected image.
In one embodiment all data relating to a given patient are converted into time series, parallel in time but not necessarily all comprehensively covering the complete time span evaluated. The objectified time series matrix is constructed by adding each new occurrence in a manner, which defines an inheritance-based hierarchy of ascending complexity. The objectified matrix is progressively built to transform the electronic medical records into a highly organized data structure from which may be derived real-time motion pictures of physiologic condition, which allows early detection and intervention.
According to one embodiment, a patient safety processor including an objectified time series matrix is provided by: (1) generating a large set of time-series of data of a patient including at least data relating to the physiologic state and/or care of a patient; (2) converting the datasets, including at least the monitored datasets and laboratory datasets into parallel time series; (3) objectifying the parallel time series; (4) converting the objectified time-series into micro-domains; (5) identifying occurrences along and between the micro-domains; (6) aggregating the occurrences into a real-time hierarchal patient safety data matrix for the generation of motion pictures of the organized occurrences; and (7) recognizing and interpreting along the motion pictures, specific pathophysiologic processes or other adverse processes defined by the motion picture.
According to one embodiment, a patient safety processor including an objectified time series matrix is provided by: (1) generating a large set of time-series of data of a patient including at least data relating to the physiologic state and/or care of a patient; (2) converting the datasets, including at least the monitored datasets and laboratory datasets into parallel time series; (3) objectifying the parallel time series; (4) placing the objectified time series into an objectified time series matrix; (5) converting the objectified time-series into micro-domains; identifying occurrences along and between the micro-domains; placing each stream of identified occurrences back into the objectified time series matrix so as to progressively and iteratively increase the objectified time series; analyzing the completed objective time series matrix for the identification of motion pictures of the organized occurrences; and recognizing and interpreting along the motion pictures, specific pathophysiologic processes or other adverse processes defined by the motion picture.
According to one embodiment, a Patient safety processor (patient safety processor) including a time series matrix is provided by: (1) generating a large set of time-series of data of a patient including at least data relating to the physiologic state and/or care of a patient; (2) converting the datasets, including at least the monitored datasets and laboratory datasets into parallel time series; (3) identifying relational patterns along a plurality of time series which is indicative of failure cascade such as, for example, a sepsis cascade, a pulmonary embolism cascade, a metabolic cascade, and a microcirculatory cascade to name a few; (4) identifying occurrences such as, for example, inflammatory occurrences, metabolic occurrences, volumetric occurrences, hemodynamic occurrences, therapy occurrences, hematologic occurrences, respiratory occurrences (to name a few), and the timing of the occurrences which relationally or collectively are indicative of at least one failure cascade such as, for example, a sepsis cascade, a pulmonary embolism cascade, a metabolic cascade; and (5) identifying and outputting an indication of the cascade, the timing and type of the occurrences along the cascade, and length of the cascade.
According to one embodiment, a patient safety processor including an objectified time series matrix is provided by: (1) generating a large set of time-series of data of a patient including at least data relating to the physiologic state and/or care of a patient; (2) converting the datasets, including at least the monitored datasets and laboratory datasets into parallel time series; (3) objectifying the parallel time series; (4) placing the objectified time series into an objectified time series matrix; identifying occurrences such as, for example, inflammatory occurrences, metabolic occurrences, volumetric occurrences, hemodynamic occurrences, therapy occurrences, hematologic occurrences, respiratory occurrences (to name a few), and the timing of the occurrences which relationally or collectively are indicative of at least one failure cascade such as, for example, a sepsis cascade, a pulmonary embolism cascade, a metabolic cascade; (5) identify and output an indication of the cascade, the timing and type of the occurrences along the cascade, and length of the cascade.
According to one embodiment, a patient safety processor may generate a time series matrix for the early detection of septic shock or the pre-septic shock state by: (1) generating a large set of time-series of data of a patient including at least data relating to the physiologic state and/or care of a patient; (2) converting the datasets, including at least the monitored datasets and laboratory datasets into parallel time series; (3) identifying occurrences such as, for example, inflammatory occurrences, metabolic occurrences, volumetric occurrences, hemodynamic occurrences, therapy occurrences, hematologic occurrences, respiratory occurrences (to name a few), and the timing of the occurrences which relationally or collectively are indicative of the septic shock or pre-septic shock failure cascade; (4) identify and output an indication of the septic shock or pre-septic shock failure cascade; (5) identify and output the relational timing of the inflammatory occurrence, the hemodynamic occurrence, and the respiratory occurrence along the cascade; (6) identify and output length of the cascade; and (7) quantify the cascade as the cascade evolves and output a time series of the severity of the cascade.
According to one embodiment, the process of moving from a massive set of time-series data to the recognition of, and interpretation of a motion picture indicative of a patient's comprehensive state of physiologic failure and treatment includes the isolation, from the global domain of data, individual sets of micro-domains in which the complexity is manageable enough for hum an researchers and/or software agents directed by human researchers to define and characterize specific properties and relationships (occurrences) along and between the micro-domains which are perceived to be, may be determined to be (and/or are statistically verified to be) associated with clinical conditions. The Patient Safety matrix is then built by combining the occurrences. This process of matrix building from different micro domains is repeated over and over to build an extensive matrix which is including occurrences ordered into relational timed images having an inheritance based hierarchy. The motion picture is a timed aggregation of those images from the matrix. The motion pictures range from isolated adverse occurrences to simple physiologic failures and finally to catastrophic pathophysiologic cascades. The global “motion picture” output of the patient safety processor is a comprehensive flow of the detected images in combination as they evolve.
One embodiment of the present techniques provides a system and method for the characterization, investigation, and analysis of the complex environment of non linear relational signals (such as physiologic signals) through the repeated and layered use of micro-domains, which, provides a mechanism such that the micro-domains may be modeled by a human researcher with an expert understanding of the physiological and systemic relationships to render detected images which are added to the objectified matrix. One embodiment provides a mechanism by which both the scope and the variables available within the micro-domains that exist within a set of physiological signals may be identified, modeled, analyzed, characterized, compared and statistically investigated. In this embodiment, the process follows four basic steps: (1) Establish the type and scope of a candidate micro-domain; (2) Calculate and characterize all or various relationships, sub-elements, values, variables, properties within the candidate micro-domain; (3) Refine the scope of the candidate micro-domain given the newly understood relationships, sub-elements, values, variables and properties; and (4) Determine whether the candidate micro-domain meets criteria to be specified as a true occurrence of the micro-domain type considered.
Once micro-domains are established as objects within an object stream, the micro-domains themselves become available to be aggregated to establish scope. In other words, the scope of a micro-domain may be defined as a set of durations (and/or points) along a time series, the aggregation of other micro-domains or the combination of the two. In one embodiment this 4-step process provides symmetry of scale across levels of the analysis by which the patient safety processor uses the same approach of analysis at each level of pre-objectified and objectified data—starting with raw pre-objectified time series and going up.
Alternatively, the data from the objectified time series matrix may be converted into images, such as moving pictures. In one embodiment, data from the electronic medical records and patient monitors are used to generate graphical displays, which may include moving pictures of the patient condition. In an embodiment, such moving pictures, or animated displays, may be referred to as “motion pictures of physiologic condition” (MPPC). Provided herein is a processing system and method for generating real-time MPPC of clinical data. The data and/or images may also be analyzed to detect perturbations, aggregate and cascading perturbations, perturbation relationships, physiologic responses to perturbations, treatments associated with the perturbations, physiologic responses to the treatments, physiologic failures, testing failures, treatment failures, and communication failures to generate the MPPC. In addition, the MPPC may also include a graphical representation of any treatment applied in association with the clinical condition.
Once the image or moving image (i.e. an image that includes more data over time as the patient monitoring progresses) MPPC of the patient condition has been generated, this image may be further processed to create an operator-interpretable indicator to assist in patient diagnosis and/or treatment. For example, the image may be directly compared to a database of similar images taken from patients with clinically confirmed diagnoses. The database image or composite of multiple images with the greatest similarity to the generated image may indicate the correct diagnosis for the patient. For example, if the generated moving image, particularly as the image progresses over time, has the greatest similarity to a database image indicating, for example, “septic shock cascade”, “inflammation failure”, “pulmonary thromboembolic cascade”, “hemorrhagic failure cascade” to name a few, a processor may generate a text or other indicator to a healthcare provider indicating such a diagnosis. The processor may also indicate that additional tests should be ordered to confirm the diagnosis. The processor may also indicate and/or provide orders for specific treatments in light of the diagnosis. In an embodiment, a moving image may be indicative of two or more clinical conditions. The processor may indicate tests that may rule out one or more of such conditions. In addition, over time, one condition may be determined by the processor to be more likely while additional time-series data may also rule out another condition.
These database images may be formed from retrospective clinical data. In an embodiment, the images may be analyzed for similarity by any suitable technique, including image registration. In embodiments, the matches may be made by image similarity measures that include cross-correlation, mutual information, sum of squared intensity differences, and ratio image uniformity. In an embodiment, the individual time-series objects that make up the image may be processed as a group for similarity to other groups of time-series objects associated with a particular diagnosis or clinical condition. The Motion Picture of Physiologic Condition (MPPC) may, for example, include abnormal and/or perturbed components and in particular “Motion Pictures of Physiologic Failure” (MPPF) of the physiologic system and of exogenous forces relating to that system. Provided herein is a processing system and method for generating real-time MPPCs of healthcare signals and processing those images to timely detect perturbations, aggregate and cascading perturbations, perturbation relationships, physiologic responses to perturbations, treatments associated with the perturbations, physiologic responses to the treatments, physiologic failures, testing failures, treatment failures, and communication failures to generate and then recognize motion pictures of physiologic failures and of the treatment applied in association with the failures.
Also provided herein is a processor and processing method for the automatic generation and/or analysis of the images of physiologic and/or clinical condition and the characterization and aggregation of the image components of complex dynamic systems, such as physiologic systems and medical care systems. The processing system may generate real-time MPPC of healthcare signals and processing those images to timely detect perturbations, aggregate and cascading perturbations, perturbation relationships, physiologic responses to perturbations, treatments associated with the perturbations, physiologic responses to the treatments, physiologic failures, testing failures, treatment failures, and communication failures to generate and then recognize motion pictures of physiologic failures and of the treatment applied in association with the failures. According to one embodiment, a processor first renders parallel time-series from each of a plurality of sensors and testing sources, which are applied to broadly monitor the dynamic system for failure. In an example, a processor programmed with instructions for time series objectification of patient data detects patterns along the parallel times-series, converts these patterns into time series of discrete objects, then organizes these objects into discrete relational objects (such as binary objects, or relational binaries, derived of relational object pairs). The processor then organizes the relational binaries to render a unifying programmatic image of the physiologic system and the care provided. The processor then automatically recognizes objects in the image components and may be able to perform analysis on the images.
One embodiment may include a patient safety processor having a single processor or a combination of processors programmed to generate time series objects, a relational binaries, moving images, patient safety images, and/or patient safety visualizations. The patient safety processor outputs images of the patient's physiologic system and medical care. In an embodiment, the processor includes processing functions for time series objectification, relational binary processing, and an imaging processing. In an alternative embodiment, the patient safety processor combines multiple processing mechanisms (e.g., time series objectification, relational binary processing, and imaging processing) into a single matrix construction processor.
According to an embodiment, perturbations detected by the processor are converted to image components that may be used to generate a moving image. In an embodiment, an MPPC may be representative of a “motion picture of physiological failure” (MPPF) when a failure image becomes progressively more complete and recognizable by the processor as each additional failure image component is added. One embodiment may involve building a dynamic real-time image of disease, injury, and/or drug reactions, the care provided, and the expense associated with that care. The image is initially associated with initial image components including one or more minor perturbations, which may for example be caused by circulation of one or more toxic, acidic, and/or immunogenic material of endogenous or exogenous origin. At first these perturbations, such as toxins, inflammatory and/or thrombogenic mediators, may induce and/or cause only minor changes in cell permeability, ion flux, or hydrogen ion elevation, and trigger various minor physiologic perturbations and responses each of which may produce an image component. The measurements of various mediators, ions, biologic profiles, as well as standard blood tests, and the outputs of vital sign monitors may begin to vary as a function of these early physiologic perturbations and responses, and it is these variations that enlarge the group of image components from which the larger image is derived. Early in the process, each of these alterations in permeability, cell injury, mediator production, and physiologic perturbations, when considered in isolation, are often minor. However, collectively they may represent the early manifestations of a nascent and evolving moving image of a serious clinical condition.
According to one embodiment, each perturbation is programmatically organized to form an image component of the MPPC. Many of these detected, images components may be isolated because they are related to a benign process, and the image may self-extinguish or may not develop into an image associated with a clinical condition involving intervention or an MPPC. Yet, as noted above, others may represent the first image components of an early moving image. Provided herein are systems and methods for the detection of the early image components of an evolving moving image to provide timely detection of physiologic failures before these failure progresses to shock (including, for example, hypovolemic, obstructive, septic, toxic, cardiogenic, hypoxic, and/or hypercarbic shock.) In one embodiment, it is advantageous to detect the early image components of the moving image before shock develops to improve the prognosis for the patient and to apply goal-directed therapy while clinical intervention is still beneficial.
According to one embodiment, a patient safety processor constructs a programmatic MPPC, which is used for dynamic, motion picture responsive, protocolization of care. This motion picture is comprehensive, including not only the events including a single or few parameters such as a the heart rate, but also other parameters that may include, for example: the slope and pattern of the heart rate, the slope and patterns of the systolic pressure variation, the slope and patterns respiration rate, the slope and patterns SPO2, the slope and patterns ventilation-oximetry index, the slope and patterns drug and fluid infusion rate, the slope and patterns blood pressure, the slope and patterns of the Neutrophil count, and the slope and patterns of inflammatory and/or thrombotic markers, and various other blood, urine and/or exhaled gas test to name a few. The signals from all of these sources may be converted to time-series and may, for example, be physiologic signals, therapy signals, laboratory signals, or historical signals, which may be objectified, as by an objectification processor, to produce the discrete programmatic objects (events). According to one embodiment, the processor detects a first discrete event that includes a pattern or value of at least one medical signal, and a second discrete event that includes a second pattern or value of at least one medical signal, the processor then aggregates at least the first event and the second event to produce a first relational object, the processor further detects a third event that includes a pattern or value of at least one medical, signal, and a fourth event that includes a second pattern or value of at least one medical signal, the processor then aggregates at least the third event and the fourth event to produce a second relational object. The first relational object and the second relational object are then aggregated to produce a first image component. Additional images are built accordingly and the image components are then aggregated according to the time of occurrence to derive the moving image and care.
In an example, the pulse related components of the typical motion picture of sepsis failure cascade would include occurrences such as early rise in heart rate, rise in pulse amplitude, and rise in slope of the pulse upstroke (as measured at the finger tip) in combination and typically proceeded by a brisk rise in inflammatory markers. In contrast the typical motion picture of occult hemorrhagic failure cascade (as for example due to heparin related retroperitoneal hemorrhage) would include occurrences of an early rise in heart rate, a fall in pulse amplitude, and a fall in slope of the pulse upstroke (as measured at the finger tip) and a rise in the respiratory related pulse pressure variation and a fall in hemoglobin. According to one embodiment, all of these occurrences along the image of an occult hemorrhagic failure cascade can all be derived from a multi wavelength pulse oximeter.
According to an embodiment, a relational binary processor is provided that divides detected variations into discrete alpha events and beta events, which are combined by the relational binary processor to construct the relational events which are termed relational binaries. These relational binaries are aggregated according to timing, frequency, and/or spatial relationship to construct images. These images are then further aggregated according to timing, frequency and/or spatial relationship to construct and progressively build MPPC (from which visual images or electronic representations may be derived as desired). These MPPC are often moving images of catastrophic cascading failures, thereby allowing more reliable detection to allow timely rescue of the patient.
The signals may be chemical or physiologic measurements, as provided by patient monitors, recorded in the electronic medical record, and/or may be biomarkers specifically ordered, either automatically by the processor or manually by the clinician to indicate the potential presence of the sepsis (as those, for example, disclosed in U.S. patent application Ser. Nos. 10/704,899, 11/647,689). The presence and/or concentration of such markers may be presented in the context of the MPPC with the timed positioning relative to the others parameters, which then allows the relevance of the biomarker to be much more readily identified. According to an embodiment, the temporal and relational pattern of inflammatory markers and temporal and relational patterns of contemporaneously measured or associated physiologic parameters are aggregated to produce a progressively enlarging MPPC of an evolving patient condition.
Therefore, to achieve the detection of various pre-shock states as well as earlier detection of failures, one embodiment detects early variations and aggregates them to provide an MPPC to dynamically present expanding failure cascades of pre-shock and shock states. This allows separation of expanding images from the smaller and less expansive image components having benign characteristics, and further allows separation of the images of minor isolated failures from failures that progress to generate an expanding MPPC heralding the potential for transition to one of the shock states. Each group of images as well as the complete MPPC and care may be analyzed for the purpose of assessing patient care in a hospital, a ward, or under the care of a given healthcare worker.
The occurrence of a large number of images indicating non cascading failures which self extinguish may be indicative of an unstable patient population or poor health care delivery. In the alternative, a large number of cascading failures are indicative of risk of injury. The MPPC and the images may be used to determine if that is due to the patient population or the quality of the care.
One embodiment detects failure cascades along with the determination of the specific fundamental perturbations, or treatments, or lack of treatments that occur early in a failure cascade. Specific fundamental failures are detected before they progresses to complex failures and particularly before they progresses to the pre-shock or shock state. Furthermore, the processor builds an image derived of the relational perturbations and treatments as the cascade expands. According to one embodiment, each time series is processed to separate expected events from unexpected events. The unexpected and/or abnormal events are then aggregated further to repetitively generate relational events, images and finally the MPPC which includes a motion picture of the cascade (if present) as well as the treatment applied in association with the cascade. This MPPC is further processed to allow the detection of the probable cause or causes of the occurrence of the moving failure images well as the images of the MPPC as it evolves thereby allowing detection of the nature and cause of the failure cascade.
As noted above according to one embodiment, an analysis is provided wherein the fundamental components of the analytic process include a basic relational variable that includes a plurality of events. In a contemplated embodiment, the basic relational variable is that includes two events (a relational pair) and this is called a relational binary. In one embodiment, the relational binaries are initially selected by the users as from a menu (or by a drag and drop interface) of relational binaries and/or of events from which the user builds the desired object binaries the binaries are then used as by drag and drop to build the definition of images for detection. This may be performed by, for example, by national or regional expert groups, or by specific departments in a hospital, or by an individual physician to provide custom management. This may also be automatically performed by the processor (as, for example, through the investigation of a large number of historical data sets that have been comprehensively analyzed and categorized according to outcomes. The objectified time series matrix and/or the MPPC may be may be outputted in various interactive, hierarchical, and relational, formats for review and automatic or manual adjustment. The MPPC may detect a wide range of failures, such as physiologic failures, treatment occurrence failures indicating the absence of expected treatment in relation to a given perturbation, testing occurrence failures indicating the absence of expected testing in relation to a given perturbation, treatment response failures indicating the absence of the expected correction of perturbation or the occurrence of a new potentially complicating perturbation in relation to a given treatment and/or dose.
The processor combines the complex data of the electronic medical record into a single motion picture of perturbations, treatments, physiologic responses, diagnostic testing, recoveries, diagnoses, missing data, patient locations, and/or other datasets. Dynamic images are generated of relational variations of a set of time series associated with a complex system to generate a real time motion picture of a failure of the system and/or of forces applied to the system. According to one embodiment, the patient safety processor automatically outputs a unified timeline, for example, derived of detected images of a given type. According to another embodiment, the processor, upon detecting a failure cascade, may present and highlight the evolving M-PC in real time on an outputted display of an image diagram for the physician to review. The portion of the motion picture, which has already been completed, may be reviewed backward and forward to review in a single summary snap shot view.
Many physiologic failures such as, for example septic shock, pulmonary embolism, congestive heart failure, respiratory arrest due to narcotics in the presence of sleep apnea, thrombotic thrombocytopenia purpura (TTP), hemorrhage due to anticoagulation, respiratory failure due to bronchospasm, and adult respiratory distress syndrome, but not limited to these clinical conditions, begin with one or two non-specific perturbation(s). Physiologic failure is commonly a relational expansion, often beginning with a fundamental physiologic perturbation at a single focal point in time. In fact, this initial perturbation is often completely masked once the cascade has progressed past a certain point. In such cases, testing or monitoring for the single perturbation may not be useful for making a diagnosis. In many cascading clinical conditions, the first perturbation(s) of the cascade may often only be detected in retrospect after the cascade has further progressed when the first perturbation(s) is no longer present. This provides a basis for optimizing the detection of the first point(s) by real-time imaging of the cascade as it develops and then examining the image to determine the first perturbation(s).
