Patent Application
20130080187
This disclosure relates to healthcare at medical facilities and to the documentation and editing of medical records.
In the medical field, clinical documentation is of paramount importance for monitoring patient well being and accurately representing medical conditions to insurance companies, governmental agencies, or other payers. The emergence and use of electronic medical records and electronic documentation can help clinical documentation processes, but may present many challenges.
This disclosure describes systems and techniques for processing medical data via one or more computers. The systems and techniques may be used by a medical reviewer (sometimes referred to as a “documentation specialist”). The techniques and systems described herein can help to automate (or partially automate) the coding process associated with medical record review. In this manner, the process can be improved and/or simplified. The techniques may apply one or more rules to define when documentation for medical records is sufficient and when further review of the medical records is needed. The rules may define when the documentation specialist should review the medical records and may automate the process by avoiding the display of medical records to the documentation specialist when the documentation in the medical records is sufficient. The rules may further define when physician review is needed, and may automate the process by avoiding physician review when the documentation in the medical records is sufficient or when review by the documentation specialist may suffice prior to (and possibly in lieu of) physician review. Machine learning techniques are also described which may be used in conjunction with, or in lieu of, one or more of the rules.
In one example, this disclosure describes a method of processing medical data via one or more computers. The method comprises identifying, via the one or more computers, a medical code within a medical record, and identifying, via the one or more computers, whether the medical code is one of a plurality of specified medical codes or one of a plurality of unspecified medical codes, wherein the specified medical codes are sufficient to reflect and accurately represent a medical condition to a payer and the unspecified medical codes are defined as requiring additional information for medical condition clarity. If the medical code is one of the specified medical codes, the method includes avoiding display, via the one or more computers, of clinical edit options for the medical record without generating a query for further input by a physician. Wherein the clinical edit options are used by a documentation specialist to determine whether a query for further input by a physician should be generated and allow a documentation specialist to edit one or more aspects of the medical record. If the medical code is one of the unspecified medical codes, the method includes determining, via the one or more computers, whether one of a plurality of suppression codes associated with the medical code appears in the medical record. If one of the suppression codes appears in the medical record, the method includes avoiding display, via the one or more computers of the clinical edit options for the medical record without generating the query for further input by the physician. If one of the suppression codes does not appear in the medical record, the method includes searching for one or more key terms in the medical record via the one or more computers. If the one or more key terms exist in the medical record, the method includes causing display of the clinical edit options for the medical record via the one or more computers. If one or more key terms are present in the medical record, the method includes determining whether or not to generate a query for further input by the physician via the one or more computers. The documentation specialist, based upon displayed clinical edit options, determines whether or not a clinical edit warrants a physician query as only a physician is authorized to directly edit medical documentation. In essence, clinical edit options are displayed to provide clarity and enable documentation specialists to generate queries when documentation is insufficient to reflect and accurately represent a medical condition. For some queries, the determination by the documentation may be automated through statistical machine learning techniques where the absence of suppression codes and presence of key terms from the medical record have a high probability of generating a particular query.
In another example, this disclosure describes a computerized system for processing medical data, the system comprising a computer that includes a processor and a memory, wherein the processor is configured to include an editing module. The editing module identifies a medical code within a medical record stored in the memory, and identifies whether the medical code is one of a plurality of specified medical codes or one of a plurality of unspecified medical codes, wherein the specified medical codes are sufficient to reflect and accurately represent a medical condition to a payer and the unspecified medical codes are defined as requiring additional information for medical condition clarity. If the medical code is one of the specified medical codes, the editing module avoids causing display of clinical edit options for the medical record without generating a query for further input by a physician. If the medical code is one of the unspecified medical codes, the editing module determines whether one of a plurality of suppression codes associated with the medical code appears in the medical record stored in the memory. If one of the suppression codes appears in the medical record the editing module avoids causing display of the clinical edit options for the medical record without generating the query for further input by the physician. If one of the suppression codes does not appear in the medical record, the editing module searches for one or more key terms in the medical record stored in the memory. If the one or more key terms exist in the medical record, the editing module causes display of the clinical edit options for the medical record stored in the memory. It will be apparent to one of skill in the art that clinical edit options may be displayed or be stored in memory to be accessed and viewed at a later time. If one or more key terms are present in the medical record, the editing module determines whether or not to generate a query for further input by the physician.
In another example, this disclosure describes a device for processing medical data. In this example, the device comprises means for identifying a medical code within a medical record, and means for identifying whether the medical code is one of a plurality of specified medical codes or one of a plurality of unspecified medical codes, wherein the specified medical codes are sufficient to reflect and accurately represent a medical condition to a payer and the unspecified medical codes are defined as requiring additional information for medical condition clarity. If the medical code is one of the specified medical codes, the device comprises means for avoiding display of clinical edit options for the medical record without generating a query for further input by a physician. If the medical code is one of the unspecified medical codes, the device comprises means for determining whether one of a plurality of suppression codes associated with the medical code appears in the medical record. If one of the suppression codes appears in the medical record, the device comprises means for avoiding display of the clinical edit options for the medical record without generating the query for further input by the physician. If one of the suppression codes does not appear in the medical record, the device comprises means for searching for one or more key terms in the medical record. If one or more key terms are present in the medical record, the device comprises means for causing display of the clinical edit options for the medical record. If the one or more key terms are present in or absent from the medical record, the device comprises means for generating the query for further input by the physician.