While a pattern of a single time series provides a larger image of a dynamic process than a single value or range, such a pattern is still only a tiny image fragment of the process. The determination of thresholds and even the detection of various patterns of perturbations include incomplete analysis, which will inevitably allow an unacceptable rate of progression to catastrophic failure. Even in situations wherein a measurement or test may seem definitive as a stand-alone test, action or conclusions based on a single value (or an average of a plurality of values) will have a reasonable probability of being incorrect. Consider, for example, a single measured spot SPO2 value of 94. This value is largely meaningless without knowing if the SPO2 is rising, falling, or cycling. Yet this infinitesimal image fragment of a patient's complex physiologic system is used everyday in hospitals to determine care. Furthermore, even if the pattern of the SPO2 is known (for example the SPO2 has been stable at about 94 for at least 12 hours) this is an incomplete image, which is largely useless and, in fact, a potentially misleading piece of information. Without knowing the relational pattern of the minute ventilation during the related time interval of the measured SPO2 pattern, the healthcare worker may be lulled into a false sense of security even as the patient is dying of septic shock or heart failure. Furthermore, an alarm or interpretive output which is based on a programmatic image of both the patterns of both the SPO2 and the related minute ventilation without additional relational elements of the image, such as, for example, the associated pattern of the white blood cell count, temperature, pulse, blood pressure, microbiologic values, and medications will be incomplete leaving too much synthesis for the healthcare worker. In another example, consider the detection of a pattern of a sustained rise in pulse or respiration rate. Each such pattern represents a tiny fragment of the present physiologic state and each pattern may be benign or alternatively may be an early image component of a much larger dynamic process of failure often associated with an evolving failure cascade. The difference between a benign or pathologic rise in pulse or respiration rate cannot be determined with this tiny image alone and often cannot even be known at the time of the onset of the rise. Therefore a tree diagram protocol with a branch based on a rising pulse or rising respiration rate adds a great degree of programmatic complexity with a high risk that the protocol will precede down the wrong pathway. An incomplete analysis of the physiologic system will often cause the healthcare worker to generate a large amount of investigation, testing, analysis and evaluation that is not necessary and therefore increases the cost of overall care. Further, these false paths of treatment and evaluation may distract the care worker from the determining the actual operative failure modes.
Prior to shock, a patient's physiologic system is perturbed by both disease and treatment. A given treatment provided to correct a perturbation might reduce the perturbation, have no effect on the perturbation, exacerbate the perturbation, cause another perturbation and/or make another perturbation worse or better. To determine which effect a treatment is having and to assure that this determination of treatment effect is complete, it is necessary to collect and, just as importantly, as provided by one embodiment, organize and analyze large amounts of relational data in a timely manner.
Another problem is that, within present hospital systems the healthcare worker is forced to do a great deal of archeology (digging, isolating, identifying, etc.) before synthesis may be effectively completed. For this reason, the synthesis of information by the healthcare worker is often not executed in a manner, which allows immediate searching, filtering, re-analysis, etc. This friction combined with the typical workload of healthcare workers limits the number and range of high-level scenarios, which may be investigated. Also the healthcare worker may, because of lack of available organized data and time, execute decisions without a complete set of synthesized information and worse, may not realize that this is the case.
For these reasons, even with conventional electronic medical record embedded protocols, patients remain subject to a range of failures across a broad range of failure modes based on the complexity of their individual condition and the complexity of the environment facing the care giver. In fact, because failures often overlap, one protocol may reduce the risk of one failure while increasing the risk of another. For example, oxygen given to treat hypoxemia under one protocol may delay the detection of pulmonary embolism by stabilizing the SPO2 and hiding the early signs of impending shock from the healthcare worker.
Because so many confounding and overlapping occurrences can be present, the time series objectification processor, the relational binary processor and imaging processor execute multiple iterations of analysis and refinement. In one embodiment this analysis would begins with a phase one execution in which each processor in order (time series, relational binary, and imaging) operates on the specified set of time series inputs storing the interim analysis results in memory and/or in the patient safety image database. After this has been completed, the patient safety processor may execute phase two preferably in the same order providing each processor with the original time series data as well as the full analysis from the previous phase(s). This second, phase may refine the analysis in terms of the first phase analysis. This process may contain as many phases as required for complete refinement of the analysis.
In one embodiment, each phase uses the same definition sets (event definition set, binary definition set, image definition set), but determines which property evaluations, rules and constraints are available per phase. The order of time series is determined to maximize the availability of constraints if a rule and/or constraint cannot be enforced in the current phase then it is ignored allowing for a relaxed set of rules to be executed and a greater number of objects to be identified. With each subsequent phase additional rules and constraints are applied as they become enforceable until all rules and constraints have been applied and all properties have been calculated, evaluated and assigned.
As an example, if an oximeter supplies a time series for oximetry and motion artifact, the multiphase analysis will allow the rule to reject an event on the existence of motion artifact unless the motion artifact is determined to present a particular density, pattern, frequency, magnitude, and/or have a relationship (as for example occurring at substantially the same time) with an arousal or other event or image. The definition of an arousal (an image) is created in terms of oximetry relational binaries and pulse events. Since the definition of an oximetry event depends on the existence of a higher-level object of which it will be an element, the patient safety processor recognizes that a single-phase approach will be inadequate and sets up a two-phase approach. Within the first phase the times series objectification processor creates oximetry events without the artifact constraint and the relational binary processor and image processor execute with the expanded set of oximetry events. In phase two, the results of phase one are available to the processors and the times series objectification processor may evaluate the complete rule.
If the processor determines that motion artifact of a particular density, pattern, frequency, and/or magnitude exists in relation to the point of time in which an oximetry event would be created the processor may query the phase one analysis for the given time window to determine whether an arousal exists. If an arousal does exist, then the motion artifact may be eliminated as rejecting the event and the analysis may continue as before. If an arousal does not exist, then the oximetry event may be rejected and, as the analysis continues, any higher-level objects that were dependent on the existence of that event will fail to be created. In one example, a micro-arousal (an image) may be defined as occurring when a new onset of motion occurs at a time when motion is not dense and near the end of a desaturation event and wherein the motion is temporally associated with a recovery from the desaturation. In addition the image of the micro-arousal may include a positive reciprocation of the in slope of the ascending portion of the pulse waveform, a positive reciprocation of the heart rate, a positive reciprocation of the pulse amplitude, to name a few events, binaries and/or images which may be included.
In an alternate embodiment, each phase has a specific group of definition sets (event definition set, binary definition set, and image definition set), which are constrained by the inputs of the phase. Within the multiphase approach of analysis the definition of events, relational binaries and images may be defined in terms of a related time series (e.g. an artifact indication stream), a derived/transformed time series (e.g. a time series derived from a calculation from two or more time series) and/or higher-level elements from a previous phase. In an alternative embodiment, post-processing phases are designed to refine resultant analysis. In this embodiment, rules are applied after the analysis to alter elements if they meet certain criteria. The patient safety processor would be supplied with a post-analysis rule set for identifying elements that should be added, altered or deleted. Using the above example, the event definition set used by the time series objectification processor will not include the constraint with reference to the motion artifact. Rather, the post-analysis rule set would include the rule to remove all events if motion artifact is found at the same time and the event is not part of an arousal. The post-processing execution would look for all events that meet this rule and mark them for removal. After all post-processing criteria have been evaluated the patient safety processor would begin the process of analysis alteration. In this embodiment, as well as other embodiments, each object provides the functionality to determine all objects upon which it is dependent down to the event level. In one embodiment, this functionality is accomplished through a recursive database procedure. Once all dependent elements are identified, each will be evaluated as to the action to take. For example, a relational binary will be removed if an event is removed, but a cycling relational binary may or may not be removed with the removal of a single event.
In an alternative embodiment, the construction of the objectified time series matrix is accomplished by a single matrix construction processor consuming two inputs—a set of raw time series and an occurrence definition set. In this embodiment, occurrence definition, including sub-elements of definitions, would contain explicit dependency requirements for construction. Dependencies would be defined in terms of raw time series required, occurrence stream required, and/or the specific sub-elements within, the specified time series and/or occurrence streams. Each occurrence definition would include a potential occurrence stream which may be added to the matrix. The matrix construction processor would examine each occurrence definition iteratively and in succession to determine whether the dependencies for the specified occurrence are available for analysis. If all dependencies are available then the matrix construction processor executes the necessary analysis to identify, qualify and completely construct all occurrences associated with the given occurrence definition. The resultant occurrences, if any, are aggregated into an occurrence stream and added into the matrix.
For example, an oxygen rise event may be defined as an occurrence with a dependency only on the existence of a raw oximetry time series. In this case, the matrix construction processor would be free to construct the oxygen rise occurrence stream and place it into the matrix as soon as a raw oximetry time series is added to the matrix. Once the oxygen rise occurrence stream has been placed into the matrix any occurrence definitions that are dependent on the oxygen rise occurrence (for example the oxygen reciprocation occurrence) may become avail able for construction. To provide additional flexibility within this embodiment, construction may be broken into the 4 stages described above: (1) type and Scope establishment; (2) property, relationship and sub-element creation; (3) Scope refinement; and (4) Occurrence qualification.
During the first stage a candidate occurrence is created. In one exemplary embodiment candidate occurrences are aggregated into stream s and placed into the matrix. Occurrence are not marked as qualified (e.g. true) occurrences until stage 4 of construction has been completed for the given occurrence definition. Construction may be interrupted between and within the stages as described above. If the dependencies required to accomplish stage 1 is available but the dependencies for stage 2 are not available, the matrix construction processor will proceed with stage 1 of construction but leave the subsequent stages for later processing. In one embodiment, candidate occurrences may be used within stage 1 construction of other occurrence definitions. In this embodiment, the matrix construction processor identifies all occurrences as being in one of four possible states: candidate, qualified, disqualified and suspect. Candidate occurrences are occurrences that have completed stage 1 but not stage 4 of construction. Qualified occurrences are occurrences that have met all the requirements specified in stage 4 of construction. Disqualified occurrences passed stage 1 but either explicitly failed stage 4 of construction or was one of its required sub-elements or scope objects was later determined to be disqualified or altered in such a way to fail scope requirements and/or qualification. Finally, suspect occurrences are occurrences that have passed stage 1 but one or more of its dependent sub-elements or scope objects has been determined to be disqualified or altered in such a way that further analysis may reveal it will fail scope requirements and/or qualification. The inclusion of the suspect state allows for flexible analysis execution in cases where candidate occurrences have been used by other occurrences to establish scope. In the case that the candidate occurrence is found to be disqualified, the matrix construction processor marks all occurrences which were dependent on the candidate occurrence as suspect. Suspect occurrences may be subsequently analyzed to determine whether the disqualification of the associated candidate occurrence (or alteration of a property on which the parent depended) does in fact disqualify the parent occurrence. This embodiment has several advantages, including the maintenance of a complete set of dependencies for occurrences, occurrence properties, relationships and sub-elements provides for additional information within the patient safety console. For example, if the patient safety console displays an index (for example, an instability index), the availability of the dependency tree allows the user to examine what elements go into the evaluation of the index and from which properties, relationships and sub-elements they were derived. In addition, evolutionary states of analysis may be persisted and analyzed by researchers and/or the patient safety processor to further understand the relationships within the resultant MPPC. Disqualified occurrences may be evaluated for research purposes and especially during the construction of occurrence definitions. Disqualified occurrences may be identified as “near misses” such that the researcher or the automated process within the patient safety processor may evaluate the result of changes within the occurrence definitions. The independent and iterative nature of the analysis approach along with the persistence of interim states lends itself to parallel, concurrent and/or distributed processing of the matrix. In other words, matrix construction may be executed within multiple threads of execution on multiple processors within a single machine or by any number of independent machines.
As discussed, according to one embodiment, the relational binary processor generates relational binaries. Such relational binaries are including an alpha occurrence object and a beta occurrence object. An early step in this process includes the defining the relational binaries by the user or by the processor. To define a relational binary, first, the alpha occurrence may be defined (as by the user or adaptively). The alpha occurrence may be, for example, an event defined both in terms of its channel and the object along the channel. In a contemplated embodiment, the objects along each channel are defined by characteristics (such as the slope, amplitude, or other features defining the object including as discussed in the aforementioned patent applications). As well, a beta occurrence defined as, for example, an event in terms of its channel and its characteristics. Alpha and beta occurrences may be events, other relational binaries, images, repeating images or pattern images to name a few.
The definition of events within the patient safety processor depends on the time series mode and type. The patient safety processor supports two time series modes: numeric and non-numeric. For each mode, there are several time series types. For the numeric mode, the patient safety processor supports several types that specify the type of data point that may be stored within the time series. Numeric types include integer, floating point, double precision, decimal, positive integers to name a few. Non-numeric types include Boolean, domain and freeform string to name a few. With numeric time series directional events may be defined in terms of characteristics of the segment including magnitude, duration, and slope to name a few. Threshold violations also may be identified as an event. With non-numeric types events may be defined as either a state matching event or a state transition event, directional events may also be defined in terms of best-fit or least-squared linear regression approach, by an imaging approach, a polarity defining approach, to name a few.
With state matching events a set of values may be defined (either by the identification of individual values or by a pattern-matching mechanism such as a regular expression) and optionally a minimum and maximum duration may be selected. An event may be said to have occurred if and when and for the length of time that the “points” within the time series match the target set for at least the specified minimum duration (if any) and for no greater than the specified maximum duration (if any).
State transition events may be defined by selecting two sets of values (either by the identification of individual values or by a pattern-matching mechanism such as a regular expression. An event may be said to have occurred if and when a first “point” is in the first set defined and the next subsequent point is in the second set defined. The patient safety processor provides for a unary “not” operator to handle two important specific cases of state transition. The researcher may define the first set specifically and then define the second set as being “not” the first set. In this way, an event may be created when a “point” is in the first set and the next subsequent point is in not in the first set. This is a specific type of state transition called a leaving state event. Similarly the second set may be specifically defined and the first set may be defined as “not” the second set. In this way, an event may be created when a “point” is not in the second set and the next subsequent point is in the second set. This is a specific type of state transition called an entering state event.
Alternatively, state transition events may be defined by specifying a state flow diagram or state machine definition and the beginning and end state for the transition. In this type of event, the transitional states are maintained as part of the state transition event as well as the order in which they occurred. Events may be individually defined with a statistical approach rather than an absolute or relative approach with any or all of the characteristics used for the definition. For example, a directional event may be defined as a deviation from previous (or subsequent) trends within a selected window of time. The comparison set may be the stream within which the event potentially exists or it may be a designated group of stream s (e.g. a group designated as “normal”), a randomly selected group or the entire set of stream s of the specified type available. This deviation may only consider sets already analyzed and sections only before a fixed or moving point in time (as in a real time analysis or simulated real time analysis) or may use all sets and times available. In an alternative embodiment, the search for deviation may be applied with reference to a definitive object or set of definitive objects selected by the processor or by an expert.
Both alpha and beta occurrences may also be defined in terms of the relationship of its characteristics to the characteristics of the candidate correlating occurrence (i.e. the alpha for the beta or the beta for the alpha). In one embodiment the user may define the relational objects, (as by using a drag and drop designer), by selecting the channel or stream (which defines the time series type), and by selecting the occurrences which meet specified range of criteria, and by identifying the timed relationship (such as the time interval) of the beta occurrence in relation to at least a portion of the preceding alpha occurrence, and/or by identifying the spatial relationships and/or frequency relationships of one occurrence to the other occurrence, may in the most fundamental relational binary, the event binary, the alpha and beta occurrences are events identified by the time series objectification processor, for example such as a time series objectification processor in U.S. patent application Ser. No. 11/280,559 and U.S. Pat. No. 7,081,095, the disclosures of which are hereby incorporated by reference in their entirety for all purposes as if completely disclosed herein. The relational binary processor then aggregates the relational binaries according to their time of occurrence and/or to specific criteria for aggregation set by the user processor to derive images and the images are aggregated according to their time of occurrence to derive the MPPC and care derived of events and patterns across hundreds of parallel time series. In a sense, the relational binaries and events become the discrete “pixels” from which MPPC of a patient's physiologic system are constructed by the patient safety processor.
According to one embodiment, the patient safety processor may be also programmed to organize the events and relational binaries into larger aggregate factorable objects, which may also be constructed as a unified object timeline rather than a motion picture. Each aggregate factorable object may include a specific aggregation of events and relational, binaries objects. In some aggregate factorable objects, the individual relational binary and event objects occur in a specific sequence or range of sequences (which may be overlapping) and the objects have a specific temporal relationship (or range of temporal relationships) with respect to each other. One specific type of object timeline may be specified as simply a grouped set. In another example, relational binaries are ordered in specified sequence in which the event and relational binaries objects were detected thereby defining the object timeline.
According to one embodiment, objects of specified types may also be combined derived to render a “unified patient timeline” which may be a simple summary of the patient's physiologic system and care. The MPPC and care provides the information at more comprehensive level. Both may be configured to provide further simplified summarization or image detail revealing drill down. The unified patient timeline may for example, represents an instance of at least one factorable aggregate object derived from a plurality of parallel time series into a single time-series or time line, often of relational binary objects of a specific type or plurality of types. In one instance the unified patient timeline and/or the MPPC and care may be constructed to be a life long time line and/or motion picture, which preferably may be recorded whenever signals are available, such as during a hospitalization or when connected to a home monitor or when blood testing is made. The beginning of the motion picture or time line may be defined by the time of the earliest date of data (which may be derived from archived patient data) the unified patient timeline does not end until a patient dies, segments of the timeline (or motion picture) may be separated for examination by location of the patient such as a hospitalization segment, or by actions taken to treat the patient, such as a peri-operative segment, or by events relating to altered patients states such as the segment immediately preceding death or while sleeping. According to one embodiment, an object nomenclature may be provided which designates the timed and sequence relationships of the binary objects and events of a plurality the parallel patient related time series, thereby converting a large plurality of datasets into this single time series of factorable objects, which may be readily outputted interpretable through application of a succinct nomenclature.
In one embodiment, the physician may mark a test result or other data point as mistaken or anomalous. In this case the processor splits the analysis into two—the working analysis (which removes or alters the test result or other data point) and a background analysis (which maintains the original data). The processor may run scenarios in which the original test result stays in effect to determine if conditions occur that might have been expected from the “so-called” anomalous test. The background will not affect the working analysis but notification may be generated if a correlation of events is found in a sufficiently suggestive pattern to warrant a consideration that the original test results may not have been mistaken and, in fact, would account for conditions that do not fit the current working state (e.g. the state with the test results removed). Background analyses may be deleted according to time (e.g. after a certain amount of time in which no correlation to following events is found) or at the request of the user or system operator (e.g. to reduce resource requirements).
In another example the processor may be programmed to generate more frequent testing binaries to confirm or exclude an apparently evolving image. In this way the processor is trying to look as far forward as possible with additional testing to confirm the motion picture of a particular failure as early as possible so that the delay associated with waiting for the detection of a failure cascade as by various traditional threshold breaches is eliminated.
In an example, as part of assuring that the future image is complete, the testing binaries are designated such that the addition of certain drugs (the alpha event) into the image, may cause automatic orders for testing to monitor for complications related to the drug (the beta event) if selected events, binaries, and/or images are present. In an example, if the physician orders heparin, a testing binary may be generated and added to the image, which includes automatic order for a platelet count every 48 hours. According to one embodiment, the time series objectification processor is objectifying the time series of platelet counts to detect a least one fall event (as for example defined by a negative slope and/or a magnitude of fall and/or a threshold fall), if a fall event is detected a divergent binary is generated and a marker indicating a fall is added to the image along the platelet count time series, the processor may generate more frequent platelet testing binaries, to confirm the presence of these divergent binaries in the image. If multiple divergent binaries are detected then the processor may generate different types of testing binaries wherein the alpha event is the fall in platelet count. This may trigger a cascade of testing binaries such as, for example, wherein the alpha event is a binary including a heparin treatment and a fall in platelet count and the beta event is, for example, a platelet factor IV assay and/or another assay. In this way, using the imaging processor, the delay associated with waiting for an absolute or relative threshold drop in the platelet count is not required but rather the slope of the platelet count, the presence or absence of prior heparin therapy, the patient's risk of bleeding or thrombosis, may all be included in the image to trigger the automatic measurement of additional testing. In addition the image of cascade as it evolves may trigger additional testing binaries (as for hepatic function tests, as required to determine the safety of Argatoban, a medication which may be ordered if the images are consistent with heparin induced thrombocytopenia). Here the advantage of having these binaries and images as part of a MPPCF is evident, because the processor will be examining the images of the motion picture for other causes of the fall in platelet count which may include cascades indicating TTP as will be discussed and/or occult hemorrhage to name a few.
One embodiment programmatically images the parallel physiologic time series to render a relational pyramid of data with the top of the pyramid representing data at the highest level of analysis and abstraction while data moves down through layers of analysis, the bottom layer being the raw data stream s. The healthcare worker may investigate the pyramid in the following ways to name a few: (1) Drilldown—the care worker may navigate into the details of the data and the rationale of the analysis (i.e. both the conditions that exist and the rules by which the analysis has arrived at its conclusion); and (2) Aspects—view ports into the system which emphasize certain elements/conditions and de-emphasize (and/or filter out) other elements/conditions). These two examples above may be used together allowing the healthcare worker to navigate through the relational pyramid vertically (drilldown through levels of analysis) and horizontally (through filters/aspects).
In one embodiment the relational pyramid may be manipulated by the healthcare worker and/or researcher to consider hypothetical scenarios or scenarios based on the rejection of certain test results or events which may be considered in error, anomalous or otherwise inaccurate. Alternate pyramids may be stored in whole or as differential images. Alternate pyramids may be compared against the working pyramid to understand the results of the altered data.
In one embodiment, the processor will automatically consider alternate pyramids under certain conditions—such as the existence of perturbation for which no precursors may be identified. The sudden existence of perturbation or of divergence may, by considering the range of possible precursors, suggest anomalous conditions: inaccurate diagnosis, faulty monitoring equipment, labeling mistakes, the failure of a patient to take medication as prescribed, to name a few. According to one aspect, the values and/or patterns of the blood tests such as the inflammatory mediators is/are compared to the image(s) of physiologic perturbation or to the pattern(s) or values of at least one physiologic parameter, such as the pulse rate, respiration rate, and/or ventilation oximetry index to name a few. Upon the detection of an apparent relationship, the processor may automatically order a sufficing number of sequential blood tests to confirm that the pattern of the parameter is convergent with the pattern of the blood test thereby providing strong supporting evidence, reinforcing redundant evidence, that the physiologic parameter and the mediator have a common physiologic failure based linkage, such as the failure of sepsis for example. One embodiment extends that analysis to incorporate specialized inflammatory mediators into the moving picture of failure so that comprehensive comparison of the marker or indicator to the image of the physiologic parameters and treatment may be provided. One embodiment generates dynamic images of relational variations of a set of time series associated with a complex system to generate a real time motion picture of a failure of the system and/or of forces applied to the system and condenses the complex data of the EMR into a single motion picture of perturbations, treatments, physiologic responses, diagnostic testing, recoveries, diagnoses, missing data, patient locations, and/or other datasets and further provides treatment, and/or testing, alarms, notifications, diagnosis, and/or orders based on the motion pictures.
One embodiment provides a system and method for programmatic characterization of a plurality of related complex and dynamic processes, which; converts patterns along a plurality of parallel times series derived from each of the process into discrete objects, organizes these discrete objects into relational objects, and, organizes the relational objects to render a unifying programmatic image of a complex and dynamic process, and then, applies expert systems to automatically recognize images or image portions, or the specific motion pictures which are indicative of at least one failure of the complex processes and/or provides a relational object image generating and processing system to provide characterization and quantification of physiologic systems by generating an organized analytic construct defined by a time series matrix of relational objects.
Although the number of potential modes of failure is very high in any hospital environment, the occurrence of certain modes of failure is reasonably likely under a given set of circumstances in the hospital. A failure mode diagram illustrating common modes of failure given a combination of a group of diseases is shown in
The data management system 300 may include one or more processor-based components, such as general purpose or application-specific computers. In addition to the processor-based components, the data management system 300) may include various memory and/or storage components including magnetic and optical mass storage devices and/or internal memory, such as RAM chips. The memory and/or storage components may be used for storing programs and routines for performing the techniques described herein that are executed by the processor 304 or by associated components of the data management system 300. Alternatively, the programs and routines may be stored on a computer accessible storage medium and/or memory remote from the data management system 300 but accessible by network and/or communication interfaces present on the computer.
The data management system 300 may also include various input/output (I/O) interfaces, as well as various network or communication interfaces. The various I/O interfaces may allow communication with user interface devices, such as a display, keyboard, mouse, and printer that may be used for viewing and inputting configuration information and/or for operating the system 300. The various network and communication interfaces may allow connection to both local and wide area intranets and storage networks as well as the Internet. The various I/O and communication interfaces may utilize wires, lines, or suitable wireless interfaces, as appropriate or desired.
In one embodiment, the device 306 is turned on as for continuous viewing (with a notification) by the processor 304 when images are indicative of a significant potential failure and/or cascade process or at a point wherein the patient's risk class exceeds a threshold value. The risk class may, for example, be derived as a function of a calculated instability index or a detected instability index pattern and/or detected failures. The instability index may be, for example, a confidence metric correlated with a matched image. For example, when an MPPC has a high likelihood of being associated with a serious condition, the instability index may be high. The instability index may be a numeric index, a color or graphic indicator, and/or an audio or text message.
In accordance with an embodiment, the device 306 includes an interactive screen displaying items, such as one or more working diagnoses, differential diagnosis, parameters derived from patients including laboratory parameters, monitored parameters, and subjective parameters (e.g., sedation scale, confusion scale, or pain scale) or the like. In an embodiment, the term “parameter” herein may refer to an absolute or relative data point or set, a pattern, or a deviation, a range of such data points or sets, a pattern of such data, a relationship along a single set of data and/or or between a plurality of sets of data, and/or patterns of data. The data may be an objective data type or subjective data type and may be directly and/or indirectly derived or historical in origin. In addition various outputs from the failure imaging processor 360 (
The data management system 300 further includes a medical records database 308 including laboratory data 310, historical data (e.g., diagnosis) 312 and therapy data (e.g., medications) 314. The medical records database 308 is coupled to the processor 304 and to the monitor 302 so that those systems may have access the data stored in the medical records database 308. The processor 304 may include a component or direct link to the centralized patient medical record, which contains real time data and receives data input from all hospital sources. Thus, a database containing substantially all of the components relating to the patient available to the hospital may be directly accessible to the processor 304 in real time to allow the embedded relational processor render relational binaries, and construct and detect failure image components which include these data from varied sources.
In accordance with an embodiment, the processor 304 is adapted to comprehensively engage the medical records database 308. As discussed further below, the processor 304 may be programmed to provide for formal, automatic simultaneous engagement, of physiologic failure image components, medication failure image components, testing failure image components, aggregate failure image components as derived from the relational processor and to render them in a timeline for viewing.
The processor 304 may be adapted to provide an immediate review of all failure image components and to take action based on the detection of specific failure image components. The processor 304 may be capable of responding faster and more reliably than the healthcare worker because it may be adapted to constantly monitor the evolving failure image components form the earliest onset of the first divergent binary. The processor 304 may therefore detect failure image component cascades, which originate from single divergent binaries, which might easily be undetected by the healthcare worker until it is too late. The processor 304 may also be programmed to alarm on divergent or null binaries upon which no action has been taken or upon which the action has not corrected the evolving divergent binary or failure image component. For example, in a scenario in which the processor 304 has been updated by the nurse that a blood culture has been obtained, the presence of a null binary may be generated indicating testing failure image component if after a pre-selected time the result is not available to the processor 304 whereas the presence of a divergent binary indicative of a physiologic failure image component may be detected if the culture is positive. If testing failure image component is detected the processor 304 notifies the lab of the apparent delay. The notification is an alpha event and a receipt response to that notification is a true beta event. Therefore the failure of the lab to indicate receipt may cause the occurrence of a divergent binary, which may trigger the notification of the nurse in the same manner until a convergent binary concludes the sequence. If on the other hand, a physiologic failure image component is detected (the culture is positive), the processor 304 notifies the nurse again in the same binary generating fashion.
While a positive blood culture is the beta event of the culture testing binary, it is the alpha event for another group of testing binaries such that the initial divergent testing binary may cause the processor to assure acquisition of a complete blood count, a comprehensive metabolic profile, increased frequency of blood pressure and pulse measurements, ventilation indexing oximetry and other testing as programmed into the processor 304 in response to the specific divergent binary detected (in this case a positive blood culture). These new testing binaries may generate unexpected beta events (such as a low blood pressure, a high pulse, or high ventilation to oximetry index) and these beta events may thereby define a new set of divergent physiologic binaries. This new set of divergent binaries (in aggregation) may be sufficient to meet the pre-selected criteria of an aggregate failure image component suggestive of early septic shock, which diagnostic consideration now includes an alpha event to a plurality of new binaries which have been programmed into the processor to assure timely and proper monitoring, timely proper patient location, and timely proper diagnostic testing, and timely and proper intervention in the event of the detection of this type of aggregate failure image component. In addition, the beta events of the divergent physiologic binaries which included the aggregate failure image component now become alpha events for new physiologic binaries wherein the beta event of each of the new binaries includes the return of each these values back to a normal range within a pre-selected time period (thereby assuring, that the aggregate failure image component is corrected timely, if possible). In additional, the positive blood culture is also the alpha event for a treatment binary such that the processor 304 may be expecting to see the correct antibiotic in response to positive blood culture administered within a pre-selected time interval. If this does not occur a divergent binary indicating treatment failure may be identified and assured nurse notification may proceed by the binary building method previously discussed.
According to one embodiment, in response to the detection of any significant divergent physiologic binary, the device 306 may be programmed to prevent the failure of notification by building a set notification binaries, which must end with convergence. The device 306 may also be programmed to prevent failure to timely treat by building a set of treatment binaries, which must end with convergence. Further, the device 306 may be programmed to prevent failure test by building a set of testing binaries, which must end with convergence. The device 306 may also be programmed to detect associated physiologic failure image components by identifying divergent physiologic binaries in associated with the initially discovered divergent binaries.
According to one embodiment, the processor 304 includes an associated, connected and/or embedded eventing system. In this eventing subsystem, users may designate actions to be initiated or data to be recorded when a specific occurrence is identified. This eventing system may interface with other internal or external systems including notification systems, workflow systems, asynchronous communication systems, reporting systems, decision support systems, dashboards, data warehousing and/or data mining systems to name a few.
According to one embodiment the relational processor is self-modulating and provides an automatically expanding analysis, which is rapidly responsive to the occurrence of even a minor failure image component. The analytic activity of the processing system is capable of multidimensional growth and diminishment in direct response to the magnitude and number of failure image components detected. In this regard, the processor 304 upon the occurrence of a physiologic failure image component may generate a cascade of notification, testing, treatment, and physiologic binaries even if that failure image component includes only a single physiologic divergent binary. The beta event of the physiologic binary may include the alpha event of each of a new generation of notification, testing, treatment, and physiologic binaries. Each of these new binaries also have a beta event, each which may induce the formation of other binaries wherein the beta event includes the alpha of another binary of the same or another type. A spontaneously growing cascade of binaries thereby evolves toward assuring timely notification, timely testing, and timely restoration of physiologic stability.
A rapidly expanding, cascade of these types of divergent binaries indicates evolving patient instability of the patient or poor performance of the healthcare system. An analysis (as by objectified pattern recognition or statistical analysis) of the timed patterns of the types and sequence of the divergent binaries may allow the determination of poor health or poor responsiveness of the healthcare worker is causing the cascade to be propagated. As health is restored, and provided the healthcare workers are timely responsive, the binary cascade may automatically diminish and the various failure image components may no longer be detected. The outputs of the relational binary object processor therefore provides a self modulating processing system which may be readily used and further analyzed to track the health of a single patient, or the patients on a given floor, or the patients hospital wide. The outputs of the object binary processor also provides a self modulating processing system indicative of the quality of healthcare delivery provided to a given patient, on a given floor, or hospital wide.
The processor 304 may be applied to other complex dynamic data sets other than medical data wherein a self-modulating relational analysis and control would be useful. The processor 304 has utility for the data mining, for example in association with the processing of archived datasets to identify the failure image component process from the initial spark (the first divergent binary) to extensive system failure. The processing of archived datasets provides the opportunity to review the automatic modulation of the binary cascades which are derived of various failures and to facilitate the construction of dynamic failure image component diagrams for complex processes in the hospital, as well as in industrial processing such as the food, chemical, or pharmaceutical processing. The processor may be programmed such that the user may select each alpha event and allow the processor to detect, offer, and or derive events and relational binaries, which have specified temporal, frequency, or spatial relationships with the selected event object. Alternatively the processor 304 may be programmed to construct its own set of convergent object binaries with a learning dataset by processing the outputs of healthy individuals and then the processor may be used to detect divergent binaries when applied to patients by identifying the lack of the expected beta events (which were defined by the learning dataset). Sensitivity for cascading (the initiation of further processing based on the detection of a divergence or a failure image component) may be adjusted by modifying the sensitivity for trueness of the beta event or by modifying the criteria such as slope, or magnitude of the objects during the objectification process. This provides a high degree of flexibility in defining sensitivity to the designation of a binary as divergent and this therefore allows a high degree of control over the sensitivity to cascade initiation, propagation, and extinguishment. Cascades may be modular or divergent or failure image component specific. A modular group of cascades may be selectable from a menu and then each one in the group modified as desired.
As shown in
According to one embodiment, the relational binary processor 348 and the time series objectification processor 336 may adapt to the output of each other to modify the analysis. For example, the detection of an event, a reciprocation, an incomplete reciprocation or other objects or patterns by the time series objectification processor 336 may cause an adjustment to the cascade responsive to the detection of a divergence. Alternatively or in combination the criteria for designation of a wave segment as an event object within the time series objectification processor 336 (for example the slope criteria for identifying a fall event object of serum sodium) may also be adjusted based on the presence of a specific alpha event. In an example, when an alpha event including a diagnosis of cerebral vascular infarction (CVA) is detected, this may cause the time series objectification processor 336 to reduce the absolute slope (less negative slope) for designating a fall event object of serum sodium, which, is preferably one of the betas in such patients. By automatically reducing the absolute slope for the designation of the beta event the alpha diagnosis of cerebral vascular infarction is adjusting the sensitivity of the diagnostic process allowing automatic and dynamic adjustment upon the occurrence and detection of different physiologic vulnerabilities. In this example, the increase in sensitivity for detection of a fail event object in serum sodium (which, combined with the alpha that includes a CVA diagnosis) would include a divergent binary), which may trigger a diagnostic cascade for close monitoring of the serum sodium and/or the evaluation of additional laboratory studies and/or the reduction of free water delivery. This is desirable due to the unique vulnerability faced by patients with CVA as a function of the potential for inappropriate increase in anti-diuretic hormone due to the CVA.
Since the relational binary definitions within the binary definition set 344 may be individually defined and refined by processing large populations of historical data, correlations may be verified, rather than being simply proposed and maintained as a function of consensus or expert opinion. In one embodiment, cascades originated by criteria for divergence provided by an expert, which untimely lead to extinguishment without intervention may be automatically adapted to either change the sensitivity for the detection of the divergent beta or to change the cascade resulting for the divergent binary. In another example, cascades originated by criteria provided by an expert which continue self propagate and expand despite timely action and without progression of the physiologic divergence may be automatically adapted to either change the sensitivity for the detection of the divergent beta or to change the cascade resulting for the divergent binary. The sensitively and specificity may be further enhanced because the system may be applied to archived training data sets wherein the outcomes are known so the magnitude and direction of the cascades may be compared to the desired magnitude and direction of the cascades and adjusted accordingly. With applied archived datasets the application of auto-adaptive adjustment in event criteria, divergence criteria, or cascade generation may be applied until the cascades proceed without premature auto extinguishment and excessive propagation. Furthermore the system may be applied to hypotheticals on the missing data to allow determination as to how they might affect incomplete (null) binaries.
According to one embodiment the processors, including the time series objectification processor 336, the relational binary processor 348 and failure imaging processor 360, may output the results of their analysis into the MPPC Database 368. The MPPC Database 368 contains the time series 328 on which the analysis was performed as well as the results of analysis including the event stream s 340, the relational pairs 352, the aggregate failures 364 as well as aggregations, relationships and alternative images of these elements. In one embodiment, the metadata rule-sets (both primary and alternative and/or temporarily overridden or altered elements) are persisted as XML (event definition set 332, binary definition set 344, Failure image component definition set 356) in the patient safety image database 368.
According to one embodiment, a processor is programmed to render sequential time series components (which may be discrete and/or succinct) and which may be subsequently linked other sequential time series components along and across parallel time series to produce a comprehensive relational image of physiologic failure and/or patient care. These components may be rendered by methods defining polarity reversal or inflection points, state changes, by imaging methods to detect pattern components (for example subsequent to time series rendering into a particular format for example), and/or by another method for defining and/or programmatically “packaging” events, image components, and/or occurrences for relational imaging and analysis. According to one embodiment “time series objectification” is employed, for this purpose. Time series objectification may be rendered by a time series objectification processor 336, an embodiment of which is discussed below. In one embodiment, time series objectification is the process of converting a set of time series into a stream of sequential discreet elements or objects such that substantially the entire time series of data is converted to a time series of objects in a relational hierarchy of ascending complexity. In another embodiment these objects are created by identifying boundaries within the time series based on the values of the points within the time series (for example threshold violation and/or state match to name a few) and/or the relationship between these points (for example polarity reversal, inflection point, state transition to name a few) using a set of rules based on an understanding of phenomena within the system from which the time series are derived or which are learned adaptively. The discrete objects which are created represent and characterize an occurrence providing a time location and a set of properties derived from the aggregated data within the boundary defined. These objects are differentiated by location and the properties derived and therefore individual objects can be qualified and the stream of objects can be searched against. Further, the conversion to discrete objects provides for the identification, qualification and searchability of relationships between elements. Relationships can be converted into aggregations and/or hierarchy of elements within which properties can be derived from components of an aggregation/hierarchy to the aggregation/hierarchy itself and/or from the hierarchy/aggregation to the participating components.
A time series objectification processor 336 may for example, contain instructions as provided in U.S. patent application Ser. No. 11/280,559, and Ser. No. 11/351,449 the specifications of which are incorporated by reference herein in their entirety for all purposes. Accordingly, such processors may function by constructing a time series matrix including of substantially all of the parameters derived during, the process of the hospitalization and then objectify each time series in the matrix to produce an objectified time series matrix. The time series which include the matrix may, for example, include objective measured values, drug dosing, infusion rates, and subjective clinical scores to name a few. At least some of the time series may be provided as a step function. For example, time series of the weights, serum sodium values, SPO2, ventilation volume or rate, heart rate, pulse amplitude, pulse slope, drug infusion dose, sedation score, pain score, stupor score, working diagnoses, an instability score, a severity of illness score, to name a few, may all be included.
The objectification processor can define objects by a wide range of methods which may be programmatic and/or image based or by another method of defining objects. In an embodiment, a time series objectification processor applies a linear and/or iterative dipole slope approach to the recognition of waveform events, as for example respiratory or oxygen saturation events. For example, the events associated with airway collapse and recovery are generally precipitous and unipolar, for this reason the linear method suffices for the recognition and characterization of these nonlinear waves. However, the iterative dipole slope approach is particularly versatile and may be used in situations whereby the user would like an option to select the automatically identification of a specific range of nonlinear or more complex waves. Using the iterative dipole slope method, the user may select specific consecutive sets of points from reference cases along a waveform as by sliding the pointer over a specific waveform region. Alternatively, the user may draw the desired target waveform on a scaled, grid. The user may also input or draw range limits thereby specifying an object or set of objects for the microprocessor to recognize along the remainder of the waveform or along other waveforms. Alternatively, the processor may automatically select a set of objects based on pre-selected criteria that may be empirically determined. Since the iterative dipole process output may be shape-dependent (including frequency and amplitude) but is not necessarily point dependent, it is highly suited to function as a versatile and discretionary engine for performing waveform pattern searches. In accordance with embodiments, the waveform may be searched by selecting and applying objects to function as Boolean operators to search a waveform. The user may specify whether these objects should be in the same order. Recognized object sequences along the waveform may be scored to choose the degree of match with the selected range. If desired, (as for research analysis of waveform behavior) anomalies within objects or occurring in one or more of a plurality of simultaneously processed tracings may be identified and stored for analysis.
After the process of objectification and further processing of the time series matrix (e.g., generated from object stream s 340) the images are transferred to the patient safety visualization processor 372 which presents and highlights the detected MPPC on an outputted display or patient safety console 306 or through a patient safety report 380 (either electronic or paper) or within an electronic representation as an interface 376 (for example the European Data Format (EDF)) which may provide an interface to external systems. In one embodiment, the aggregation of data, analysis and metadata provide the source of data for the patient safety visualization processor 372. In one embodiment, the Patient Safety Visualization Process 372 provides a visualization of a patient's condition in a comprehensive grouping defined by rows of timelines of specific signals and/or grouping and/or categories of signals and/or signals. In one embodiment the global state of each row is represented by color in a spectrum with a different color moving from stability to failure (for example, Sustained Stability [deep blue], Stability [light blue], convergence [green], Perturbation [yellow], Divergence [orange], Null [black], Failure [red]. Cascading Failure [bright red]), in another embodiment colored arrows, icons, blinking or highlighted text, and/or other visual representations along each time line represent these states.
In one embodiment the patient safety visualization processor 372 represents the patient condition as a set of pixel stream s moving from left to right to show evolution of condition over time. The processor provides the navigation backward and forward in time as well as up and down through levels of analysis within the patient safety image database 368. In this embodiment the levels of analysis may be, for example:
time series—Unanalyzed data stream s in the form of time series
events and Perturbation—events, state matches, state transitions and threshold violations characterized within their respective channels as to whether they represent clearly defined perturbation according to the event definition set 332
Systemic Response—convergent, divergent and null binaries representing the relationships between events, state matches, state transitions, threshold violations, perturbations and expected elements according to the binary definition set 344
Failure—Failure images that have been identified within a single patient
System Failure—Failure images within a specific category (such as the respiratory system) representing images of failure that have been identified within a single patient
Failure Patterns—Trends of failure and failure images within patient population or a specific region, such as a specific hospital ward for example.
In one embodiment the patient safety visualization processor 372 composes an image on computer monitor (the patient safety console 306), which may be composed by a series of pixels oriented horizontally representing data and analysis stream s. These pixel stream s may be stacked vertically with the position on the x-axis representing a specific point in time. The processor provides for the movement of the pixel stream s horizontally to provide a pan through time.