The techniques of this disclosure may be implemented at least partially in hardware, such as a processor or discrete logic circuits. The techniques may also be implemented using aspects of software or firmware in combination with the hardware. If implemented at least partially in software or firmware, the software or firmware may be executed in one or more hardware processors, such as a microprocessor, application specific integrated circuit (ASIC), field programmable gate array (FPGA), or digital signal processor (DSP). The software that executes the techniques may be initially stored in a computer-readable storage medium and loaded and executed in the processor. The processor may execute modules to perform the techniques of this disclosure, and the modules may comprise combinations of software and hardware, e.g., software routines executing on the processor.
Accordingly, this disclosure also contemplates a computer-readable storage medium comprising instructions that when executed in a processor cause the processor to process medical data, wherein upon execution the instructions cause the processor to identify a medical code within a medical record, and identify whether the medical code is one of a plurality of specified medical codes or one of a plurality of unspecified medical codes, wherein the specified medical codes are sufficient to reflect and accurately represent a medical condition to a payer and the unspecified medical codes are defined as requiring additional information for medical condition clarity. If the medical code is one of the specified medical codes, the instructions cause the processor to avoid display of clinical edit options for the medical record without generating a query for further input by a physician. If the medical code is one of the unspecified medical codes, the instructions cause the processor to determine whether one of a plurality of suppression codes associated with the medical code appears in the medical record. If one of the suppression codes appears in the medical record, the instructions cause the processor to avoid display of the clinical edit options for the medical record without generating the query for further input by the physician. If one of the suppression codes does not appear in the medical record, the instructions cause the processor to search for one or more key terms in the medical record. If one or more key terms are present in the medical record, the instructions cause the processor to cause display of the clinical edit options for the medical record. If one or more key terms are present in the medical record, the instructions cause the processor to generate the query for further input by the physician.
In other examples, this disclosure describes hybrid techniques that use fixed or pre-defined rules in conjunction with adaptive rules that can change based on statistical machine learning. For example, this disclosure describes a method of processing medical data via one or more computers. The method may comprise parsing a medical record via the one or more computers, determining a first outcome for coding the medical record based on one or more pre-defined rules, determining a second outcome for coding the medical record based on one or more adaptive rules, wherein the adaptive rules are defined based on statistical machine learning based on processing of other medical records, selecting between the first and second outcomes, and causing output related to coding the medical record based on the selected outcome.
In another example, this disclosure describes a computerized system for processing medical data, the system comprising a computer that includes a processor and a memory, wherein the processor is configured to include an editing module. The editing module parses a medical record stored in the memory, determines a first outcome for coding the medical record based on one or more pre-defined rules, determines a second outcome for coding the medical record based on one or more adaptive rules, wherein the adaptive rules are defined based on statistical machine learning based on processing of other medical records, selects between the first and second outcomes, and causes output on an output device based on the selected outcome.
In another example, this disclosure describes a device for processing medical data, the device comprising means for parsing a medical record via the one or more computers, means for determining a first outcome for coding the medical record based on one or more pre-defined rules, means for determining a second outcome for coding the medical record based on one or more adaptive rules, wherein the adaptive rules are defined based on statistical machine learning based on processing of other medical records, means for selecting between the first and second outcomes, and means for causing output for coding the medical record based on the selected outcome.
In another example, this disclosure describes a computer-readable storage medium comprising instructions that when executed in a processor cause the processor to process medical data, wherein upon execution the instructions cause the processor to parse a medical record via the one or more computers, determine a first outcome for coding the medical record based on one or more pre-defined rules, determine a second outcome for coding the medical record based on one or more adaptive rules, wherein the adaptive rules are defined based on statistical machine learning based on processing of other medical records, select between the first and second outcomes, and cause output for coding the medical record based on the selected outcome.
The details of one or more examples of this disclosure are set forth in the accompanying drawings and the description below. Other features, objects, and advantages associated with the examples will be apparent from the description and drawings, and from the claims.
This disclosure describes systems and techniques for processing medical data via one or more computers. The systems and techniques may be used by a medical reviewer (sometimes referred to as a “documentation specialist”). The documentation specialist may be assigned the task of reviewing medical records for purposes of billing a payer and ensuring accuracy in the medical records. The payer may be a governmental agency, such a Medicare or Medicaid, or a private entity, such as an insurance company. The documentation specialist may be a nurse, clinician, administrative person, or any person given the task of reviewing medical records, and verifying or updating medical codes in the medical documentation. In general, the documentation specialist typically reviews medical records and ensures that the medical records include the correct medical codes associated with the medical tasks or medical conditions defined in the medical records.