Each pixel stream may be composed of a set of pixels, which indicate the state of the data and/or analysis at the specified point in time. The pixel has a state (e.g. represented by color) and granularity (the length of time it represents [for example 1 minute]). The size of the view as well as the selected span of time determines the granularity of the pixel. In the contemplated embodiment, the pixel is displayed by the highest level of instability found within the time span represented by the single pixel within the pixel stream. Further, each pixel has a level of abstraction, which determines which objects from the patient safety image database 368 contribute to its state. The contributing objects are shown below by level of analysis:
time series—Data points within the channel (e.g. oxygen Saturation Values)
events and Perturbation—events and threshold violations
Systemic Response—Relational binaries
Failure—Failure images
Failure Patterns—Failure trends and correlations.
In one embodiment, groups of pixel stream s are stacked vertically to create a patient safety visualization. Patient safety visualizations may be composed of pixel stream s of different patients or of data and analysis stream s within a single patient. Patient safety images provide the ability of the care worker to filter the analysis quickly to identify problem areas or areas of a specific nature. Sorting may be provided highlight emerging failure cascades or other pattern failures. In an embodiment patient safety images may be composed of different levels of analysis displayed on the patient safety console 384 at the same time correlated by time. The use of mixed-analysis level visualizations provides the care worker with the ability to quickly understand the relationship between the lower levels of data (e.g. incomplete recovery within oximetry) and the higher levels of analysis (e.g. the identification of narcotic-induced ventilation instability).
In an embodiment the patient safety console 384 provides the user the ability to trace a failure condition back to the earliest events associated with the failure to provide a visual display of a failure cascade. Alternatively, individual events and threshold violations may be selected to identify which higher-level objects in which they played a part. In other words, low-level events may be traced forward to understand their relationship within evolving patient instability. This tracing may be accomplished in many ways. For example, the processor 304 may exploit the fact that Alpha events of a relational binary are often the Beta event of a preceding relational binary. This chain of relational binaries provides a powerful tool of analysis. The patient safety visualization processor provides the ability to isolate these binary Chains showing their origin, evolution and resolution.
Alternatively, and in concert, the processor 304 may use the trend of probabilistic momentum. In one embodiment, visualizations may be filtered by the existence and character of binary Chains or a recognizable trend of Probabilistic Momentum. In one embodiment, and if selected by configuration, the patient safety visualization processor provides the ability to navigate into the metadata models at any point within the visualization, event, convergence and image Diagrams or other occurrence definition Visualizations are accessible from objects, which were composed using specified elements within these diagrams within the event definition set 332, binary definition set 344 and image definition set 356. Navigation into the metadata models provides expert care workers and researchers the ability to further understand and/or alter the analysis.
The patient safety console 384 presents a complex set of data and analysis that meets the immediate need of the busy care worker. In one embodiment, analysis at the highest levels may be collapsed, into a single pixel stream or group of pixel stream s per patient that provides a simple representation of the evolution of overall patient safety. Within and from that pixel stream the care worker may drill down into the most complex displays: multiple levels of analysis, binary Chains, trends of Probabilistic Momentum and metadata models to name a few. Alternatively this drill down may be provided by for example mouse over, touch screen, or may appear automatically when the processor detects certain adverse patterns or thresholds.
In one embodiment, the object stream visualization focuses on the relationships and cascading of the onset of perturbation within the patient. This is an alternate, and complimentary, view to the pixel stream s described above which focus to a greater extent on the state of discrete elements within the system at various levels of analysis. These two visualizations may be used in parallel and/or provide navigation between them. In the contemplated embodiment, the object stream visualization represents events and threshold violations as icons along a time series in which the icon is placed at the first point in time in which the event or threshold violation occurred. Icons indicate their character by color, size and decorations. The basic icon is an arrow pointing either up or down (as in
Decorations on the arrow may be presented to provide visual cues as to the nature of the event. A line underneath the head of the arrow indicates that he event that occurred was a threshold violation. A circle around the arrow (see 979 of
In one embodiment, the patient safety visualization processor will provide automated visual navigation for a specified period of time and/or specified images. This automated visual navigation acts as an analysis-driven video playback of the selected period of time. The healthcare worker selects “Play” and allows the patient safety visualization processor to move visually through the evolution of a specified condition. The healthcare worker may choose navigation movements including “Play”, “Pause”, “Fast-Forward”, “Rewind”, “Skip Forward”, “Skip Backward”, to name a few. In the contemplated embodiment, during Play mode the patient safety visualization processor moves at different speeds through the automated visualization depending on the severity of the condition being displayed. If the time series being displayed have little perturbation (or little perturbation related to the specified failure cascade) the processor will move very quickly through time (i.e. from left to right). When an area of interest, as determined by the processor, comes into vision the patient safety visualization processor will slow the movement from left to right. Further, the patient safety visualization processor will highlight elements that indicate, clarify and specify the evolution and/or cascade of failure as well as their relationships with other elements. The patient safety visualization processor will further display translucent pop-up panels that provide further textual and/or visualization elements to describe the current view and elements within the current view. At any point, the healthcare worker may “Pause” the automated visual navigation to review the displayed data and/or drill into what has been displayed.
In a complimentary embodiment, the healthcare worker may select from a summary view a time span to review and also indicate sections of the time span for which they are interested. The patient safety visualization processor will slow for the areas selected that are of interest and will increase the textual and visualization display appropriately for the highlighted sections. In one embodiment the patient safety visualization processor chooses the object stream s to display and may include or remove stream s as they become important in the video navigation. The healthcare worker may choose to include additional stream s or to “pin” stream s so as to make them always available in the video navigation, missing stream s are also indicated. The patient safety visualization processor may further indicate to the healthcare worker the time required for automated visual navigation (e.g. “Standard visual, navigation will require 2 minutes and 37 seconds”).
The patient safety visualization processor will include audio and visual elements corresponding to and synchronized with the time series data along with time series data if video and audio feeds are available, in the contemplated embodiment, healthcare workers may include audio and/or video comments into the data stream s to communicate and collaborate regarding elements displayed within the patient safety visualization processor. The patient safety visualization processor may be directed to include all or a specified subset (e.g. “Include Comments from Doctor X”) of these elements within the automated visual navigation or may be directed simply to indicate their presence such that the healthcare worker may invoke them as needed.
In one embodiment, the patient safety visualization processor may “record” an automated visual navigation session into a non-interactive video format which may be viewed on standard video equipment, with streaming technology or in a standard media player such that automated visual navigation sessions may be shared with healthcare workers who do not have access to the patient safety image database or the patient safety visualization processor (e.g. as an attachment to an e-mail or accessed from a video-enabled phone).
Having a flexible time series 404 supports the construction of the time series objectification matrix. For example raw signals from monitoring equipment may be contained within numeric point stream 410 whereas complex micro-domains such as images may be stored in occurrence stream s 414. Both of these stream s may be treated similarly as a time series 404. The time series matrix 400 is simply an aggregation of supplied and derived time series 404. The fact that all elements within the streams are time locatable gives the matrix the parallelism that allows time relationships to be identified and aggregated into the scope of an occurrence 418.
Each time series 404 has a single type (time series type 402). For signals this may be defined as the signal type (e.g. oximetry, EKG). The matrix 40) may contain many time series of the same type. For occurrence stream s the time series type is defined by the occurrence definition associated with the definition (e.g. oximetry fall event, Narcotic Induced Instability image). A time series may be broken up into sub-spans called time series segments 416. Both the time series 404 itself and all associated time series segments 416 are time bound 420 (e.g. have a start and end time).
The processor 304 defines an isolated event as an event identified independent of a relational binary, in other words, an event that does not belong to any relational binary. In an example of a embodiment, a range of events may be defined within a single definition. Range-based definitions allow the user to define all but one parameter within the event definition. For that final parameter, rather than a single value or expression, the user is allowed to define a set of value ranges in which the final parameter may fail. For each range provided, the type of event is specified. In this way a set of related events may be defined.
The processor 304 employs a multistage approach to the identification, creation and refinement of event objects. An initial set of criteria is used to establish the time location of a time series segment, which is established as an event candidate with a specific start and end time within a single point stream. Once this time series segment has been established it becomes a micro-domain within which further properties and elements may be derived. For example, within events established on a numeric time series that are specific to a unipolar trend of data (e.g. directional events 480), the processor 304 will further refine the trend by looking for smaller changes in the trend. The processor 304 calls points of change within the trend Inflection points 486. Once inflection points are determined then the event is broken up into smaller segments called event segments 488. In one embodiment Inflection points 486 are derived in three ways:
The creation of sub-objects (e.g. inflection points and event segments) within an event provides a wide array of elements against which the definition of the event may be refined. For example, the researcher may decide to create an event type 476, which must include a “Slow Descent” segment followed by a “Rapid Descent” segment. The researcher may choose to remove the “Slow Descent” portion of the candidate event from the final event, or split the candidate event into two separate events. Alternatively, the researcher may not choose to use these elements and their properties as part of the criteria (e.g. used to accept or reject) but may use them to characterize the event (as per, for example, the assigning of attributes described below).
For each event 474 there is an event type 476 which identifies the type of micro-domain which has been created. The event type is associated with the definition used to identify, construct and qualify the event. For example, a researcher may create an oximetry Rise event as an event type 476 which defines a positive directional event 480 within the oximetry point stream. The definition for the event (depicted in
Another subclass of occurrence 448, within this embodiment, is the relational binary 462. The relational binary comes in two subclasses: occurrence binary 466 and event binary 468. The relational binary is depicted more completely on
During matrix construction, the processor 304, by default, aggregates and analyzes repeating objects. The processor 304, after identifying an occurrence, reviews the object stream for other occurrences proximate within the associated stream s. By default, the processor 304 looks for repeating occurrences 432 (e.g. occurrences of the same occurrence mode (e.g. relational binary) and occurrence type (e.g. oxygen reciprocation)). Each occurrence type definition specifies a minimum recurrence count as well as a recurrence threshold time, span by which the processor 304 may determine whether or not to aggregate occurrences of the same type into a repeating occurrence object. In one embodiment, the recurrence threshold may be specified either as a fixed time span or as a series of time spans depending on the state of the aggregation process. For example, the definition may include an initial recurrence threshold (a maximum time span allowed between the first and second occurrence) and a subsequent recurrence threshold (a maximum time span allowed between any subsequent two occurrences). In one embodiment, this threshold may be a function (e.g. the time being increased incrementally by the number of occurrences that have already been aggregated).
Once the processor 304 has identified a repeating occurrence 432 and has aggregated occurrences within this object, the processor 304 will create a point stream 442 for all properties of the specific type of occurrence being aggregated called a property channel 442.
These property channels 442 are attached to the repeating occurrence 432 and may be accessed through the repeating occurrence 432 object. Further, these property channels may be analyzed through the time series objectification processor, the relational binary processor and the Imaging Processor to create occurrences within and between the property channels and/or between occurrences within the property channels and other point streams or occurrence streams in the system. For example, in
One species of repeating occurrence 432 is cycling. Cycling is a specific type of repeating occurrence in which there is a repeating occurrence 4332 of a relational binary 462 made up of a perturbation event is followed by an event returning the system to normal A repeating occurrence 432 with a property moving in a specific direction is called a trending recurrence. A trending recurrence is a specific type of repeating occurrence 432, repeating occurrences 432 may be trending in more than one property. The relationship between occurrences within a repeating occurrence supplies additional properties, which are placed in a point stream and analyzed. For example, the time between occurrences is tracked. As an example, consider 9 fall events that are aggregated into a repeating occurrence 432. In this case a property channel 442 of 8 points will be created of the time spans between the occurrences within the aggregation.
The processor 304 may not create a definition for each repeating occurrence. A definition, may be constructed from the minimum recurrence count and recurrent thresholds supplied within the occurrence type definitions. The processor 304 allows repeating occurrence definitions to be created to add additional criteria to the identification process and/or to add calculated properties and/or attributes to the repeating occurrence 432. In these cases where a specific definition is created, the researcher must indicate whether the user-defined definition replaces the default definition, or is to be identified in addition to the default definition. In this way, the researcher may identify any number of specific types of repeating occurrences 432 for a single occurrence type 444 (e.g. oxygen fall event).
The researcher may also select to suppress repetition for any occurrence type 444 causing the processor 304 to ignore repeating occurrences 432 for the specified occurrence type 444. A repeating occurrence 432 is itself an occurrence and therefore may repeat. Recurrent repeating occurrences (which in themselves are simply repeating occurrences 432 of repeating occurrences 432) will create property channels for the properties of the repeating occurrences 432 contained within. For example, the number of occurrences within a repeating occurrence will be aggregated into a property channel 442. As with the repeating occurrence 432, the Recurrent repeating occurrence does not require a definition, but the researcher may provide a definition to add additional criteria and/or specify calculated properties and/or attributes. repeating occurrences 432 and Recurrent repeating occurrences may be elements of an image 452. In particular, the processor 304 recognizes a particular type of image designated a deviation image. In a deviation image a repeating occurrence 432 is followed by a specific image or set of images that represent a deviation from an established cyclic phenomenon. narcotic-induced ventilation instability is one such deviation image. In this case, the physiologic response to an oxygen desaturation event has set up a repeating occurrence 432 of convergent oxygen reciprocation binaries. If this repeating occurrence 432 is followed by a narcotic therapy event and then a divergent oxygen reciprocation binary the processor 304 will identify the narcotic-induced ventilation instability deviation image. In one embodiment, deviation images include additional fundamental properties including the point of deviation—the time at which the repeating occurrence ended.
repeating occurrences 432 may be identified for any occurrence types including images 452, therefore many repeating patterns of occurrence may be addressed with this mechanism. For example, if A and B are images, then the pattern ABABAB may be identified as the repeating occurrence of an image defined as AB. Patterns that are more complicated than simple repetition are addressed in the processor 304 with the pattern occurrence 446. The pattern occurrence 446 defines a pattern in terms of other occurrences. Researchers first define the set of occurrence types which will be used at least once in the pattern and define each a mnemonic. For example, the researcher may choose 3 images and assign them the mnemonics A, B and C. Mnemonics are alphanumeric representations of occurrences used as a reference to an occurrence type. Once the set of occurrence types has been selected and mnemonics assigned the researcher may define the pattern in terms of the mnemonics separating them with brackets. For example, the researcher may define a pattern as [A][B][A][C][A][B][A]. In one embodiment, a pattern occurrence 446 may have a set of patterns, any of which, if matched, will cause a pattern occurrence to be created. pattern occurrences 446 are defined as a subclass of repeating occurrence 432 and therefore have property channels 442 in the same way that repeating occurrences 432 have property channels 442, but, since the members of the pattern occurrence 446 are of different types, the property channels are only created for properties that match within the pattern occurrence. For example, in the example above (with the pattern [A][B][A][C][A][B][A]), there will be property channels 442 containing 7 points for properties that exist in A, B and C. For properties that exist only in A and B, there will be property channels with 6 points. pattern occurrences 446 are occurrences themselves and therefore the processor 304 will automatically identify repeating pattern occurrences as described above.
When the processor 304 identifies an occurrence 448 (or candidate occurrence) the aggregation of all elements within that occurrence becomes a micro-domain. Relationships, properties and elements within this micro-domain have meaning because they exist within the specific context of the occurrence. The processor 304 provides for mechanisms to exploit this fact and to allow the researcher to specify properties that reflect the new meaning of properties and the relationship between elements within the micro-domain of a specific occurrence type 444. When an occurrence 448 is identified it has a set of properties 456 or values that the processor 304 identifies, calculates and/or assigns. There are three types of properties 456—fundamental properties 468, calculated properties 470 and attributes 472. Fundamental properties 468 are defined by the processor 304 depending on the occurrence mode and the nature of the associated point stream or occurrence stream. For example, all occurrences have the fundamental properties of start time, end time and duration to name a few. As another example, all occurrences that are events 474 on a numeric point stream have a magnitude and slope to name a few. As another example, repeating occurrences 432 have a member count and mean separation duration to name a few.
Calculated properties 470 and attributes 472 are properties 456 that have an expression-based definition. The processor 304 uses this mechanism to refine processor 304-specific occurrence objects and elements (as an example Inflection points 486). The researcher is also able to define these properties and store these definitions with the occurrence definition (See
In summary the three modes are described as follows:
The three modes are described below in more detail. A convergent binary 506 represents a relational binary wherein the processor 304 has identified the Beta occurrence which has been defined as having an expected relationship to the Alpha occurrence. A convergent binary 506 may have either a true occurrence 512 or a missing occurrence 514 as a beta depending on what has been specified as the expected condition. If a true occurrence 512 was specified in the relational binary definition (See
divergent binaries 508 represent a pair of occurrences identified in a relationship that contradicts the expected relationship as described in the binary definition. Therefore a divergent binary 508 may have either a true occurrence 512 or a missing occurrence 514 as a Beta depending on what has been specified as the expected condition. If a True occurrence 512 was specified in the binary definition then the associated divergent binary 508 may have a missing occurrence 514 as a Beta. If a missing occurrence 514 was specified then the associated divergent binary 508 may have a True occurrence 512 as a Beta. The class structure therefore allows for zero or one occurrence 512 and zero or one missing occurrence 514. According to one embodiment, a divergent binary 508 may not contain two Beta occurrences.
Null binaries 504 represent the existence of a condition in which an alpha occurrence was identified but the data stream (or stream s) from which the expected beta event is to be derived is unavailable to the processor 304. A missing occurrence 514 is associated with a set of time series segments 518 that represents the areas within the matrix that was searched for the occurrence described as expected in the binary definition. Null occurrences 510 are not associated with time series segments 518 because at least one time series to which they would have been attached or the relevant section of that time series is unavailable or corrupted. The processor 304 will convert null binaries 504 to convergent 506 or divergent 508 binaries as new time series or time series segments become available and analysis is executed. An occurrence 512 may be the Beta occurrence of a first relational binary 500 and the Alpha occurrence of second relational binary 500.
In one embodiment occurrence stream s are stored as divergent binary streams and convergent binary streams. In the alternative or in combination, all such stream s or a portion of specific stream s or a grouping of stream s filtered for severity of divergence (for example) may be aggregated and rendered for periodic viewing wherein for example, the temporal relationships of for example divergent binaries or of the occurring images are easily recognized or specifically indicated.
To further clarify this structure it may be useful to describe the order of operation within an example of a embodiment of the processor 304 as it constructs the analysis according to one embodiment.
The actual relationship between the alpha and beta events, which include the object binaries, is not necessarily defined by cause and effect (which may not be known with complete certainty) but is rather defined by the pattern relationship such as a temporal spatial, and/or frequency relationship of the occurrences, or simply by their prior designation as a relational pair. For example, the actual relationship between the alpha and beta occurrences including a given relational binary could be a cause and effect, two effects resulting from an unmonitored cause, a relationship between two monitoring technologies measuring the same physiological phenomenon, an expected compensatory response, or a pathologic response, to name a few. Once the general association has been captured within the binary then the relationship between the occurrences within the binary is examined to determine if the relationship meets the requirements for the determination of a higher level of relationship (e.g. cause and effect, compensatory response to name a few). Though the focus of
Specifically, the search for an image may be reduced to successive searches for occurrence binaries. An image is defined by a set of occurrences and their relationships within the image. Two sets of occurrence sets are defined—Sequenced and Non-Sequenced. Sequenced occurrences are defined with a time relationship to at least one other occurrence within the image. Non-Sequenced occurrences are defined only with a time relationship to the span of the overall image. (See
The process of searching for an image starts with making sure that all of the constituent occurrences types have already been processed for construction and the resultant occurrence stream s have been placed into the matrix. At this point the overall search region (a single patient stay, for example) is reviews to gather the counts of the constituent occurrences within the region. If any of the required occurrences has a count of zero within the target region then the search for the specific image type may be abandoned. (In an alternative embodiment, the processor 304 continues to construct the image as much as possible so as to store “Near Miss” images for research or analysis purposes.) Next the Sequenced set of occurrences is processed first. Sequenced occurrences have a specified, time relationship to at least one other occurrence type within the image and therefore may easily be aggregated into dynamic binaries 500. For example, if three occurrence types are specified as A, B and C; A is defined as preceding B and B is defined as preceding C then the processor 304 may aggregate this set of 3 occurrence types into 2 dynamic binaries to search for. The first dynamic binary is created by choosing A as the Alpha occurrence and B as an expected Beta occurrence. This dynamic binary is searched for within the target region and the resultant dynamic occurrence stream is placed back into the matrix. Once the [A→B] dynamic binary has been processed, then the second dynamic binary may be created as a binary having the dynamic [A→B] binary as the Alpha and the C occurrence type as the Beta (described as [[A→B]→C]. If only the time relationship has been specified between the constituent sequenced occurrences (as opposed to, say specific related properties within the created micro-domain), then the order of the dynamic aggregation of binaries is not required to follow the sequence. In other words, the processor 304 may, for performance or other reasons, choose to aggregate B and C first as [B→C] and then aggregate A as [A→[B→C]]. Since the process of searching for binary occurrence is equivalent for searching for a single occurrence (when only the time relationship is specified) then the results are the same. In this way, any number of sequenced occurrence types may be processed to find their existence within the target time region.
non-sequenced occurrence types are searched somewhat differently because of their unstructured nature. First the occurrence counts of all specified non-sequenced occurrence types are placed into a set and sorted from least to greatest as far as their count within the target region. In an example of a embodiment the processor 304 begins with the least count occurrences and proceeds through greater count occurrences to maximize performance. For each occurrence type within the non-sequenced set a dynamic binary is created first with the resultant occurrence created by aggregating all of the Sequenced occurrence types (as [A→[B→C]] in the above example) and then with all of the other Non-Sequenced occurrence types. Continuing the example above, if two Non-Sequenced occurrence types are specified D and E then the following set of dynamic binaries would be created and processed: [D→[A→[B→C]]], [D→E], and [E→[A→[B→C]]]), [E→D]. In certain cases, the reverse dynamic binaries (both [D→E] and [E→D] are not required.