Unfortunately, medical records can be long, complicated, and sometimes incomplete. This can make the coding and review process very time consuming and difficult for a documentation specialist. The techniques and systems described herein can help to automate (or partially automate) the coding process associated with medical record review. In this manner, the process of coding medical records and verifying codes in medical records can be improved and/or simplified. The techniques may apply one or more rules to define when documentation for medical records is sufficient and when further review of the medical records is needed. The rules may define when the documentation specialist should review the medical records and may automate the process by avoiding the display of medical records to the documentation specialist when the documentation in the medical records is sufficient. The rules may further define when physician review is needed, and may automate the process by avoiding physician review when the documentation in the medical records is sufficient or when review by the documentation specialist may suffice prior to (and possibly in lieu of) physician review. Physician review of medical records is generally undesirable when it can be avoided as it is time consuming, labor intensive, and adds cost. Additional techniques are also described, which may rely on machine learning to replace one or more of the rules described herein. With machine learning, the rules or techniques may adapt over time based on statistics associated with the selections or activities of documentation specialists with respect to coding of prior medical records.
As described in greater detail below, the methods of this disclosure may be performed by one or more computers. The methods may be performed by a stand-alone computer, or may be executed in a client-server environment in which a documentation specialist views medical records at a client computer. In the later case, the client computer may communicate with a server computer. The server computer may store the medical records and apply the techniques of this disclosure to facilitate medical record review and coding addition or modification by the documentation specialist at the client computer.
In one example, a method may include identifying, via one or more computers, a medical code within a medical record, and identifying, via the one or more computers, whether the medical code is one of a plurality of specified medical codes or one of a plurality of unspecified medical codes. The specified medical codes may be defined as sufficient to reflect and accurately represent a medical condition to a payer and the unspecified medical codes are defined as requiring additional information for medical condition clarity. If the medical code is one of the specified medical codes, the one or more computers may avoid display of clinical edit options for the medical record without generating a query for further input by a physician. If the medical code is one of the unspecified medical codes, the one or more computers may determine whether one of a plurality of suppression codes associated with the medical code appears in the medical record. If one of the suppression codes appears in the medical record, the one or more computers may avoid display of the clinical edit options for the medical record without generating the query for further input by the physician. If one of the suppression codes does not appear in the medical record, the method may include searching for one or more key terms in the medical record via the one or more computers. If one or more key terms are present in the medical record, the method may include causing display of the clinical edit options for the medical record via the one or more computers. If one or more key terms are present in the medical record, the method may include generating the query for further input by the physician via the one or more computers.
The output device 130 may comprise a display screen, although this disclosure is not necessarily limited in this respect, and other types of output devices may also be used. Memory 114 stores raw medical data 118 comprising medical records. Processor 112 is configured to include an editing module 102 that executes techniques of this disclosure with respect to raw medical data 118, and in some cases, editing module 102 may generate coded medical data 120 comprising edited medical records.
Processor 112 may comprise a general-purpose microprocessor, a specially designed processor, an application specific integrated circuit, a field programmable gate array, a collection of discrete logic, or any type of processing device capable of executing the techniques described herein. In one example, memory 114 may store program instructions (e.g., software instructions) that are executed by processor 112 to carry out the techniques described herein. In other examples, the techniques may be executed by specifically programmed circuitry of processor 112. In these or other ways, processor 112 may be configured to execute the techniques described herein.
Output device 130 may comprise a display screen, and may also include other types of output capabilities. In some cases, output device 130 may generally represent both a display screen and a printer in some cases. Editing module 102 may be configured to cause output device 130 to output specialist prompts 136 and physician prompts 138. Specialist prompts 136 may be generated, e.g., as output on a display screen, so as to allow the documentation specialist to add or modify coding edits to the medical records. In this manner, coded medical data 120 may be generated and stored based on raw medical data 118 and based on additional information from a documentation specialist operating in response to specialist prompts 136 generated at output device 130. In addition, physician prompts 136 may be generated, e.g., as output on a display screen or printouts of one or more request forms for a physician. This can allow a physician to provide additional information so as to improve and/or supplement the medical records, when necessary. The techniques of this disclosure may serve to automate the coding and the review process with respect to medical records, minimizing both specialist prompts 136 and physician prompts 138. For example, specialist prompts 136 can be minimized to situations in which the medical records do not include the necessary information, but do include key words from which a documentation specialist may be able to add or modify medical codes based on the information in the record. Physician prompts 138 may be minimized to situations in which the medical records do not include the necessary information, and also lack sufficient key words from which a documentation specialist would be able to add or modify medical codes based on the information in the record.
In one example, editing module 102 identifies a medical code within a medical record stored in the memory 114. The medical record may be one of many medical records within raw medical data 118 needing review by a documentation specialist. Editing module 102 identifies whether the medical code is one of a plurality of specified medical codes or one of a plurality of unspecified medical codes. The specified medical codes are defined as sufficient to reflect and accurately represent a medical condition to a payer and the unspecified medical codes are defined as requiring additional information for medical condition clarity. The payer typically comprises either a governmental payer, or an insurance company, although the techniques of this disclosure may apply to other payers.
If the medical code is one of the specified medical codes, editing module 102 avoids causing display of clinical edit options via specialist prompts 136 on output device 130 for the medical record. In this case, editing module 102 also avoids the generation of a query for further input by a physician, e.g., avoids generating physician prompts 138 on output device 130.
If the medical code is one of the unspecified medical codes, editing module 102 determines whether one of a plurality of suppression codes associated with the medical code appears in the medical record stored in raw medical data 118 in memory 114. If one of the suppression codes appears in the medical record editing module 102 avoids causing display of clinical edit options via specialist prompts 136 on output device 130 for the medical record. In this case, editing module 102 also avoids the generation of a query for further input by a physician, e.g., avoids generating physician prompts 138 on output device 130.
At this point, if one of the suppression codes does not appear in the medical record, editing module 102 searches for one or more key terms in the medical record. If one or more key terms exist in the medical record, editing module 102 generates specialist prompts 136 on output device, e.g., causing display of the clinical edit options for the medical record stored in raw medical data 118 of memory 114. When code additions or modifications are received from a documentation specialist, editing module 102 stores an edited version of the medical record in coded medical data 120 within memory 114. On the other hand, if one or more key terms are present in the medical record and do not contain sufficient detail as to define a code, editing module 102 generates physician prompts 138 on output device, e.g., causing display or printout of a query for further input by the physician.
The medical codes within the medical records may comprise codes defined by the International Classification of Diseases (ICD), such as ICD-9 codes or ICD-10 codes, although the techniques are not necessarily limited to ICD medical codes and could apply with respect to other types of medical codes as would be apparent to one of skill in the art. In particular, other medical codes may be used with the techniques of this disclosure, particularly for billing to insurance companies or other non-governmental organizations, which may define their own code system or may adopt that of the ICD. Like the medical codes, the suppression code may also be defined by the ICD, wherein the suppression codes are more specific than the medical codes. According, a given suppression code may override and “suppress” a broader medical code by providing more specific information on a given condition or procedure coded in the medical record.
In some examples, the key terms are pre-defined, and editing module 102 automatically searches for the key terms within the medical record when one of the suppression codes does not appear in the medical record. In other examples, at least some of the associations between key terms and queries may be adaptively defined, in which case machine learning techniques may be used over time to associate key terms with queries to medical records that are made by the documentation specialist. Accordingly, in this case editing module 102 may adaptively define at least some of the key terms based on previous searches for terms performed by one or more users (e.g., other documentation specialists that performed review and edits or similar types of medical records). For example, editing module 102 may cause the display of possible terms to the one or more users (e.g., as specialist prompts 136), and editing module 102 may then search for ones of the possible terms within a medical record based on selections by the one or more users (e.g., user input in response to specialist prompts 136). In this case, one or more of the associations between key terms and queries may be adaptively defined by editing module 102 over time based on the selections of the possible terms by the one or more users. Moreover, once one or more of the associations between key terms and queries are adaptively defined over time based on the selections of the possible terms by the one or more users, editing module 102 may be configured to automatically search for the adaptively defined associations between key terms and queries when one of the suppression codes does not appear in the medical record. In this manner, machine learning techniques may be used over time to associate key terms with selections and/or queries to medical records made by documentation specialists. Additional machine learning techniques are also discussed below.
When causing display of the clinical edit options for the medical record, editing module 102 may cause any of a wide variety of specialist prompts 136 to appear on output device 130. In some examples, specialist prompts 136 may display of at least a portion of data from the medical record in raw medical data 118 to allow for review by a documentation specialist. Once code edits e.g., additions and/or modifications are reviewed and confirmed by the documentation specialist, editing module 102 may cause the edited version of the medical record to be stored in memory 114 as coded medical data 120.
When generating a query for further input by the physician, editing module 102 may automatically or manually, through a documentation specialist, generate physician prompts 138. Physician prompts 138 may comprise a physician documentation request that requests additional details for the medical record. As examples, the requested details may pertain to the medical code, the suppression code, or one or more key terms. In this way, physician prompts 138 can be automated, yet limited to situations in which physician input is actually needed. Accordingly, unwanted or unnecessary queries to the physician can be substantially minimized.
The system of
Server computer 210 may perform the techniques of this disclosure, but the documentation specialist (i.e., the user) may interact with the system via client computer 250. Server computer 210 may include a processor 212, a memory 214, and a communication interface 226. Client computer 250 may include a communication interface 252, a processor 242 and an output device 230. Of course, client computer 250 and server computer 210 may include many other components. The illustrated components are shown merely to explain various aspects of this disclosure.
Output device 230 may comprise a display screen, although this disclosure is not necessarily limited in this respect and other output devices may also be used. Memory 214 stores raw medical data 218 comprising medical records. Processor 212 of server computer 210 is configured to include an editing module 202 that executes techniques of this disclosure with respect to raw medical data 218, and in some cases, editing module 202 may generate coded medical data 220 comprising edited medical records.