An image may also specify that a specific occurrence must not be found proximate to the rest of the image. In this case processor 304 will use the verify non-existence relationship within the dynamic binaries created. If multiple occurrences of a dynamic binary are found during the image identification process the processor 304 will select which of the occurrences to include into the image. (In one embodiment, pruned constituent occurrences are stored for research or analysis purposes.) The selection may be directed by researcher specifications so as to minimize time relationships within the image or may be directed to maximize the overall span of the image 434. Dependencies may also be a factor. For example, if a more proximate occurrence being selected destroys a dynamic binary required for image completion the processor 304 may select a different occurrence to be maintained though it is less proximate. This decision may be directed by the researcher according to the nature of the micro-domain and its relationship to physiological forces.
Occurrence stream s created from dynamic binary types will be maintained within the matrix if the target image is found. If the target image is not found, the processor 304 may be configured to remove the dynamic binary occurrence stream s since they are relevant only within the context of the given image micro-domain, image Identification is completed if and when all of the dynamic binaries that are created generate occurrences that match the required time relationships. When this occurs the constituent occurrences are aggregated into a micro-domain establishing the scope of the image. At this point further stages of construction may proceed to generate properties, relationship and sub-elements, refine scope and finally to qualify the image to see if it is a true image or a disqualified candidate.
In one embodiment, a researcher may specify an inheritance relationship between occurrence types. With this mechanism, all of the properties associated with the parent occurrence type will be defined in the child type. In this way, an “Is-A” relationship may be established. Calculated properties definitions are a type of property definition 540 defined with a name and an expression. The expression may contain references to any number of other properties 456 (Fundamental properties. Calculated properties or attributes). This mechanism may be used simply to rename certain fundamental properties within the context of the occurrence definition 544. For example, if a binary type called an oxygen reciprocation binary is defined as a fall event (Alpha) and a Rise event (Beta), then the oxygen reciprocation binary may include a calculated property called Magnitude Ratio which is defined as “=Alpha·Magnitude/Beta·Magnitude”. The processor 304 verifies submitted expressions for validity. Calculated properties have access within their expressions to a set of functions. Functions available include mathematical functions (Absolute Value. Square Root, to name a few), aggregation functions (Mean, Count, Sum, to name a few) and relational conversion functions (Relative To, time Span From, to name a few) to name a few.
In one embodiment, to assist with cross-channel expressions the processor 304 provides functions and values that may be included in these expressions. The Relative To function provides the ability to derive a percentage relative to a target element. Further, the processor 304 allows evaluations relative to the ranges of associated time series. This value is called Normalized and has a specific function. The processor 304 and researcher may create additional properties called attributes using an attribute definition, attributes are properties that are assigned based on condition being true. attribute definitions have three parts: Conditional Expression, attribute Name and Value Expression. If the Conditional Expression is found to be true then the Value Expression is evaluated and the result is assigned to the property with the name provided in the attribute Name. The Value Expression is of the same format as the calculated property expression. The Conditional Expression uses the format of the calculated property expression but also requires at least one Boolean operator (e.g. “alpha·Magnitude >4) or property.
In an alternative embodiment. Calculated properties and attributes may be defined and/or assigned with an embeddded programming language (e.g. a scripting language). In one embodiment, to assist with expressions, the processor 304 provides a mechanism to define Synonyms 538 within the occurrence definition 544. Both elements and properties may be defined as a Synonym. For example an oxygen reciprocation binary may contain a Synonym called Recovery which is defined as “=Beta” and another Synonym may be called desaturation and be defined as “=Alpha”. Synonyms are simply symbolic replacements for use within expressions, conditions and search criteria. The processor 304 will pre-process all expressions to replace synonyms before evaluation. For example, now the Magnitude Ratio above could be defined as “=Recovery·Magnitude/desaturation·Magnitude”. In one embodiment each occurrence must go through a qualification stage to determine its veracity as an occurrence of the type specified. This qualification process uses Qualification Rules 546 that are attached to the occurrence definition. In one embodiment the subsequent stage of qualification is accomplished through a rule set applied to the properties within the candidate occurrence. In an alternative embodiment, the candidate is run through a series of preservation and rejection tests. Preservation and rejection tests may be designated as absolute (e.g. if this state obtains, reject the candidate with no further evaluation) or may be aggregated into a set of overrides. In the second case the rules are created in the format: “Preserve if X unless Y” and/or “Reject if M unless N” where X, Y, M and N may represent any Boolean expression over the occurrence candidate's properties.
In the contemplated embodiment, the processor 304 maintains (or is able to derive) the Dependency 548 of all occurrence definitions 544, property definitions 540 and Qualification Rules 548. This dependency works down to the Fundamental property level. The processor 304 may then, given these dependencies, determine whether certain Fundamental properties, Calculated properties and/or attributes may be left unevaluated or their evaluation delayed. For example, if an attribute is created by the researcher but not used (directly or indirectly) then the processor 304 will leave the attribute unevaluated. If at some future time, for example during display by the patient safety visualization processor, this attribute is requested then the processor 304 will execute evaluation. The management of dependencies also allows for caching strategies in which the processor 304 stores the results of evaluation during execution and/or persistence, but may accurately recognize the need for re-evaluation if any of the dependent properties have been changed. Further, within the instance of an occurrence, not only are the dependencies well understood, but the values of each element within the dependencies are known. Given this fact, the researcher or care worker may drill down into occurrence instances to understand both the metadata and data dependencies of the selected occurrence. For example, within a oxygen Cluster (for example, defined as a repeating occurrence of oxygen reciprocation binaries), the caseworker may see not only the value of a specified property (for example, an instability index), the structure of calculation that created the property and the specific occurrences and the values of their properties that participated in the final value.
Finally, an occurrence definition is associated with a set of ordered Construction Phases 532. Elements within the occurrence definition 544 may be associated with a specific Construction Phase to indicate to the processor 304 when it is appropriate to identify, create or evaluate the sub-element. For example, if constituent occurrence types are marked with the Construction Phase “Scope”, then the marked occurrence types are required for establishing the occurrence as a candidate occurrence. If they are marked with a subsequent phase then the constituent occurrence is added after scoping to refine the scope, properties marked with a specific Scope are not evaluated until the specified phase is being executed. Qualification Rules are automatically considered in the Qualification Construction Phase.
Occurrence definitions 544 are subclassed into a set of definitions classes that represent the type of patterns that the processor 304 may search for. Specifically, occurrence definition 544 is subclassed into event definition 552, relational binary definition 554, image definition 555 and repeating occurrence definition 536. The repeating occurrence definition is further subclassed into pattern occurrence definition. The relational binary definition is subclassed into event binary definition and occurrence binary definition (for more detail see
An event definition 584 is a subclass of an occurrence definition 582 and each event definition is associated with a unique event type 586, event subclasses are described in detail in the text associated with
There are two subclasses of relational binaries—event binaries and occurrence binaries. The event binary definition 616 requires that the Alpha and Beta are events and therefore uses event definitions 622 whereas the occurrence binary may use an occurrence definition 624 as Alpha and Beta, relational binaries that are between an event and an occurrence will be an occurrence binary since an event is a subclass of occurrence. A Search definition 620 includes parameters, expressions and other descriptions to tell the processor 304 the relative time series segment(s) to search for the Beta event or occurrence. The Search Mode 626 indicates different types of searches including Expected, Verify Non-Existence and Reoccurring Verification. A Search definition 620 must have at least one Search Mode 626 but may have more that are used in combination (e.g. a Reoccurring Verification of Non-Existence). For each relational binary, the relationships may be further defined to select a more specific Relationship type 610 (or to match one of a set of relationships), for example cause and effect, subordination to name a few. The relational binary definition may optionally include rules for the indication of a specific Relationship type. In an alternative embodiment the selection of the Relationship type is implemented through the attribute mechanism.
This structure may best be understood within the context of the User Interface modeled in
The box containing a pair of time offsets (e.g. 768) represents the Search definition 620. This definition contains the Start and End time offsets from the end point of the Alpha occurrence for which the Beta occurrence should be searched in the target Beta time series. Finally the icons represent the Alpha and Beta event types. These types reference a unique definition which provides the parameters with which the relational binary processor may search the identified time series segment(s) for the existence of a pattern. Further, criteria is contained within the binary definition itself based on the micro-domain established by a candidate aggregation of an alpha and beta paired by the spacial requirements alone. These criteria, established by the researcher or automatically determined by the processor 304 (as per guided image discovery described below), may utilize all of the properties (including Fundamental properties. Calculated properties and attributes described below to name a few) of the candidate binary to accept or reject the binary as a true representative of the specified binary type.
c shows a subset of the occurrence definition set as it relates to the image definitions. In one embodiment, the classes described here represent the image definition set used by the image processor to create the image analysis. An image definition 642 represents the set of element definitions and their relationships, which allow the image processor to determine whether the pattern of elements meets the criteria of the specified image. This Structure of
Each relationship added to the design surface 764 must have at least one time interval provided (e.g. 768) which represents the Search definition 620 for the relational binary definition 612. Etch relationship may be directional. The line includes an arrow end-style on the end that represents the Beta definition, either a Beta event definition 622 or a Beta occurrence definition 624. The end without an arrow represents the Alpha definition, either the Alpha event definition 622 or the Alpha occurrence definition 624. Within
Each pair of events, which has a connecting relationship, represents a single event binary definition 616. In the above figure, the following seven binaries:
This diagram does not represent all of the relationships of each of these events. It is an example of a subset view into the overall convergence model with a focus on sleep apnea. Relationships and elements may be removed from this diagram without removing them from the entire model (i.e. the editor distinguishes between “Remove” which removes the element from the diagram but not the model and “Delete” which removes the element from the diagram and the model [including all other diagrams]). A diagram may be constructed that shows all of the events and relationships, but it would likely be so large and complex as to be unreadable.
The editor will check the diagram for validity before persistence or at the user's request. For example, a relationship without a Beta occurrence would invalidate a diagram. An invalid diagram may invalidate the convergence model. It is contemplated that a convergence model cannot be persisted into a binary definition set. The editor allows for an invalid state to provide flexibility during diagram construction. Further, if the target binary definition set is associated with image definition sets that are available to the editor, the editor may warn of conflicts with associated models by changes to the diagram. Depending on editor settings, these changes are disallowed, or the changes may be propagated into the images. Each diagram element may be manipulated in a more detailed way through property editors associated with the element type and access to the occurrence property Subsystem described below. The property editors provide access to all editable properties of the associated definition objects such that the editor is sufficient to construct a complete binary definition set. The editor provides for adding text, notes, lines and other visual elements to the diagram to increase human readability and to communicate between users. These additional visual elements have no affect on the binary definition set.
Zero or more sequences may be specified, but if an element is placed in the Sequenced section it must be part of a sequence. Elements placed in the Non-Sequenced section cannot have relationships. Only existence is specified within an overlapping Span of Influence (defined below). The image diagram differs from the binary diagram in that the diagram itself represents an entity—the image definition 650—and is not simply a collection of other entities (e.g. binaries in the case of the binary editor). Removing elements changes the definition of when a image will be identified. All elements added to the image diagram represent an “And” relationship for identification purposes (i.e. all elements and sequences must exist for the image to be identified). In one embodiment, to create “Or” scenarios, multiple image diagrams are created with variation representing the “Or” combinations. An image may include any combination and number of occurrence types. The editor may check the diagram for validity before persistence or at the user's request. The editor allows for an invalid state to provide flexibility during diagram construction. Each diagram element may be manipulated in a more detailed way through property editors associated with the element type. The property editors provide access to all editable properties of the associated definition objects such that the editor is sufficient to construct a complete image definition set.
The spatial configuration (in time) of the occurrences are preferably satisfied to create a candidate image. Once a candidate image is established, the processor 304 may use this set of objects as a micro-domain to establish all of the properties using the occurrence properly Subsystem (described below). The properties derived within this micro-domain may then be used to refine the definition of an image, to more specifically characterize the image or be used in the decision to accept or reject the candidate image as a true (i.e. qualified) image. The image editor may be used to enter into the occurrence property Subsystem to further define calculated properties and attributes (defined below) of the image. The editor provides for adding text, notes, lines and other visual elements to the diagram to increase human readability and to communicate between users. These additional visual elements have no affect on the image definition set.
Since
Using these basic designations the image of
Two drug treatments are evident in the image, the antibiotics Vancomycin 982 designated by its dose on the time line, and Levofloxacin 983, similarly designated. Also a rise in IV fluids in the form of normal saline 984 is indicated. All of these treatments come late after the image has lone been indicative of a high probability of sepsis. (This delay, which may be detected in real-time by the patient safety processor, suggests poor and ineffective care, which has ignored or otherwise been poorly responsive to the patient safety processor. The processor may be programmed to provide an indication of the quality of the care provided. time lines, which include the care worker or ward may be provided so that delays may be linked to particular locations or care workers.)
The image of the progressive cascade 980 shows the drug treatments components 982, 983 of the image are too late because they appear within the image very late along the cascade 980. The late portions of the image of the cascade 980 also include a very ominous beta including a rise in anion gap 985. The addition of this new image provides a mature image of cascade 980, which is now strongly indicative of a stage of septic shock. Other image views may be for example; specific expanded portions of the time lines, specific expanded views of images (or other occurrences) along the timeline portions, specific groupings of the timelines, overviews of perturbation progression from group to group (an example of this is shown in
Unfortunately. TTP shares many of systemic response features of the very common disorder of sepsis (
Since TTP is associated with the accumulation of large multimers of Von Willebrand factor which damaging red blood cells and induce extensive micro vessel thrombosis producing confusion, renal failure and microangiopathic anemia which is associated with sentinel schizocytes which may be detected in the peripheral smear of blood (if the diagnosis is suspected and the test is ordered). Thrombocytopenia, renal failure, and hematuria may appear earlier in this process than with sepsis but these early findings are only an “image clue” and does not differentiate the two moving images. The decision to diagnosis a rare condition instead of a common one on the basis of a clue is a dangerous human tendency and a pitfall, which may result in patient complications. Alternatively the decision to diagnosis a common condition despite the clue because as the trite medical student saying goes “common condition occurs commonly” is equally dangerous. Indeed it is tragic that patient have to die because of such trite and oversimplified thinking. However it is the nature of many humans often approaches the analysis of overwhelming complexity with unknowingly capricious, summary judgment. This combined with the overlapping complexity of disease and healthcare is one of the most important reasons that comprehensive real-time physiologic and care rendering by the generation of digital MPPC and care is important.
The MPPC suggestive of TTP may be generated by the processor, with the processor indicating a image consistent with the possibility of sepsis and/or TTP and other less likely conditions such as an acute vasculidity. The processor may output non-specific characterizations of the image such as “image consistent with a life threatening acute or sub-acute thrombotic and inflammatory augmentation” and may present a differential diagnosis of the processes, which may generate such an image.
Also, as for example upon the detection of a threshold frame or frames, automatically order the peripheral smear, blood cultures, urine cultures, sputum cultures, Chest X-Ray, ANA, pancreatic enzymes, renal sediment, and ANCA study to enlarge and fill in the gaps of the image as rapidly as possible. It is the hospital experts who will ultimately decide the cost effective balance of ordering these tests as defined by the position the tests are ordered along the cascade. If desired the reports from the Chest X-Ray may include a section which will appear as a time series (as for example a step function). The radiologist in the interpretation (and in comparison with the last test and the last number selected by the last reading radiologist) may enter an indication of pulmonary infiltrate, pulmonary edema, and the like and may indicate a value between 1 and 5 which may result in a step change of the processor 304 from the last test. In this way the results of studies such as Chest X-Rays and other such interpreted tests become a source for dynamic time series rendering and incorporation into the imaging process. This will also provide an objective tool for comparing subjective quantification between radiologists and between various testing modalities in relation to the actual MPPC thereby identifying radiologists who are not generating reasonably reproducible or comparable subjective quantification in relation to themselves, others or the MPPC. In an example if a radiologist consistently calls the level of pulmonary edema a 1 or 2 in patients who have MPPCs consistent with of acute severe CHF and acute severe pulmonary edema or if the radiologists quantification consistently fails to follow or predict the clinical course then instruction can be provided or in the alternatively it can be recognized by the processor 304 that the input from that specific radiologist is not useful in further defining the images along the processor 304.
The presence of an image including images defining a failure cascade 1108 including inflammatory-hemodynamic respiratory-augmentation 1116 with an early fall in platelet count 1104, a fall in the Ventilation oximetry Index (VIO) 144, a fall or threshold value of hemoglobin 1144, an rise or threshold value of a confusion score 1148, and/or a rise or threshold value of red blood cells in the urine 1150, and/or a rise or threshold value of Creatinine 1152. Together the combination of images produces a MPPC suggestive of the possibility of TTP and/or sepsis and/or other less common processes. For example, if the patient had just received blood it would suggest a possible transfusion reaction.
It is not as important for the processor 304 to make the diagnosis as it is for the processor 304 to indicate to the healthcare worker the gravity of the image, a differential diagnosis as suggested by the image, and the general type and/or physiologic description of failure cascade present, and perhaps a notification that the detection by the processor 304 of this type of image requires prompt notification of the attending physician and transfer to ICU. If the image has insufficient binaries because results are not available to define enough beta components to define the presence of the image suggestive of TTP with a sufficient confidence level to take action, the unavailable tests are ordered upon the detection of the partial image in an attempt to complete the image. Note in
In this case, failure to rescue is not preempted because of human delay in physically following the orders of the processor 304. The delay in carrying out the order is determined by the processor 304 and the processor 304 may be programmed to up-indicate the warning upon increasing delay. To prevent this delay, the processor 304 may be programmed notify another station if action is not taken in response to detection of various evolving images such as the one in
Rather than or in combination with an index, if desired the processor may be programmed to provide an indication of the severity and number of the aggregate perturbations in each group. These may be for example designated by many enlarging or colored arrows, other icons, and/or timed instability scores, to name a few. Many such options may be included so that the user may define his or her preference to visualize the sequence and patterns of cascade progression across groups.
A range of expert and pattern recognition systems may be applied to analyze the images and the images generated by the image processor. These include the image identification Processor. In one embodiment the image Identification Processor works with the image editor which allows the user to select the images for detection using for example a from a drag and drop interface. In an embodiment the drag and drop interface provides for the discretionary selection of, for example, the time-series type to be selected, then occurrences are selected on each time-series type in order and the ranges of relative positions and orders occurrences is selected. In this example, the image editor allows customization of the desired ranges for the components of the images (and therefore the ranges of the images themselves) to be selected as well as the response of the image Identification Processor to the detection of a given image and/or images. The image editor may allow for selecting the ranges of timing and order of the occurrences to generate a specific output such as a proposed diagnosis, warning order for more testing or imitation or termination of treatment. The image Identification Processor may also be adaptive such that a physician inputs the diagnosis present, such as for example septic shock, with a given image. The physician may also capture a given image or set of images into the image editor to then select ranges about the occurrences within the image which also would have indicated the presence of septic shock so that the adaptive image processor may learn more quickly.
In one embodiment researchers work from the top down to define images. Researchers begin by selecting a set of channels in which they want to “paint” the image.
The image editor creates and modifies image definition sets. Furthermore, the image editor works in concert with both the convergence editor and the event editor to create and modify the binary and event definition sets. In one embodiment (shown in
In one embodiment researchers work from the bottom up to define failures from a set of time series. Researchers may begin with a set of actual time series from patients diagnosed with known failures, with a set of time series generated by the processor to simulate certain conditions or a set of time series simulating no perturbation at all within a patient. This set of time series may be designated as immutable (for example with the set of actual time series) or may be edited to provide a sample of the patterns being defined. Researchers may select portions of the time series, which the image editor then will analyze to provide candidate event definitions. Alternatively the researcher may select parameters to define an event and the time series displayed will indicate the results of that definition overlaid on top of the time series to provide visual guidance to the researcher. Once the researcher completes the definition of an event the image editor will compare that definition with other definitions within the same channel. If similar patterns are found the researcher is alerted and allowed to create a new event type or select one of the event types already selected. If the event is a relational event, the researcher may select a corresponding event from which relational parameters may be defined and experimented with or the researcher may simply define a function (e.g. >2×Relative Magnitude. Once an event has been fully defined then the researcher may choose to relate the event to another event within the image or to a search window within the image (e.g. to indicate a missing or null event). The researcher may indicate that a processor-ordered event as the beta of a relational binary. Groups of events and relational binaries or any other occurrence may then be selected to define a images, images already defined within the image definition set are highlighted such that they may be included into the image the researcher is working with or the researcher may simply select to alter its definition. Access into the occurrence property Subsystem is available and the expression editors included indicate immediate results with respect to the current image or other occurrences selected. This allows the image editor to work on all aspects of the image including scope definition, qualification rules to name a few.
In one embodiment, the image editor may be presented with a large collection of time series sets provided with the indication of the presence or absence of a particular known image. The image editor creates a set of candidate definition sets refining them to create the right specificity and sensitively to match the sample set. Once the best-fit definition sets are created, a second large collection of times series sets are provided with the indication of the presence or absence of a particular known image. The image editor first uses the candidate definition set, determining sensitivity and specificity and then refines the definition set to be better suited if possible to both the first and the second collection of sample data. This process may be executed iteratively until a best-fit set of definition sets is created or the process is deemed not to be asymptotic and is abandoned.
In one embodiment the image may be “played” or executed by the image editor as an MPPC to provide further time-specific markers. A default execution of an image is “played” by placing all events as specified in their default (e.g. midpoint) location within their respective search windows as defined by the image definition. A sample result of this is displayed in
In an alternate and/or complimentary embodiment, the image editor provides the ability to split the execution of an image into multiple intermediate and/or end states. Each different branch within the image definition may be defined as a state within an image or a different, albeit related, image. Trees of related images may be composed to provide alternative evolutions of failure within the image definition.
The case presented in
The patient presented with an acute stroke but was recovering and alert. Then he slowly began to develop confusion and less alertness. As the stroke was large the nurses and physicians managing the case thought that the patient's confusion and obtundation was due to brain swelling and called the family in to adjust code status. For this reason the family consulted one of the present inventors. The patient SPO2 and ventilation rate were normal, he had no signs of sepsis and because of recently normal electrolytes the attending physicians did not think that a metabolic cause for the confusion was a reasonable option. In other words they misdiagnosed the pathophysiologic failure pathway (illustrated on the failure mode diagram 200 of
Since the stroke caused the SIADH (which cased the fall in serum sodium) the actual modes of failure were significantly different than suspected by the hospitalist in this case. The actual failure followed to path 171 from the stroke 208 to the hyponatremia 242 and then followed the path 172 from the hyponatremia 242 to the confusion 220. In this case the patient survived the missed diagnosis but he experienced several extra days unnecessary days in the hospital because of delay in detection and treatment of this failure.
Now referring to
Here the problem is simple but the early signs of failure were at first subtle at a time when intervention would have prevented the increased length of stay later the pathways of failure were confused leading to further delay and considerable family since they were told the incorrect diagnosis. In this case the nurses and physicians may have been busy or may have been inexperienced or simply not familiar with the subtle decline in mutation, which may attend the development of SIADH in a stroke patient. The reason subtly findings are missed is myriad. Note also, in defense of the healthcare team, as illustrated in the failure mode diagram of
According to one aspect of the present embodiment failure mode diagrams, such as the one in
The patient safety processor is not constrained by these definitions during analysis in an absolute way, but rather compares actual data to a plurality of images and image states to find best-fit matches. The patient safety processor will indicate all possible images and image states ranked by level of confidence. For example the patient safety processor may indicate that a MPPC is consistent the Systemic Inflammatory Response Syndrome with a high degree of confidence and early septic shock with a medium degree of confidence and that TTP (and other potential alternatives) or overlapping failure modes are remotely possible in view of the image and remain to be excluded. The physician may be asked if it is desired to order the focused testing to exclude these remote alternatives or overlaps and/or the processor may be programmed to automatically add this testing based on a specific range of images (as defined, for example, using the drag-and-drop editor discussed previously).
The identification of failure within the patient safety processor is not the single selection of a failure mode or a failure state, but the ranking of a set of images with regard to their fit within the data presented. The identification of multiple images is not simply the selection of alternatives. Multiple failures may, in fact, exist and be interacting with each other. Early states of some images may be very similar, or in fact exactly the same, as the early stages of other images or of a combination of images. The patient safety processor provides the analysis and visualizations that allow the health worker to understand the current state of the patient (and patient environment) in terms of possible future states—alternatives and candidate overlaps—along with confidence levels as to their specification. Further, the Patient safety monitor allows the health care worker to query the patient safety processor with regards to confidence levels and, in particular, the comparative confidence level between two images and/or image states. For example, the confidence level for sepsis is low with the frame shown in
In one embodiment, the patient safety processor may be trained by a pathophysiological engine (such as a human simulator as are known in the art) for the creation of failure and response images. Given a specified event definition set and binary definition set the Patent Safety Processor provides a dynamic image derived from the input of the pathophysiologic engine and the Processor is instructed as to the nature of the images so that when these images are detected in the future they are recognized. In one embodiment a human simulator is connected to the processor 304 to provide an improved teaching tool for healthcare workers. Researchers may select to be presented with a normal, unperturbed patient with various conditions. Once a dynamic image of the patient is displayed researchers may introduce perturbation into the pathophysiological engine, which will result in new dynamic images from the Patient safety processor. For example, a research may select relationships presented according to a convergence and toggle them to divergence. Also, random divergence may be configured into the system. Divergence with respect to a single or a set of response system(s) may be specified to model the breakdown of systemic response. Divergence may be configure globally or for a specific timeframe indicating that systemic response fails, or is delayed. In this way both perturbation and failure of systemic response may be selectively introduced to create images. These images may be persisted to be further edited within the image editor or other tool. The researcher may select several different variations and save them as failures and/or failure states. These failures and/or failure states may be persisted within a image definition set to be used by the image processor. Further, resultant images may be compared with actual patient data to refine image, binary and event definition sets.
Alternately or in combination, according to one embodiment, an MPPC may be carried from the processor 304 to the processor driving the human simulator so that healthcare workers may observe the reanimation of the MPPC of the processor 304 either as a digital animation or as a reanimation derived from output of a human manikin.
One utilization of the embodiment, which combines the pathophysiological engine to the processor 304, is to model treatment protocols. The engine may output expected or unexpected parameters (divergence) in response to treatment and the image output of the processor 304 may be observed, and/or recorded for protocol modeling. Further, using the ability to introduce divergence, allows processed protocols or other protocols to be verified for reasonable redundancy to cover failures of systemic response.
One of the problems with conventional diagnostic process is that, even with diligence, the time to detection of the cause of a complex process is a direct function of care worker experience. One of the present inventors, a trained critical care physician, has over many years managed perhaps several hundred septic shock patients and for this reason he learned to visualize the complex cascades embodied in
The bandwidth of the MPPC is defined by the number of tests, historic data, and treatments et al. which include the MPPC. When potentially treacherous images of perturbation are identified along the MPPC the Patient safety processor is programmed to quickly broaden the bandwidth to investigate the alternative causes. This is important because the longer the duration an undetected failure mode the greater the increase in cost and mortality because complications develop with widen the cascade and make salvage more expensive and difficult.
A narrow bandwidth (fewer tests and/or simpler tests per unit of time) is, on the other hand, (without considering the cost of allowing a longer duration of failure) less expensive than a broader bandwidth. The “effective bandwidth” includes those components of the bandwidth, which actually contribute to characterize the factors actively defining the images, which are components of the MPPC. Poorly conceived testing and treatment increases the bandwidth and the medical cost but may not increase the effective bandwidth. The goal of the processor 304 is to increase the effective bandwidth as rapidly as possible without broadening the bandwidth inordinately. The ideal system monitors with a few monitors and tests but uses these as sentinels, increasing the number of monitors and tests automatically if specified occurrences are identified or failure cascades begin.
Therefore, one function of the present invention provide a mechanism to automatically increase the effective bandwidth of the MPPC at any time (for example at 2 AM in a rural hospital), to optimally shorten the duration of failure without the application of a continuously wide and expensive bandwidth. One optimal mechanism to broaden bandwidth is with improved testing, such as focused tests, which have a high sensitivity and specificity for a specific failure mode. The MPPC archive 16 of the patient safety processor Network 10 may be examined for opportunities to increase the motion picture bandwidth and achieving a balanced mechanism for mortality and cost reduction by shortening the duration of failure through earlier detection and improved treatment response.
The use of smaller bandwidths of monitoring and testing combined with the ability to auto adjust to a wider bandwidth upon detection of events, binaries, images, repeating occurrences and patterned occurrences reduces cost, and allows application to a larger populations of patients for the same or less cost. The occurrences, which precede physiologic failure, are recorded and this relationship is monitored by the interaction processor 304 via the network. If a relationship of physiologic failure to these specific occurrences is identified, the processor further examines the MPPC for these now sentinel occurrences, which, if obtained earlier using a wider testing bandwidth, might have provided a more diagnostic image earlier. These test are then added to the processor 304 as automatically ordered upon the detection of the in the sentinel occurrences. If a test or set of tests fails to provide additional image detection or quantification, which is not provided by another cheaper or less invasive test or set of tests, then the test or set of test is not used. In this way the processor 304 is auto adaptive, learning to optimize the processing of the MPPC by eliminating unnecessary or redundant testing and by adding testing which allows earlier detection of failure thereby reducing cost as a function of facilitating timely prevention of cascade development or prevention.
This same approach, using the adaptive processor 304 communicating with one or more centralized processor 304s over the national or international network to continuously or intermittently adjust and improve the testing components of the MPPC may also be applied to the treatment components of the MPPC to determine evidence that a given treatment had a measurable effect (positive or negative) on the MPPC. Treatment components include, for example, drugs, fluids, nutrition, surgical treatments, inhaled gas treatments, pressure treatments, rehabilitation, exercise, positioning, splinting, to name a few. The MPPC provides the images of the treatment relationships, including relationships related to timing of the treatment order, treatment delivery, treatment done or procedure, the pattern of the individual treatment dose, the pattern of dose administration over time, the method of administration, and other images for comparison with the images of the rest of the MPPC. New treatments such as new drugs, different doses, different procedures, and/or the elimination of a treatment may be applied using the processor 304 to determine, as by the application of statistical comparison, the effect of the new drug on the MPPC in relation to the MPPC without the drug or treatment. In one embodiment, this is determined by the centralized processor 304, which then adjusts the protocol to produce the most favorable treatment regimen as determined by the processor 304 (and after approval by the physician group overseeing the processor 304). Locally at each hospital they may choose to accept or refuse the uploaded change from the centralized processor 304. In either case the subsequent MPPC of the refusing hospital may be compared by the processor 304 to the MPPC of those hospitals which accepted the uploaded changes into their processor 304 to determine if the change indeed had a positive, negative, or no impact on MPPC (including the expense components of the MPPC.
The use of one or more centralized processor 304s in communication with local processor 304s through a national or international Patient Safety Processing Network allows local healthcare delivery systems hospitals, and even different floors to use different testing and treatment protocols. This diversification is a great strength of the system because the centralized processor has a very wide range of alternative MPPCs to choose from in selecting, what it, (and the overseeing physician experts of the centralized processor 304 if desired) determines is the optimal treatment and testing protocols from the diverse sets of MPPCs (which include a divers set of treatment and testing images).
As discussed, according to one embodiment, the patient pafety processing network includes a set of local processor 304 located at hospital ward or unit. The Local processor 304 under the direction of the healthcare workers at that location. This allows the local healthcare workers to control the treatment and testing protocols, and variation of the testing bandwidth, deployed for the patient under their control. The local attending physicians individually or as a group as well as the hospital pharmacists and nurses may prescribe these protocols though the use of the Local processor 304. The local processor 304 records worker(s) (for example as a step time series or non-numeric time series in which a state transition event occurs when the physician, or nurse for example assumes responsibility and a different state transition event occurs when he or she is replaced by another. Alternatively, and possibly in combination, a state match event may be used to indicate the duration for which the physician or nurse was on duty. Those caring for the patient are therefore part of the Motion Picture of Physiological Condition (MPPC). Protocols may be decided by a group or by an individual physician caring for the patient. The extent to which a particular healthcare worker or group is statistically or otherwise associated with favorable or unfavorable MPPC may be assessed by the processor. The protocol choices for the local processor 304 may be made through the use of pre prepared MPPC protocols as previously discussed.
The local processor 304 may recognize the physician time-series and adjust the protocols and MPPC to match those selected by this physician. The physician may override the processor 304 and, if this occurs, this override includes at least one event and includes a new time series until the override is withdrawn. The extent to which a particular override is statistically or otherwise associated with favorable or unfavorable MPPC may be assessed by the Hospital processor 304. The Hospital Group processor 304, or the International processor 304. These may provide modifications in future protocols, and even provide incorporation of the modification of the override or the prevention of this type of override according to the MPPC associated with the override.
The local processor 304 throughout the hospital communicate with a central “Hospital processor 304” which is preferably under direction of the quality improvement committee and the hospital experts in each field. The Hospital processor 304 communicates with all the Local processor 304s and may be used to upload treatment/and or testing and/or bandwidth adjustment protocols and or comparison MPPC, which have been agreed upon for application hospital-wide to the local processor 304s.
Each Hospital processor 304 communicates with (and may be controlled by) central Organization processor 304. The healthcare “Organization processor 304” allows standardization of the hospital protocols through the Hospital processor 304s under its control to set minimum safety treatment and testing standards and may be controlled by a centralized quality assurance group with expert representatives from all of the hospitals. Since the individuals caring for the patient represent at least one time series and the ward represents at least one time series and the hospital represents at least one time series and the organization represents at least one time series. The MMPP at the Organization processor 304 therefore includes all of these caseworker and location time series for each patient. Alternatively, or in combination, if the Patient is wearing a monitored GPS unit, this may include a location time series, which provides continuous real time location as part of the MMPP. (The processor 304s may compare with the entered locations to the GPS location to produce a confirmatory binary.)
One embodiment demonstrates an example of how a new set of time series derived from a new test or testing device may be evaluated for cost effectiveness. In this example, a pulse oximetry reflectance probe is mounted (as by hat or headband or other fixation device one or both eyes to the patient's head and the probe is wirelessly or otherwise connected to pulse oximeter and the local processor 304 (as by Bluetooth for example). The transmitter may be mounted in the probe, or in or on the headband, hat, or behind the ear (for example, in the position of a hearing aid if desired). A position sensor may also be provided mounted on the patient. A maneuver such as a change in body position from supine to standing may be detected and included as a state transition or state match event by the processor 304 and a fall of a component of the photoplethysmographic pulse (indicative of the perfusion of the capillary bed distribution of the supra-orbital artery, a distal branch of the internal carotid) in relation to a maneuver to produce a divergent binary. In this way the flow of the capillary bed above the eye in relation to a standing maneuver becomes a surrogate marker of other capillary beds supplied from the internal carotidin relation to a standing maneuver. Real time perfusion may be compared with that of the ear, fingertip, or the pulse pressure (as by an invasive arterial line for example) to identify disparate in perfusion in one or both of the internal carotid distribution. The local processor 304 processes the MPPC with these as additional time-series. The local processor 304 uploads the MPPC to the Hospital processor 304, and the Hospital processor 304, Organizational processor 304, and/or International processor 304 where the MPPCs may be evaluated using the patient safety comparison processor to determine if, after adjusting for disparities in the MPPCs as a function of co morbidities and the like, the MPPCs which includes orthostatic reductions in supraorbital perfusion (in combination with therapy or corrective action to reduce falls upon the detection) is associated with a statistically significant decrease in the number of falls in the hospital. These time series ma; be automatically added (by automatically ordering the intermittent or continuous supra-orbital monitoring used the study) to increase the testing bandwidth when it is detected that the MPPC of a given patient is similar to those of the study population (for example as by type of surgery, age, co morbidities, or other events or images) where the addition of those processed time-series data had a positive impact on outcome. If one the other hand disparate hypo-perfusion above one eye is identified ultrasound evaluation for carotid disease may be ordered by the processor 304 and the result added as an event.
All the Organization processor 304s (or Hospital processor 304s if the hospital is not under a central organization) are preferably connected to an International processor 304. The International processor 304 is preferably controlled by a healthcare information corporation, such as Google or Microsoft, which maintains the international processor 304 and the network as part of a centralized repository of electronic medical records for example Health Vault™ by Microsoft). Each subordinate processor 304 (below the international processor 304) is capable of operating independent of the processor 304 Network so extensive redundancy, lack of subordinate dependency, and therefore greater safety against network failure is built into the processor 304 Network.
This processor 304 Network structure allows a wide range of minimum standards to be set by each government and allows the monitoring of the effects across the range of minimum standards to determine relative cost and benefit of individual standards.
Any domain-specific processor 304 will preferably include a patient safety comparison processor (PSCP) which compares the MPPC and all occurrences within the MPPC, such as events, binaries, images, and cascades, to other MPPCs and all of the occurrences of the other MPPCs to identify statistical differences between the MPPC which are associated with improved or adverse expenditure, outcome, length of stay, morbidity, mortality, resource consumption, and/or complications to name a few.
One advantage of the processor 304 is that the objects of the MPPC are discrete and are therefore readily incorporated into statistical software components of the PSCP. Specifically the processor 304 provides the ability to positively identify the existence of an occurrence within a patient's set of stream s given a specific occurrence definition and search window of time. In the contemplated embodiment, the process of identifying any of these occurrences is accomplished through a polymorphic image identification method. The unified method of searching and/or identification according to this embodiment, allows the incorporation of a range of statistical tools thereby providing access to the functionality of a wide array of statistical approaches and methodologies, as are well known in the art, for identifying differences in discrete time related data collections.
The PSCP provides further optimization through basic occurrence query capabilities. For example, a query may be crafted to include a specific occurrence, the domain (e.g. Local, Hospital, Organizational, International), and a set of conditions. The result set would provide patients along with the time spans in which the specified occurrence obtained. Aggregation of result sets provide an indication of the percentage of identification within the specified population (e.g. 434 (0.003%) patients within the specified domain (˜2.3 Million) meeting the conditions (234.046) were identified as having the specified occurrence).
The objects also include organized collections of an ascending hierarchy of complexity and the organized collections, which may be compared statistically at each ascending level of complexity to identify associated differences. In one embodiment the PSCP divides the MPPCs into groups having a least a portion of substantially the same images. For example a grouping may be derived having substantially the same initial sepsis cascade picture and similar co morbidities and age and sex but different physicians, hospitals and/or treatments. Differences in length, progression, compilations and mortality associated with the cascade may be identified and statistically compared with the differences in physicians, hospitals, treatments, testing, and/or treatment timing.
In one embodiment, the processor 304 is supported by an Online Transaction Process (OLTP) system to optimize data inclusion and relational management and image identification while the PSCP includes an associated Online Analytical Processing (OLAP) database optimized for the data mining with regard to occurrence presence and analysis within large populations, occurrence identification within a patient provides a primary measure while dimensions are established for time, patient characteristics personnel, etc. The multidimensional data environment allows for the inclusions of dimensions as data is available and/or determined as providing important data segmentation. Hierarchies (e.g. domain hierarchies, personnel hierarchies) are included to “roll up” aggregations.
When a particular testing, treatment, bandwidth variation, ward location, or hospital location is identified as statistically associated with improved outcome, then the International processor 304 may offer, as for download, new protocols which incorporate those identified particulars into the Hospital processor 304s and/or Organization processor 304s for their consideration. New medication or treatments may be assessed in this way with blinding accommodated by the processor 304 such that the time series of the experimental medication is labeled with an experimental code.
In one embodiment, the PSCP applies a top down approach to statistical analysis of objects wherein an expert or panel of experts at any level from the ward to the international level, defines the statistical comparisons, which they desire. This may be performed by a Probability Assessment Studio, for example as by drag and drop of occurrences (selecting relational timing ranges for the objects), such as events, binaries, images, etc. into a first window and then drag and drop of occurrences also selecting relational timing ranges for these objects such as a blood culture result, a diagnosis, a medication, a combination of medication and another object, to name a few. The processor may then provide the statistical comparison indicating the statistical relationship of the compared objects.
In one embodiment, the Probability Assessment Studio indicates Near Misses and their percentages as well as the probability of hits. The definition of a near miss must be specified within the occurrence definition. Here it is important to contrast as for example by multiple regression. The processor 304 is applied to allow individual and combined events, binaries, and images to be compared statistically to targets (which target may be each other). This mitigates the primary weakness of the application of multiple regressions to define relationally statistical values as a function of a plurality of different relational variables. While many reading the medical literature are compelled by the assumption laden mathematical embellishments, which commonly include many multiple regression equation based evaluations of complex relational physiologic data sets, a truly skilled mathematician is aware that the application of multiple regression equations to relational physiologic datasets will commonly produce outputs of dubious mathematical integrity and reproducibly. This is one of the advantages of statistical processing using the processor 304 which allows statistical analysis to be applied with a much wider discretion over a priori separation and combination of images.
According to one aspect of the present invention a Patient Safety Discovery Processor (PSDP) is provided at one of the levels but ideally at the level of the International Patient safety processor. Like the patient safety comparison processor, the PSDP compares (as for example by statistical analysis) occurrences such as events, binaries, images, cascades, repeating occurrences and patterned occurrences to other objects termed “highly definitive objects” such as final diagnoses, results of “highly definitive” lab tests (such as a blood culture, pregnancy test, or HIV test), length of stay, expense of care, number of care workers providing care, number of care workers providing care to others (workload of the care-workers), consensus based diagnosis, to name a few. However, in contrast to the PSCP, the PSDP applies a “bottom up” approach by performing statistical analysis on occurrences (potentially substantially all objects) in the processor 304 database to identify previously unknown relationships. These relationships may include statistically significant relationships with each other and especially with objects including at least one final result, diagnosis, and/or outcome.