Processors 212 and 242 may each comprise a general-purpose microprocessor, a specially designed processor, an application specific integrated circuit, a field programmable gate array, a collection of discrete logic, or any type of processing device capable of executing the techniques described herein. In one example, memory 214 may store program instructions (e.g., software instructions) that are executed by processor 212 to carry out the techniques described herein. In other examples, the techniques may be executed by specifically programmed circuitry of processor 212. In these or other ways, processor 212 may be configured to execute the techniques described herein.
Output device 230 on client computer 250 may comprise a display screen, and may also include other types of output capabilities. For example, output device 230 may generally represent both a display screen and a printer in some cases. Editing module 202 may be configured to cause output device 230 of client computer 250 to output specialist prompts 236 and physician prompts 238. Specialist prompts 236 may be generated, e.g., as output on a display screen, so as to allow the documentation specialist to add or modify codes based on the medical records or provide additional information from the medical record to help clarify the query. In this manner, coded medical data 220 may be generated and stored based on raw medical data 218 and based on additional information, e.g., code edits, additions or modifications, from a documentation specialist operating in response to specialist prompts 236 generated at output device 230 of client computer 250. In addition, physician prompts 236 may be generated, e.g., as output on a display screen or printouts of one or more request forms for a physician. This can allow a physician to provide additional information so as to improve and/or supplement the medical records, when necessary. Again, the techniques of this disclosure may serve to automate the coding and the review process with respect to medical records, minimizing both specialist prompts 236 and physician prompts 238. For example, specialist prompts 236 can be minimized to situations in which the medical records do not include the necessary information, but do include key words from which a documentation specialist may be able to review based on the information in the record. Physician prompts 238 may be minimized to situations in which the medical records do not include the necessary information, and also lack sufficient key words from which a documentation specialist would be able to review based on the information in the record.
Similar to the stand alone example of
If the medical code is one of the specified medical codes, editing module 202 avoids causing display of clinical edit options via specialist prompts 236 on output device 230 of client computer 250 for the medical record. In this case, editing module 202 also avoids the generation of a query for further input by a physician, e.g., avoids generating physician prompts 238 on output device 230 of client computer 250. Again, communication interfaces 226 and 252 allow for communication between server computer 210 and client computer 250 via network 240. In this way, editing module may execute on server computer 210 but the output may appear on output device 230 of client computer. A documentation specialist operating on client computer 250 may log-on or otherwise access editing module of server computer 210, such as via a web-interface operating on the Internet or a propriety network, or via a direct or dial-up connection between client computer 250 and server computer 210. In some cases, data displayed on output device 230 (including any specialist prompts 238 or physician prompts 238) may be arranged in web pages served from server computer 210 to client computer 250 via hypertext transfer protocol (HTTP), extended markup language (XML), or the like.
If the medical code is one of the unspecified medical codes, editing module 202 determines whether one of a plurality of suppression codes associated with the medical code appears in the medical record stored in raw medical data 218 in memory 214. If one of the suppression codes appears in the medical record editing module 202 avoids causing display of clinical edit options via specialist prompts 236 on output device 230 for the medical record. In this case, editing module 202 also avoids the generation of a query for further input by a physician, e.g., avoids generating physician prompts 238 on output device 230 of client computer 250.
At this point, if one of the suppression codes does not appear in the medical record, editing module 202 searches for one or more key terms in the medical record. If one or more key terms exist in the medical record, editing module 202 generates specialist prompts 236 on output device 230 of client computer 250, e.g., causing display of the clinical edit options for the medical record stored in raw medical data 218 of memory 214. When code additions or modifications are received from a documentation specialist operating on client computer 250, editing module 202 stores an edited version of the medical record in coded medical data 220 within memory 214. On the other hand, if one or more key terms are present in the medical record and do not contain sufficient detail as to define a code, editing module generates physician prompts 238 on output device, e.g., causing display or printout of a query for further input by the physician. Both the presence of key terms and the absence of key terms in the medical record may be used, in some cases, to determine the outcome (e.g., display of edit options to the documentation specialist or display or output of a query for further input by the physician.
A documentation specialist may select a patient record in order to review the documentation that physician created in response to a patient visit. The documentation specialist may also be referred to as a coder, a reviewer, or a user of the system described herein. After the documentation specialist reads the patient record, they may add the diagnosis code, such as 428.0 to code unspecified congestive heart failure (CHF). As shown in
Manual review of CDI reference material can be difficult and time-consuming for a documentation specialist. Review of CDI reference material like that shown in
For example,
For example, the screen shot of
In the cross-reference view depicted in
The documentation specialist may also be able to review the entire medical record by clicking on the document title, e.g., a GUI link labeled as “progress notes 1/03/2011” on the left hand side of
The example of
If a search code is found (“yes” 1002), editing module 102 may check the ICD-9 codes for suppression codes (1004). If a suppression code is found (“yes” 1005), then editing module 102 does not cause the display of any clinical edit options to the documentation specialist (1003), e.g., since sufficient information should be included for the medical record in the form of a suppression code. However, if a suppression code is not found (“no” 1005), then editing module 102 may search for key terms in the medical record (1006). At this point, if at least one required term is found in the medical record (“yes” 1007), then editing module 102 does not cause the display of any clinical edits to the documentation specialist (1003), e.g., since sufficient information should be included for the medical record in the form of a required term (possibly in conjunction with other elements in the medical record). However, if at this point any required terms are present in the medical record (“no” 1007) and do not contain sufficient detail as to define a code, then editing module 102 may cause output device 130 to display the clinical edits and search terms (1008), e.g., which may correspond to the display of specialist prompts 136. It will be apparent to one of skill in the art that other medical classification systems may be used in place of ICD-9 codes.