In one embodiment The Patient Safety Discovery Processor may determine the probability that a first occurrence is associated with a highly definitive object and then determines the probability that a second occurrence which occurs after the first object. The processor then determines the probability that each consecutive object is associated with the highly definitive object. This generates a time series of probabilities with each probability data point being derived with each new occurrence type (which may be generated each time a new object is added inside the occurrence type definition or each time the probability of the occurrence of the highly definitive object changes as a function of a new object which is added to the definition). Here a plurality of probability time series may be derived, one or more for each highly definitive object. These time series of probabilities may be converted into objects and processed to detect and characterize events of rising or falling probability (termed probability events), by the objectification processor incorporated into the processor 304. In a presently contemplated embodiment, the baseline probability is established (pretest probability) based on non-volatile risk factors and then the occurrence type definitions are built (after the initial probability has been determined by time insensitive or pretest probability assessment) so that the slope of the probability time series is less dependent of non-volatile factors.
The patterns of the time series of Probabilities, (TSP) generated from sequential occurrences within an MPPC may exhibit characteristics, which provide more information then is provided by individual probabilities alone. For example, when a rise event (along the time series of Probabilities) is detected the slope and magnitude of the rise is known. A rise event along a TSP, which is of high magnitude and exhibits a rapid slope is said to define “probabilistic momentum” (PPM) which may be positive or negative. The presence of a high positive PPM provides strong evidence of the presence or future detection of the highly definitive object for diagnosis) under test. TSP may be generated by the processor for each diagnosis for which an increased probability is identified. Treatment of a pathophysiologic process such as sepsis (if effective) will cause the generation of positive reciprocation along the TSP indicating first a positive PPM, followed by a negative PPM. This is a desirable pattern, since if the negative PPM is maintained; this provides evidence that the catastrophic event has been aborted. Therapy, which is statistically associated with a positive probabilistic reciprocation along the TSP is identified by the International processor 304.
The first step for the researcher using guided image discovery is to identify a domain of patients (or patient-grouped time series sets) to which discovery will be applied (The aggregate of 1240 and 1242). Next, the researcher must attach each patient with a Boolean flag indicating the presence or absence of the condition or diagnosis to which the discovered images will be related. In an example the researcher may identify a definitive object, which includes a “stand alone diagnostic meaning” (as determined by a blood culture object or pathology object for example). In one embodiment, this definitive object is used to begin the process of identification. The processor 304 marks all patients that contain this definitive object as positive (placing them logically and/or physically into the Positive Patient set 1242), but the researcher may override this designation to weed out subsequently identified false-positives. The PSDP provides tools to make individual positive or negative designation or designation by the satisfaction of a rule-set. In this way, either explicitly, by rule all patients are placed logically (and/or physically) into one of two patient sets—the Positive Patient set 1240 or the Negative Patient set 1242.
Once all patients in the domain have been designated as positive 1242 or negative 1240 then the PSDP may begin to build images correlated to the condition or diagnosis of interest. The PSDP assumes access to the results of standard processor 304 and PSCP analysis against all patients within the positive and negative groups using the supplied event 1250) and binary 1252 definition sets. An initial image definition 1254 set may be supplied as well, but is not required.
The process of building statistically enhanced occurrence definition sets 1246 depends on the ability to construct occurrence definitions using occurrences (found in the positive patient set) and finding the statistical difference between their existence in the positive 1242 and negative 1240 patient sets. (As defined above, an occurrence may be an event, relational binary, image, repeated occurrence or pattern occurrence to name a few). This process is iterative and may begin with basic occurrences (e.g. events) and then moves to more and more complex occurrences (e.g. images, repeating occurrences and pattern occurrences). Each occurrence definition created is placed into the Candidate occurrence definition cache 1256.
In an alternative and complimentary embodiment, depicted in
For each candidate occurrence definition created the PSCP query engine 1244 is used to execute two queries, which are identical except for the domain (one for the positive 1242 domain, and one for the negative domain 1240) to determine the percentage of identification. These queries can, as shown above, at the researcher's request (or in a way automated by the processor 304) may be limited by any number of population, risk factor, preexisting condition, or other segmentation dimensions to identify statistically significant subpopulations.
In an example, a primary element in determining correlation may be the Distinct Patient occurrences, which may be defined as the number of patients in which the occurrence was identified at least once. Multiple occurrence identification is communicated to the researcher and may be weighted as an additional factor in the inclusion of an occurrence type. As well, when recurrence of the same or similar occurrence is identified, the PSDP may attempt to add a repeated or pattern occurrence to the working occurrence type.
The difference between the percentages returned from the above queries indicates the statistical significance index (SSI) of the occurrence. The researcher will set a threshold by which the PSDP may determine whether the occurrence type is retained or discarded. If the occurrence type is retained it is stored (as depicted for example by 1247) along with its SSI 1248 with regard to the specified condition or diagnosis. At the completion of the process a set of occurrence type definitions are stored and ranked by SSI. The occurrence definition group, along with their respective SSI designation, is stored in the statistically enhanced occurrence definition set 1246 as related to the specified condition or diagnosis.
In one embodiment, the PSDP begins with basic occurrences; occurrences, for example, including a single perturbation event or single relational binary. Once these images have been analyzed the statistically significant occurrence s definitions are stored, and the PSDP begins the process of creating more complex occurrence definitions. To accomplish this process, a plurality (such as three) key time span elements may be designated. Each time series involved ma) be specified with an indication the maximum phase shift of their data. The maximum phase shift indicates the maximum time delay from the real-time occurrence of the data and the recording of that data. Secondly, events and binaries may be designated with an indication of their maximum span of influence. This maximum span of influence may be defined in terms of real time (e.g. 30 seconds or 2 hours) or may be a relative identification (e.g. patient stay, 3 times event duration). A hierarchy may be provided to determine the contextual maximum span of influence. For example, a time series may be designated to have a maximum span of influence and therefore, unless otherwise overridden, substantially all events that occur along the time series will inherit the maximum span of influence from their time series, binaries, which are including events from two time series, will inherit the greatest of the span of influence of the two time series (unless specifically overridden). In some cases, when fine-tuning is required, the maximum span of influence may be designated as from one specific time series of the two. Further, a maximum span of influence may be designated within the relationship between two time series. In other words, a time series may be known to have a specific span of influence on another time series. The direction of this relationship may be specified as well, since the influence from one to another time series may not be the same as the influence of the second time series to the first. Individual events and binaries may also be designated as having a specific maximum span of influence. In one embodiment, when multiple maximum spans of influence are in effect for a binary or image, the greatest maximum may be applied. Finally a third key time span may be a single tolerance element that the researcher may indicate to widen the search.
In one embodiment, specifically for the construction of occurrences (e.g. binaries and images) the PSDP uses; these three time spans to determine the scope of search for elements to combine to produce the occurrence definition. In this embodiment, when the PSDP is looking for an element to combine with a given occurrence, the search looks for the overlap of the following time spans:
(OS−T) to (OE+T)
(ES−EMPS) to (EE−EMPS)+EMSI))
The PSDP uses this scope of search to identify all occurrences that should be combined with a working occurrence to determine if adding the new occurrence to the definition results in a statistically significant occurrence, in this example, the search is conducted within a single patient and only occurrences that exist within the positive patient domain are used. For efficiency, the researcher may further indicate that the occurrence must be found at least X number of times within the positive patient domain before creating a new occurrence definition to test against the negative domain. The analysis of the timing relationships among multiple variations of the image within the positive patient domain provides time interval data for determining the search windows within the occurrence definition itself, and an additional tolerance may be configured here as well. Once a new element has been chosen as providing the core for building a new occurrence definition, the occurrence definition is then used to query for SSI and to determine whether it meets threshold SSI so as to be part of the statistically enhanced occurrence definition set. Regardless of its inclusion in the statistically enhanced occurrence definition set, the new occurrence definition is stored to build additional combinations since the statistical significance of component occurrences is not always correlated with the statistical significance of a complete occurrence. In fact, a given occurrence component may predict the existence of a specific definitive object with modest probability, whereas once the complete occurrence definition has been built, the presence of the complete occurrence definition may actually predict that the definitive object will not occur and vice-versa. In this way, the PSDP continues to build more and more complex occurrence definitions to be analyzed statistically until no more combinations within the scope of search are identified.
To support the incremental creation of occurrence definitions, the processor 304 supports query capabilities to return potentially significant occurrences. This query returns a result set of components as follows: Given (occurrence definition to be Enhanced, Positive Domain); Return (Element) Where (Element is in Scope of Search) Further an aggregation query may be submitted as follows: Given (occurrence definition to be Enhanced and Positive Domain) Return (Element type, Count of occurrences Found, Count of Distinct Patient occurrences Found) Where (Element is in Scope of Search) Order By (Count of Distinct Patient occurrence)
During this analysis, the PSDP further maintains the definitions of antecedent relationships between occurrences identified. An evaluation of these antecedent relationships provides a further element in the statistical analysis. The PSDP may identify the most predictive (e.g. most statistically significant) paths of evolution of images along the timeline of the MPPC. This information will help refine protocols to be targeted not only images but image paths. The image (or other occurrence) with the maximum SSI may be designated as the primary image. Alternatively, multiple images (or occurrences) may be designated as primary the fact that they are the maximum SSI within their associated evolutionary path.
The PSDP may assist in the location of binaries, but more readily assists in the detection of certain binary types. For example. Expected and Analogous binaries may be identified with the PSDP, but Verify Non-Existence binaries require much more processor power since the PSDP uses non-existence of an event as a criteria for search, which greatly extends the scope of the search. In another instance the existence of events, which are similarly or identically, predictive may suggest the definition of an analogous binary, an analogous image, and an indication of potentially redundant or superfluous testing and the PSDP may provide this indication to the researcher.
In some cases, if a binary is identified as not predictive, an event (or both events, within that binary may be identified as predictive. Here it may be seen that one or more relationships between events may be more predictive then the presence of the events. This is one of the reasons that conventional statistical regression as applied to even a very wide range of detected events (without building the relational binaries and more complex occurrences (e.g. images) and more specifically without considering the complex relational patterns and the timing of the patterns of the events in relation to each other) is incomplete and potentially misleading.
Because the PSDP is very sensitive to event segmentation, the researcher may instruct the PSDP to break individual event definitions into a continuum of event sub types (as, for example, pertaining to slope, magnitude, threshold breach, components of the definition, to name a few). The PSDP will search both on the original event type and on each of the event subtypes to look for a threshold breaking SSI. Once this process is completed and the statistically enhanced occurrence definition set is available for use in the processor 304, the processor 304 may search for all occurrences within the statistically enhanced occurrence definition set. Once all occurrences have been identified, the processor 304 may obtain the Maximum SSI (MSSI) among all of the identified occurrences. Further, the MSSI for specific conditions and diagnoses under test then may be added as additional parallel time series and available for analysis by the processor 304 and visualization in the patient safety visualization processor, events within these MSSI time series are of particular significance. An up event in an MSSI indicates the movement of the system toward a positive predictive identification of a condition or diagnosis. From the slope of a time series of statistical values and the instantaneous statistical value of that event is derived Probabilistic Momentum within the processor 304 and patient safety visualization processor. According to an embodiment of the present invention, a high product of the positive slope, wherein the outputs of the testing used to derive the slope are time sensitive (i.e. the outputs change with time as a function of the presence or absence of the disease state or the presence of the definitive object, provides additional predictive indication beyond the that of the instantaneous predictive indication. The presence of high probabilistic momentum is highly predictive and may also provide an indicate acuity with certain disease states. The patient safety console may be configured to show the parallel time series of MSSI for the top X diagnoses or may be configured to show the MMSI with the greatest statistical momentum. Parallel probability time series; may be generated by the processor for each diagnosis for which an increased or reduced probability is identified. Treatment induced reversal of a pathophysiologic process such as sepsis will cause the generation of positive reciprocation along the time series of probabilities for sepsis.
With retrospective analysis, Probabilistic Momentum toward Primary Diagnosis is a Key Performance Indicator (KPI). The increase of this KPI may be used to determine the effectiveness of a diagnostic environment. Elements (e.g. diagnostic tests) will be able to establish efficacy by demonstrating the increase Statistical Momentum toward Diagnosis. With Probabilistic Momentum toward Diagnosis and Cost added as measures in the PSCP multidimensional database the PSCP will be able to demonstrate the cost-effectiveness of diagnostic tools. In one embodiment, the PSDP utilizes the process of Retrospective Real-time analysis (RRA) to further investigate probabilistic momentum. RRA uses a time-slicing technique to examine a set of time series, and the associated object stream s, in the way they would have presented in real time. To accomplish this, the processor 304 creates a set of MPPCs, each of which represents what the MPPC would have been at a specific point in time. In other words, the processor 304 selects a point in time and truncates the MPPC to only include the data that existed at that point in time and before. In this way, and in combination with the true final MPPC, partial occurrences may be analyzed to investigate their evolution and the definitions that may be employed to identify them. The PSDP may compare a partial occurrence with the final occurrence to look for indicative elements. As well, the PSDP may compare partial occurrences across a wide range of patients and conditions. For example, the PSDP may determine partial occurrences that tare very similar and that will complicate early detection and look for key elements that differentiate them.
In this manner future binaries, images, and cascades are constructed from the bottom up as more and more patient data sets are available. In essence, the entire pool of patients monitored by the processor 304, include experimental datasets whereby positive or negative statistical associations may be determined automatically and new relationships, disorders, predictive sets of testing, and superfluous testing, effective treatment, superfluous treatment, ineffective treatment and harmful treatment may all be identified from the bottom up as a function of statistical relationships.
The application of Probabilistic Momentum evaluation exploits one of the most important features of living organisms; the time dependency of pathophysiologic processes and of physiologic relationships. As previously described, perturbations and the pathologic or physiologic responses of organisms to perturbations are relationally time dependent. Each living organism is a specifically structured chemical, electrical, and mechanical entity. The relational structure of each of these components of the organism defines the timing for chemical, electrical, and mechanical action. Therefore actions, reactions, and failures of this structure sill occur with order within definable limits as a function of the relational structure, which will define the temporal relationship limits of actions, reactions, and failures.
In the presence of a pathophysiologic cascade, increased probabilistic momentum may be induced by a more rapidly progressive pathophysiologic cascade but may also be induced by more rapid data sampling, broader bandwidth of testing (more testing per unit of time), and/or better focused testing. Probabilistic momentum is therefore both a function of the rate and magnitude of pathophysiologic cascade progression as well as the quality and scope of the testing applied to render the parallel time series. When the testing and expansion of the bandwidth is the same, a higher degree of probabilistic momentum suggests a greater rate of pathophysiologic cascade progression.
In one embodiment, a system and method according to the present invention is provided wherein, events, binaries, and images are defined by time aggregations wherein event aggregations are primarily or solely) based on time (as by windowing for example). This exploits the time dependency of physiologic systems and of pathophysiologic processes and is simpler than the rendering of binaries images, and/or cascades as previously discussed. Such time aggregations may be used to complement the more complex building of the motion pictures of physiologic failure. In addition binaries, images, and/or cascades may be derived from a time based aggregation (as by combining all permutations for specific sets of permutations) of the event objects. Alternatively, or in combination, events binaries, and images may be additionally or similarly aggregated by magnitude, slope, and/or pattern and/or type or other characteristic.
According to another aspect of the invention each time relationship may be converted to an object such that at east one portion of an event, binary, image, and/or cascade is comparable with another portion of an event, binary, image, and/or cascade. In an example, the time relationship of at least one portion of alpha event is compared to at least one portion of the beta event to derive at least one intra binary temporal relationship (which may be an object or a characteristic of the binary object). The intra binary temporal relationships (or another temporal relationship) include values, which may be converted to a time series, objectified and incorporated into the MPPC as desired. Similarly this may be performed for images and cascades. In an example, acceleration of the cascade may thereby be readily identified.
In one embodiment the processor is programmed to detect a sentinel event object (such as a rise in WBC count, a rise in respiration rate, or a rise in temperature). The processor then identifies all event objects occurring within a retrospective and prospective time window of the detection of the sentinel event. In an example, upon the detection of a rise in temperature the processor may aggregate all event objects, which occur within 48 hours after the temperature is elevated, and this aggregation may be compared to aggregations (built and/or derived from statistically derived aggregations in relation to definitive objects.) The primary relationship rendering the aggregation of these event objects is time but the event objects within the aggregation may be comprehensively compared.
With a time aggregation, upon the addition of each new event, the probability of the future occurrence of a plurality of definitive events may be determined. As discussed, Probabilistic momentum is derived when a plurality of time sensitive (volatile) events, each adding greater probability is detected and the time series of the probability of a plurality of definitive events may be plotted, objectified and analyzed.
In another embodiment, the processor repeatedly cycles (which may be a cycle of windowing) through the time series looking for (for example) objects (such as perturbation events). Each time an object is identified it is placed (for example) in a perturbation event set if this is the object type being detected. Consecutive sets are derived for each cycle. The set derived from each cycle may include no perturbation events or hundreds of perturbation events depending on the state of the organism. This approach essentially aggregates objects (events) by cycles (the cycling frequency may range for example from about 0.01 second to about 1 hour depending on the definitive object). Consecutive sets of perturbation events include objects and therefore have the characteristics of the objects, as well as the number of the objects, and the probabilities associated with the objects. These characteristics may be used to derive various time series, which may be objectified and analyzed. In an example, if along a time series of perturbation events, the number (and/or magnitude) of perturbation events shows a high positive slope, but none of the time series of the probability for any definitive process shows a high positive slope, this suggests that the processor is failing to identify (and may not timely identify) the cause of the perturbation and this may provide a warning that intensive expert physician diagnostic evaluation is timely required. The processor 304 in this case is saying, the patient's physiologic system is progressively more perturbed, and I am not identifying a likely cause at a rate matching the growth of the perturbation and the patient needs expert human brain help soon.
Perturbation object sets may be derived globally (from all available time series) or may be derived from a focused set of a particular group of time series (such as time series which relate to inflammation for example). The relative pattern of growth or decline of each different set may be compared by objectifying selected time series of characteristics of the sets. In an example, during sepsis, a set of inflammation perturbation objects may grow rapidly (with an increasing number and/or magnitude of perturbation objects and or decreasing time interval between increases in number and/or magnitude), and then this growth may be followed a rapid growth of a hemodynamic perturbation object set. The comparison of objectified time series of the sets renders parallel rise objects producing a binary with the inflammation set rise including the alpha event and the hemodynamic set rise including the beta. Upon detection, the occurrence of a respiratory set rise may then be added to produce an image (derived of time series of characteristics of sets) and finally with additional sets a motion picture of physiologic failure derived of time series of characteristics of sets (wherein the characteristics may for example include number of perturbations, magnitude of perturbations, frequency of perturbations, slope of perturbations, probability of a given definitive object to name a few).
One embodiment provides an example of the continuous derivation of specific objects (such as events, binaries and images) between two or more time series to both corroborate the significance of an object as well as to exclude perturbation due to artifact. In an example, it is well known that motion may cause artifactual desaturation events along an SPO2 time series. However, motion may also occur in response to a desaturation event (particularly in response to the arousal from the desaturation). The proximate existence of arousal motion corroborates the SPO2 event, binaries and images and helps to establish that the patient was sleeping or sedated at the time of the desaturation. Motion inducing artifactual desaturation and arousal motion responding to a true desaturation are generally different in relational timing, relational spatial pattern, and the relational frequency pattern and characteristics. After adjusting for any phase shift, a arousal motion object inducing artifactual desaturation object will generally early occur at the same time or precede the onset of a desaturation object. Whereas arousal motion object will occur after the onset of the desaturation object or adjacent the recovery from the desaturation object. In addition after adjusting for any phase shift, arousal motion objects commonly occur in a specific pattern, which mirrors and/or corresponds, in at least one aspect, with the pattern of SPO2 objects, with the pattern of desaturation objects within the pattern of the SPO2 objects preceding the pattern of motion objects. Furthermore motion arousals within a cluster are usually brief (about 2-6 seconds) and then followed by a period of little or no motion. In one embodiment the high frequency components of the frequency spectrum of the plethysmographic pulse waveform is used to indicate motion or another known method is used to either indicate threshold motion as a step function time series (as is for example outputted by the Minolta 300i pulse oximeter) or preferably a graded motion time series between the values of 1-10 is provided.
The following provides an example of an occurrence of an image of “Hypoxemia Induced Micro-arousal” (HIM), which is commonly indicative of sleep apnea induced hypoxemia. In one embodiment the HIM image is derived from a motion time series (as generated using detection of motion from the plethysmographic time series or by the use of atigraphy or another method), a SPO2 time series, and a pulse time series (such as a plethysmographic pulse time series). All of these time series may be generated by a patient mounted pulse oximeter. The processor 304 detects a SPO2 event, binary, or image within the span of influence of the motion event, binary, or image and the pulse event, binary, or image within the span of influence of the SPO2 event, binary, or image. Together, the onset of a fall in oxygen saturation followed by the onset of a brief episode of motion (e.g. 1-6 seconds), a rise in heart rate (and/or another property of the pulse time series) and a rise in SPO2 produces an occurrence of an image of HIM. In some cases the motion event, binary, or image may be indicative of pathology as a function of their characteristics alone. In an example the detection of motion occurring in a periodic cluster pattern typical of a cluster of HIM induced by a cluster of apneas as defined for example by objectification, FFT processing, by a combination of frequency domain and spatial domain processing, or another method) whereas at other times the decision to reject a given SPO2 event, binary, or image may be made with consideration of the relational timing and relational pattern of the motion and the other signals. In an example the presence of a cluster of motion reciprocations in relational combination with a cluster of SPO2 reciprocations provides stronger evidence of the presence of a cluster of HIM. This is further supported by the detection of a prolonged period of motion (for example 15 seconds or more at the end of a SPO2 reciprocation cluster (of high SPO2 desaturation magnitude). This terminal motion is typical of a hypoxia-induced awakening (HIA) induced by the cumulative effect of a cluster of severe apneas. This is dangerous image to identify in a patient being treated with patient controlled analgesia because the awakening may trigger the perception of pain causing the patient self medicate at a vulnerable time. The processor 304 may therefore be programmed to detect the combination of image of HIA in combination with PCA treatment and to send an indication to the PCA device to prevent a patient bolus from the PCA device upon the detection of HIA and during an interval (for example 30 minutes) after any image of HIA has been detected. The detection of HIA in the presence of PCA if followed closely by a self-administered bolus would provide strong evidence of overmedication (or under treatment of the sleep apnea).
One method for inclusion of a SPO2 pattern as non-artifactual includes comparing the SPO2 time series to a time series indicative of motion and determining the SPO2 time series as non-artifactual based on both the timing and characteristics (which may be a spatial and/or frequency characteristic) of perturbations along the SPO2 time series to the timing and characteristics (which may be a spatial and/or frequency characteristic) of perturbations along the motion time series. In one embodiment this may include identifying a desaturation as non artifactual by identifying that at least 4% of the desaturation occurred before the onset of threshold or trend of perturbation along the motion time series or that at least ⅓ of the desaturation magnitude occurred during a period without motion. Another method for inclusion of a cluster of desaturations as non-artifactual includes identifying a cluster pattern of desaturations, identifying a cluster pattern of brief motion events wherein the cluster patterns correspond in frequency or wherein the motion events correspond to a greater extent to the recovery than to the desaturation. Another method for identifying a desaturation or cluster as non artifactual includes transforming the plethysmographic waveform into frequency components (as for example by FFT), identifying the desaturation or cluster as non artifactual if the frequency spectrum shows a higher degree of high frequency components (or a broader bandwidth) during the recoveries from the desaturation than during the first portion of the desaturations.
In another example time series data sets of the slope of the recoveries, the magnitude of recoveries, and the recovery magnitude ratio (magnitude rise/magnitude fall of each reciprocation), duration ratio (duration rise/duration fall), area ratio (area above or below the curve rise/area above or below the curve fall), of each reciprocation) of the SPO2 is generated and compared to a time series of narcotic infusion. Narcotics may both diminish the slope and magnitude of the recovery and may increase the magnitude of the fall and may reduce the recovery ratio. Therefore according to one aspect of the present invention a perturbation pattern or threshold change in one of these values (such as a 50% fall in the mean slope of the recoveries for 10 minutes may be used to trigger an indication of possible excessive narcotic treatment or to lock out the PCA.
These provide examples of relational images at the event, binary, and image level, along with specific properties and relationships which are defined in each micro domain and added to the processor 304. In this example the images of sleep apnea encapsulates the images of HIM, HIA, and motion-induced artifact, which represent a few of the basic images of the comprehensive processor 304.
In an example, the processor 304 may generate and output a cycling severity index (CSI) for the entire night) by the formula 1 and 2:
Ts=Tt−Tm 1.
CSI=[(Tc/Ts)/(Sum Tr)/Tc]mean delta SPO2 2.
WCSI=[(Tcw/Tw)/(Sum Tr)/Tc]mean delta SPO2 3.
CSI(baseline SPO2−mean recovery SPO2) 4.
In one embodiment the objects derived from narcotic and or sedative infusion is combined with the objects of recovery to produce an image indicative of drug associated recovery failure wherein, for example, the image contains an occurrence of drug infusion followed by an occurrence of a fall in recovery peak values and/or a fall in recovery slope values. The presented embodiments represent a few examples of patient safety processing technology applying the objectified time-series matrix. As noted earlier when a new micro domain is added to the matrix the question may be asked, given the addition of the new occurrences within this micro domain, does any new image within the matrix have a greater probability of being associated with a definitive object then without the addition. Many new significant relationships and measures will be identified using this method or by automating the process to provide adaptive processor 304 processing wherein the processor 304 builds and cycles though a broad range of micro domains looking for occurrences which, when added to the matrix change the probability in a significantly positive or negative way.
Patient monitoring data, including the constructed Objectified time-series Matrices flow this direction into the Aggregate patient safety image database 1212 increasing the available data against which research and other reporting activities may be conducted. This data flow may be facilitated by database synchronization, message queuing, EDI, an Enterprise Service Bus, Asynchronous Web Services, Business Process Management software or other enterprise orchestration servers to name a few. In an example of a embodiment, the Centralized image Data center 1210 contains three databases: the Aggregate patient safety image database 1212, the safety image definitions Database 1216 and the Patient Safety Data Warehouse 1214.
The Aggregate patient safety image database 1212 contains the aggregated data from the Patient Safety images Databases at any number of medical facilities. The Patient Safety Data Warehouse 1214 is a derived database generated by analysis of the Aggregate patient safety image database and is an Online Analytical Processing (OLAP) database optimized for the data mining with regard to occurrence presence and analysis within large populations.
In an example of a embodiment, the safety image definitions Database 1216 is a centralized repository of occurrence definitions (for example image definitions) that research has determined to be most accurate with respect to probabilistic momentum. These may be accessed, as described above, by any number of facilities to enhance their ability to identify and treat conditions, occurrence definitions added, changed or deleted from the Safety image definition Database 1216 may be distributed to the medical facilities by individuals or committees specializing in this distribution or by predefined transmission protocols.
The Research Facility 1220 depicted in
The first section is the Selection Bar 1324 at the top of the screen. As configured this section includes the ability to select and the display of the occurrence type 1328 and Selected occurrence 1330. The selected occurrence type specifies the occurrence definition that the researcher is viewing and/or updating. This occurrence type may be one of the simpler types (event or binary) or may be a more complex type (image, repeating image or Pattern image). The occurrence type may be a candidate type which has not been persisted into the occurrence definition set. The occurrence editor also allows for the Researcher to create Derivative types in which he/she begins with the definition of a different type and makes changes to create a new type.
The Selected occurrence 1330 allows the researcher to choose an occurrence within an available occurrence stream to be a reference as he/she works on the definition of the occurrence micro-domain. Once a Selected occurrence is chosen, it may be displayed (with all or some of its constituent parts) in the bottom (fourth) section of the screen. This allows the researcher to immediately see the results of changes in the definition within a reference case. The Researcher may switch between several reference cases during the course of editing. The second section of the screen is made up of several subsections: The occurrence type Explorer 1332, the Characteristics Box 1334, the Other Participation List 1336 and the Dependency Viewer 1338. The occurrence type Explorer is the primary element of this section and all of the other subsections relate to it and to the occurrence type that is selected within it (e.g. oximetry reciprocation in
The other three subsections provide information as follows: The Characteristics Box 1334 provides a list of characteristics of the occurrence type selected in the occurrence Explorer. These characteristics include the Name, type (i.e. is it an event, binary or image to name a few), visibility (i.e. what domain does this apply to) to name a few. This section may contain workflow details as well to allow researchers to understand the state of the occurrence definition (e.g. what researcher created it, when it was created, whether it is approved or under review etc.) The Other Participation List 1336 displays a list of occurrences in which the occurrence type selected in the occurrence Explorer is a participant. For example, as shown in
The third section of the screen is the occurrence definition Diagram 1340 where the researcher draws (e.g. through drag-and-drop) the diagram that represents the scope of the occurrence. This section may be configured using the Phase Dropdown 1341 to specify the Phase (e.g. Scope Identification. Element Construction, Scope Refinement to name a few) to which the diagram applies. The View Dependencies button 1342 allows the researcher to see the dependencies on which the displayed diagram depends. The diagram itself is displayed on the Construction Surface 1344, (The diagram shown in
The fourth and final section (as configured) is the time series Section. This section contains occurrence stream s 1346 and Associated point streams 1346 (e.g. signals). These streams are added ether when a specific occurrence is selected (Using the Selected occurrence functionality 1330) or individually through the Add stream button 1348. The stream s are shown parallel in time. If an occurrence stream is added, the occurrence definition editor may ask if associated occurrence streams and/or point streams should be added.
This section allows the researcher to examine real data while constructing the micro-domain that they are working on. Changes to the occurrence definition may be immediately reflected in the time series section so that the researcher may tune the definitions to reference patients. It multiple patients (or patient stays) are required the time series section may be split into multiple sub-sections to allow for groups of stream s to be coordinated in time. In this way, the researcher may want to view a positive case and a negative case (perhaps with similar, but misleading patterns) so as to refine the occurrence definition with precision.
The occurrence definition editor is built to be one part of a rich set of tools that the researcher may use to refine occurrence definitions. For example, the occurrence definition editor links to the Patient Safety Comparison processor to run statistical analysis (for example to determine SSI) on reference patient sets at the researcher's request. In this way, the researcher may immediately assess the statistical impact of changes to the definition.
Further, the occurrence definition editor may be employed along with the image construction processor during guided image discovery. In this way, the automation (e.g. the location of images that may be statistically significant) provided by the processor may be directed and refined by the researcher
The property editor is described in detail in
As configured in
The second section of the screen is made up of several subsections: The occurrence model View 1376, the Characteristics Box 1380, and the Dependency Viewer 1382. The occurrence model View is the primary element of this section and all of the other subsections relate to it and to the item that is selected within it (e.g. instability index Calculated property in
The Characteristics Box 1380 provides a list of characteristics of the item selected in the occurrence model 1384. These characteristics include the Name, type (e.g. Calculated property as shown in
The third section of the screen, in the depicted configuration, is made up of three subsections: the property Expression Box 1340, the Evaluation Result Pane 1392 and a second Dependency Viewer 1396 connected to the Working property definition. The property Expression Box 1340 and the contained property Construction Surface 1394 provides an editing space where the researcher creates the expression (either directly or through drag-and-drop) that defines the Working property. The type of the property may be selected from the type Dropdown 1388 for example Calculated property and attribute to name a few). When the type of the property is changed, the Construction Surface 1394 is reconfigured to the property type. For example, if an attribute is selected then the Construction Surface is broken up into two parts: the Boolean condition and the Value Expression. The Dependencies View 1396 allows the researcher to see the dependencies on which the Working property depends.
The fourth and final section (as configured) is the time series Section. This section contains occurrence stream s and Associated point streams (e.g. signals) as in
In an example of a embodiment, a dependency diagram contains a single Root Element 1420 which is an icon representing the element which is being evaluated as regards dependency. In some cases, the Root Element 1420 may be omitted (for example, it is clearly implied within a software user interface). From the root element lines stretch downward to show dependency. In an alternative embodiment (not shown in
A dependency diagram may be collapsed into a dependency list. A dependency list simply lists all of the dependencies, but does not show the hierarchical relationship. The diagram in
Dependency diagrams and dependency lists may be filtered. Sample filtering include (to name a few):
In the presence of patient data (e.g. a selected occurrence stream and occurrence) a dependency tree may be enhanced. For example, for elements that may be evaluated down to a single value (e.g. a property or Rule) may display the evaluated value. Elements that represent scope elements (e.g. an occurrence type) may provide a hyperlink into the actual occurrence that were used to meet that specific scope requirement. In some cases, such as a repeating occurrence, the dependency diagram may be expanded into an instance diagram. In this case single dependencies are expanded out to a list of all of the occurrences that were found. In this way, the instance diagram creates rapid access to the elements within the selected occurrence. This may become very useful in very complicated images, particularly when the image spans a fairly large amount of time. The selection of individual nodes within the instance diagram allows the user to navigate to the element within the occurrence and/or point streams and shifts the parallel time view to be appropriate (e.g. at the right scale and time location).
In another view, of interest to re-searchers, the dependency tree may be enhanced to show the statistical significance index (SSI) of each node. In this way, the researcher may recognize the statistical dependencies. If, for example, a root node has an equal or lower SSI than a child node (with reference to a specific condition) then the researcher may want to abandon the more complex image for a simpler version represented by the node.
The occurrence Pattern definition Bar 1480 contains a Phase dropdown 1482 and the View Dependencies button 1484 with the same role and functionality as in
According to one embodiment for generating a more comprehensive time series matrix for analysis by the processor 304, one lumen of the double lumen cannula can be connected to an oxygen source and the other lumen connected with a pressure and/or flow sensor so that a time series of the nasal pressure can be generated along with a time series of the oxygen thereby and these can be combined with the various time series from the pulse oximeter. For ease of wearing such a cannula can be a short connected to a nasal pressure sensor and pulse oximeter worn about the neck (as for example configured like a bolero) where the cannula has a terminal is connectable to another tube providing connection to a source of oxygen. In one embodiment a single lumen catheter is used which bifurcates at the neck having a one connection for nasal pressure and one for the oxygen source. The presence of pressure induced by the continuous oxygen flow and changes in flow is detected as the DC pressure component in the tube upon which is superimposed the AC component of the nasal pressure signal.
According to another embodiment the processor 304 is applied to search for evidence of early ventilator associated pneumonia (VAP). The time series matrix of the datasets from the ventilator, for example; minute ventilation, I:E ratio, respiratory rate and effort, FIO2, oxygen consumption (if available), CO2 production, exhaled CO2, trend of triggered vs spontaneous breaths, exhaled gas components, to name a few. These may be further combined by the processor with time series of cardiovascular monitors for example heart rate, pulse pressure (and/or pleth) variability, pulse upstroke, pulse pressure, and maneuver responsive pressure variability to name a few. These may be further combined with time series of laboratory data such as Neutrophil percentage and/or count, stab (band percentage) and/or count, inflammatory markers, sputum parameters, and time series of other physiologic measures or scores such as the SPO2, temperature. These may be further combined with a time series of other vales such as sputum or BAL score or values, confusion score, and radiographic scores and/or scales. The finding of a cascade of inflammatory—respiratory—and/or cardiovascular augmentation in the absence of other sources of infection and/or in the presence of a rising sputum volume or purulence score is strongly suggestive of ventilator associated pneumonia (VAP). According to one aspect of the present invention patients managed with mechanical ventilation are monitored by a patient safety processor 304 which generates a data sets of at least respiratory data and inflammatory data and intermittently or continuously searches at least both data sets for relational patterns suggestive of VAP and provides an output upon detection of a relation pattern suggestive of VAP. The relational pattern can be a cascade comprised of linked or aggregated perturbations/trends/threshold breaches along a time series matrix.
Objectification is an example of a time series processing method which may be employed by the processor 304 to render the time series matrix freely searchable. As noted, objectification converts the time series matrix into stream s of discrete sequential and/or overlapping objects of ascending complexity, in 3 or more axes allowing each time series and the objectified matrix itself to be readily searchable in multiple dimensions. An example of the process of time series searching employing objectification is provided wherein as a first step to objectification; the processor 304 aggregates a set of time series together into a search vector. The combination of an occurrence type and a search vector defines the data from which a single occurrence stream is derived. A search vector is defined by a time window (whether explicitly or as a function of specified elements such as “patient stay” or “post-operative observation period+1 day”) and a set of time series that apply, by definition or by rule, to the tracking of a phenomena within the system from which the time series are derived. For example, a search vector may be created for respiratory rate and/or amplitude for the hospital stay. The processor 304 can be directed to aggregate all signals from which respiratory rate and/or amplitude can be derived. For example, the processor 304 can use a signal from a pulse oximeter or may use the chest impedance signal as derived from an EKG signal and/or from a double lumen pressure sensing nasal cannula, and/or a sound sensor on the chest or airway, and/or another source of data.
In this way a search vector may include any number of time series. The processor 304 may be directed to “prefer” certain signals such that if a certain signal is found then other signals are ignored. Alternatively, the processor 304 may be directed to aggregate all available signals and allow them to overlap. In this way, a search vector may include overlapping time series. For example, if a patient is wearing two different oximeters, the processor 304 may be directed to include both of these into a single search vector. When there are overlapping time series within a search vector, the identification of an occurrence at the same time (or within a reasonable phase shift specified) is counted as one occurrence for the purpose of search and/or statistical analysis. In this case, both objects are stored and a relational binary is created between them (typed as, for example, “equivalent occurrence”). The processor 304 may identify one of the occurrences as “primary” and all others as secondary such that statistical analysis and search may not duplicate the multiple recording of a single event within the system being monitored, in cases where an equivalent occurrence is expected to be found, but is not, a divergent binary may be created aggregating the found occurrence and the search region in which the expected occurrence failed to be found.
In one embodiment a search vector is meant to accomplish several things including the following three things. It may allow for the aggregation of time series from the same source where there have been interruptions of the recording of data. It may allow for the aggregation of times series that are from two monitors that are of the same type or which have multiple leads. It may allow for the processor 304 to utilize different sources that represent equivalent observation of the same phenomena within the system. On the other hand, in one embodiment of the processor 304, the search vector is not used to aggregate two time series that represent distinct systemic response to phenomena within the monitored system. The identification of multiple reactions to the same phenomena may be handled by separate but parallel search vectors and allows for the identification of complex occurrence aggregations (such as images) and their evolution which will generate probabilistic momentum (explained above).
While the disclosed embodiments may be susceptible to various modifications and alternative forms, specific embodiments have been shown by way of example in the drawings and have been described in detail herein. However, it should be understood that the disclosure is not intended to be limited to the particular forms disclosed. Indeed, disclosed embodiments may not only be applied to clinical diagnosis of systems of physiological failure, but may be applied to any clinical condition that may be represented by images as provided herein. Indeed, the disclosed embodiments may be applied to monitor and/or diagnose conditions in which a patient's condition is generally improving, such as post-surgical monitoring. Rather, the disclosure is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the disclosed embodiment and as defined by the following appended claims.
This application is a continuation of U.S. patent application Ser. No. 13/844,381, filed Mar. 15, 2013, which is a continuation of U.S. patent application Ser. No. 12/437,417 filed May 7, 2009, which claims the benefit of U.S. Provisional Application No. 61/126,906, filed May 8, 2008 and this application also claims the benefit of U.S. Provisional Application No. 61/200,162, filed Nov. 25, 2008, the disclosures of which are hereby incorporated by reference in their entirety for all purposes. This application is a continuation-in-part of U.S. patent application Ser. No. 12/437,385 filed May 7, 2009 entitled “Medical Failure Pattern Search Engine” the disclosure of which is hereby incorporated by reference in its entirety for all purposes. This application is also a continuation in part of U.S. patent application Ser. No. 12/629,407 entitled “Microprocessor System for the Analysis of Physiologic and Financial Datasets,” filed Dec. 2, 2009 which is a continuation of U.S. patent application Ser. No. 10/150,842, entitled “Microprocessor System for the Analysis of Physiologic and Financial Datasets,” filed May 17, 2002, now U.S. Pat. No. 7,758,503, the disclosure of which is hereby incorporated by reference in its entirety for all purposes, which claims the benefit of U.S. Provisional Application Ser. No. 60/291,687 filed May 17, 2001, the contents of which are hereby incorporated herein by reference and U.S. Provisional Application Ser. No. 60/291,691, filed on May 17, 2001. This application is also a continuation in part of U.S. patent application Ser. No. 11/369,355, entitled “Centralized hospital monitoring system for automatically detecting upper airway instability and for preventing and aborting adverse drug reactions”, filed Mar. 7, 2006, which is a continuation of U.S. patent application Ser. No. 10/150,582 entitled “Centralized hospital monitoring system for automatically detecting upper airway instability and for preventing and aborting adverse drug reactions,” filed May 17, 2002 now U.S. Pat. No. 7,081,095, which claims the benefit of U.S. Provisional Patent Application Ser. Nos. 60/291,691 and 60/291,687, both filed May 17, 2001 and U.S. Provisional Patent Application Ser. No. 60/295,484 filed Jun. 1, 2001, the disclosures and contents of which are incorporated by reference as if completely disclosed herein. This application is also a continuation-in-part and claims priority to U.S. patent application Ser. No. 12/152,747 entitled “Pulse Oximetry Relational Alarm System for Early Recognition of Instability and Catastrophic Occurrences,” filed May 16, 2008, the disclosure of which is hereby incorporated by reference in its entirety and for all purposes.
| Number | Date | Country | |
|---|---|---|---|
| 61200162 | Nov 2008 | US | |
| 61126906 | May 2008 | US | |
| 61200162 | Nov 2008 | US | |
| 61126906 | May 2008 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 13844381 | Mar 2013 | US |
| Child | 14535838 | US | |
| Parent | 12437417 | May 2009 | US |
| Child | 13844381 | US |