According to the techniques described herein, instead of having a time intensive manual process for entering codes, reviewing the CDI reference information and searching the documentation for CDI opportunities the documentation specialist may reap benefits of an automated process. With the CDI enhancement and auto-suggestions of codes, the documentation specialists may begin coding sessions with codes and CDI clinical edits already present for their review and consideration, which can result in significant productivity improvement for the documentation specialists.
In some example, the steps taken to generate either suppression or non-suppression codes that ultimately lead to a query suggestion can be reduced through either auto-suggesting the codes, or bypassing one or more choices in a decision-tree logic based encoder implemented e.g., as part of the described editing module. By annotating clinical terminology in the patient documentation and integrating decision-tree choices (referred to herein as coding paths) directly with the annotations, productivity improvements can be achieved. Essentially, the coding path choices can be prompted on an exception basis. Bypassed steps may include the manual entry of clinical terms and the manual selection of coding path prompts that can be satisfied based on the clinical terminology annotated in the medical record.
In a manual process, the user may perform the following steps:
In one automated example, annotations may be embedded in coding paths. With the coding paths embedded within a document's annotated clinical terminology, the manual steps above may be changed as follows:
In another automated example, auto-suggested codes can be generated. In this case, with auto-suggested codes, steps 1 and 2 above can be eliminated from the perspective of the documentation specialist. The system may identify a code along with evidence for the clinical specialist to validate the code. The steps above may be revised as follows:
In additional examples, one or more of the techniques and rules described herein may be replaced or supplemented with machine learning techniques based on statistics. In this case, the actions of documentation specialists can be saved and the computer may learn and adapt future coding suggestions based on statistics associated with the prior actions of documentation specialists.
In both a rules-based approach or a statistical machine learning approach, it may be desirable to determine one of three outcomes: (1) automatically coding the document (and either sending the document directly to billing or to a human review for final approval), (2) issuing one or more specificity queries back to the physician to improve the documentation, or (3) determining that the automated system was not confident in its prediction and sending the documentation to a human reviewer to choose outcome (1) or (2). In some systems, if the documentation is complete enough that a human coder (i.e., a documentation specialist) would not issue a specificity query to a physician, outcome (1) may be chosen, and only when the documentation is missing key information may outcome (2) be chosen, since it is often considered costly and undesirable to query the physician. Therefore, systems that reduce the number of situations with outcome (3), as well as systems reducing the number of documents mistakenly given outcome (2) when they should have been given outcome (1) or visa-versa, may improve conventional coding systems.
In one exemplary machine learning approach, a computer may gather data on the actions of the documentation specialist, and observe or analyze the patient documentation and the outcome (e.g., outcomes (1) or (2) mentioned above). The computer may also observe or analyze any codes or queries generated in each case. The computer may also process documents to identify linguistic and clinical evidence, including one or more of clinical terminology, non-clinical terminology, negation, ambiguity, semantic relationships of identified terms, sentence structure, word order, temporal references, document sections and document structure. The computer may then train one or more machine learning models based on the gathered data to determine statistical relationships between the linguistic and clinical evidence and any resulting action (e.g., outcome (1) or (2), codes generated, or specificity queries generated). By performing such tasks, the choice between outcomes (1), (2) or (3) can be made. In some cases, the computer may implement a support vector machine in order to implement one or more of these machine learning techniques.
For new patient documentation, a statistical machine learning approach to coding may apply statistical models described above to predict the outcome, one or more codes and any queries that may be needed. These might then be reviewed by the documentation specialist, in which case the statistical machine learning approach may provide a more desirable starting point for the documentation specialist, with suggestions for the outcome, suggestions for one or more codes, and suggested queries that may be needed
At this point, any actions taken by the documentation specialist with respect to suggestions offered by the statistical machine learning approach may be used to generate an updated or new statistical model (e.g., a “confidence” model) based on the relationship between the available clinical and linguistic evidence, the outcome, codes, and/or queries predicted by the original model, and the selections or agreement of the documentation specialist with the predictions of the automated system. Then, the computer may apply the confidence model to determine whether future patient documentation should be given outcome (3) (i.e., sent to a human reviewer when confidence is low) instead of the chosen outcome that might otherwise be proposed.
For both a rules-based approach and a statistical machine learning approach, in cases where outcome (3) occurs (i.e., cases where a human reviewer is needed to complete the coding process), the available evidence may be used to increase the speed with which a human reviewer can complete documentation by identifying and jumping to an intermediate step in the coding process from which a documentation specialist can begin the coding process. In a rules-based approach, identified clinical terms and context may be used to link directly to intermediate code path steps, such as jumping to outcome (2) or outcome (3) based on satisfying one or more rules with respect to the content of a medical record. Similarly, in a statistical/machine learning approach, the system may gather data to train a model that maps linguistic and clinical evidence to a code path, which may then be followed for new patient documents in order to predict the most likely intermediate code path based on available linguistic and clinical evidence.
In many cases, a rules-based approach and a statistical machine learning-based approach may be used together to define a “hybrid” approach. A hybrid rules-based and statistical system may be used on the same documentation, then selection of the outcome, codes, or queries may be based on the output of both approaches. The choice may be made based on rules and or statistics (e.g. for specific rules or queries, the system may choose the rules-based outputs and otherwise, the system may choose the machine learning output). Statistical confidence may also be used, in which case, the system may build a confidence model based on the rules-based system, then compare the confidence of the rules-based outputs to the confidence of machine learning outputs. In this case, the system may choose results having the highest confidence metric, with a first confidence metric being defined for the outcome of a rules-based approach and a second confidence metric being defined for the machine learning approach. In still other cases, a confidence metric may be defined only for the outcome defined by the machine learning approach, and the outcome defined by the rules-based approach may be used as the default approach whenever the confidence metric is below some threshold. In these cases, the outcome defined by the machine learning approach may be used when the confidence metric defined by the machine learning approach exceeds a threshold.
Furthermore, in some hybrid examples, machine learning may be applied to only some of the steps of the coding process, such as that associated with the identification and coding based on key words in the medical record. In this case, the computer may apply a rules-based approach up to the point where key word searching occurs. At the point of key work searching, the computer may adaptively define at least some of the key terms based on previous searches performed by one or more users (i.e., key word searches performed by previous documentation specialists on previously coded documents). The computer may cause the display of possible terms to the one or more users, and may search for possible terms based on selections by the users (i.e., the documentation specialists). Based on such selections, the key terms may be adaptively defined over time. Thereafter, the computer may automatically search for the adaptively defined key terms, e.g., when one of the suppression codes does not appear in the medical record. That is, if the documentation specialist associates key terms or phrases with particular codes or particular actions in the coding process (such as queries to the physician), the computerized system may learn over time, and automate these associations for automated output, or automated suggestions to the documentation specialist with respect to future documents being coded.
If the medical code is one of the specified medical codes (“specified” 1202), editing module 102 avoids clinical edit options or physician prompts (1203). In other words, if the medical code is specified (“specified” 1202), editing module 102 avoids causing display of clinical edit options via specialist prompts 136 on output device 130 for the medical record, and editing module 102 also avoids the generation of a query for further input by a physician, e.g., avoids generating physician prompts 138 on output device 130.
If the medical code is one of the unspecified medical codes (“unspecified” 1202), editing module 102 determines whether one of a plurality of suppression codes associated with the medical code appears in the medical record stored in raw medical data 118 in memory 114 (1204). If one of the suppression codes appears in the medical record (“yes” 1204), editing module 102 avoids clinical edits or physician prompts (1203). In other words, if a suppression code appears in the medical record (“yes” 1204), editing module 102 avoids causing display of clinical edit options via specialist prompts 136 on output device 130 for the medical record, and editing module 102 also avoids the generation of a query for further input by a physician, e.g., avoids generating physician prompts 138 on output device 130 (1203).
At this point, if one of the suppression codes does not appear in the medical record (“no” 1204), editing module 102 searches for one or more key terms in the medical record (1205). If one or more key terms exist in the medical record (“yes” 1205) and are sufficient to define a code, editing module 102 displays the medical record for clinical edit options by the documentation specialist (1206). In particular, editing module 102 may generate specialist prompts 136 on output device, e.g., causing display of the editing options for the medical record stored in raw medical data 118 of memory 114. Accordingly, when code additions or modifications are received from a documentation specialist, editing module 102 may store an edited version of the medical record in coded medical data 120 within memory 114. On the other hand, if one or more key terms are present in the medical record (“no” 1205) and do not contain sufficient detail as to define a code, editing module 102 generates a physician query (1207). In particular, if the one or more key terms do not exist in the medical record (“no” 1205), editing module 102 generates physician prompts 138 on output device, e.g., causing display or printout of a query for further input by the physician.
As noted above, medical codes within the medical records may comprise codes defined by the ICD, such as ICD-9 codes or ICD-10 codes, although the techniques are not necessarily limited to ICD medical codes and could apply with respect to other types of medical codes. In particular, other medical codes may be used with the techniques of this disclosure, particularly for billing to insurance companies or other non-governmental organizations, which may define their own code system or may adopt that of the ICD. Like the medical codes, the suppression code may also be defined by the ICD, wherein the suppression codes are more specific than the medical codes. According, a given suppression code may override and “suppress” a broader medical code by providing more specific information on a given condition or procedure coded in the medical record.
In some examples, the key terms are pre-defined, and editing module 102 automatically searches for the key terms within the medical record when one of the suppression codes does not appear in the medical record. In other examples, at least some of the associations between key terms and queries may be adaptively defined, in which case machine learning techniques may be used over time to associate key terms with queries to medical records that are made by the documentation specialist. Accordingly, in this case editing module 102 may adaptively define at least some of the key terms based on previous searches for terms performed by one or more users (e.g., other documentation specialists that performed review and edits or similar types of medical records). For example, editing module 102 may cause the display of possible terms to the one or more users (e.g., as specialist prompts 136), and editing module 102 may then search for ones of the possible terms within a medical record based on selections by the one or more users (e.g., user input in response to specialist prompts 136). In this case, one or more of the associations between key terms and queries may be adaptively defined by editing module 102 over time based on the selections of the possible terms by the one or more users. Moreover, once one or more associations of the key terms and queries are adaptively defined over time based on the selections of the possible terms by the one or more users, editing module 102 may be configured to automatically search for the adaptively defined associations between key terms and queries when one of the suppression codes does not appear in the medical record. In this manner, machine learning techniques may be used over time to associate key terms with selections and/or edits to medical records made by documentation specialists. Additional machine learning techniques are also discussed below.
When causing display of the editing options for the medical record, editing module 102 may cause any of a wide variety of specialist prompts 136 to appear on output device 130. In some examples, specialist prompts 136 may display of at least a portion of data from the medical record in raw medical data 118 to allow for edits by a documentation specialist. Once code additions or modifications are made by the documentation specialist, editing module may cause the edited version of the medical record to be stored in memory 114 as coded medical data 120.
When generating a query for further input by the physician, editing module 102 may automatically or manually, through a documentation specialist, generate physician prompts 138. Physician prompts 138 may comprise a physician documentation request that requests additional details for the medical record. As examples, the requested details may pertain to the medical code, the suppression code, or one or more key terms. In this way, physician prompts 138 can be automated, yet limited to situations in which physician input is actually needed. Accordingly, unwanted or unnecessary queries to the physician can be substantially minimized.
As mentioned above, one or more of the techniques and rules described herein may be replaced or supplemented with machine learning techniques based on statistics. In this case, the actions of documentation specialists can be saved or accumulated over time (e.g., as statistics), and the computer may learn and adapt future coding suggestions based on statistics associated with the prior actions of documentation specialists.
As discussed above, in both a rules-based approach or a statistical machine learning approach, it may be desirable to determine one of three outcomes: (1) automatically coding the document (and either sending the document directly to billing or to a human review for final approval), (2) issuing one or more specificity queries back to the physician to improve the documentation, or (3) determining that the automated system was not confident in its prediction and sending the documentation to a human reviewer to choose (1) or (2). It may be desirable, in some cases, to determine a first outcome based on a set of pre-defined rules and determine a second outcome based on adaptive rules defined by statistical machine learning. Then, the computer may select between the first and second outcomes. Confidence metrics may be defined for one or both outcomes and the selection between the first and second outcomes may be based on the one or more confidence metrics.
The techniques of this disclosure may be implemented in a wide variety of computer devices, such as servers, laptop computers, desktop computers, notebook computers, tablet computers, hand-held computers, smart phones, and the like. Any components, modules or units have been described provided to emphasize functional aspects and does not necessarily require realization by different hardware units. The techniques described herein may also be implemented in hardware, software, firmware, or any combination thereof. Any features described as modules, units or components may be implemented together in an integrated logic device or separately as discrete but interoperable logic devices. In some cases, various features may be implemented as an integrated circuit device, such as an integrated circuit chip or chipset.
If implemented in software, the techniques may be realized at least in part by a computer-readable medium comprising instructions that, when executed in a processor, performs one or more of the methods described above. The computer-readable medium may comprise a tangible computer-readable storage medium and may form part of a computer program product, which may include packaging materials. The computer-readable storage medium may comprise random access memory (RAM) such as synchronous dynamic random access memory (SDRAM), read-only memory (ROM), non-volatile random access memory (NVRAM), electrically erasable programmable read-only memory (EEPROM), FLASH memory, magnetic or optical data storage media, and the like. The computer-readable storage medium may also comprise a non-volatile storage device, such as a hard-disk, magnetic tape, a compact disk (CD), digital versatile disk (DVD), Blu-ray disk, holographic data storage media, or other non-volatile storage device.
The term “processor,” as used herein may refer to any of the foregoing structure or any other structure suitable for implementation of the techniques described herein. In addition, in some aspects, the functionality described herein may be provided within dedicated software modules or hardware modules configured for performing the techniques of this disclosure. Even if implemented in software, the techniques may use hardware such as a processor to execute the software, and a memory to store the software. In any such cases, the computers described herein may define a specific machine that is capable of executing the specific functions described herein. Also, the techniques could be fully implemented in one or more circuits or logic elements, which could also be considered a processor.
These and other examples are within the scope of the following claims.
This Application claims the benefit of U.S. Provisional Application No. 61/539,410, filed Sep. 26, 2011, the entire content of which is incorporated herein by reference.
| Number | Date | Country | |
|---|---|---|---|
| 61539410 | Sep 2011 | US |
