Patent Application
20240274252
This patent application claims priority to, and incorporates by reference the entire disclosure of, UK Patent Application No. 2301936.7, filed on Feb. 10, 2023.
The present disclosure relates to a method of generating a dosage calculator for determining a dosage of a drug for administering to a patient, a dosage calculator and a method of determining dosages.
Pharmacokinetic/Pharmacodynamic models such as compartment models are a useful way of understanding the effects of a drug on the human body. Compartment models may be used to model concentrations of a drug in the central compartment—a region of the body in which the drug takes effect. Compartment models can therefore provide a relationship between drug dosage and drug effect.
Compartment models typically employ differential equations that require iterative time-based computation. The computation is both resource intensive and time intensive.
The present disclosure relates to methods and apparatus for determining a drug dosage for a patient in an efficient manner that can be deployed in personal computing devices such as those found in a smart phone.
According to a first aspect of the present disclosure there is provided a method of generating a dosage calculator for determining a dosage of a drug for administering to a patient, the method comprising:
The machine learning dosage calculator may be configured to calculate the dosage for administering to the patient by processing patient data and a target PKPD metric for the patient.
The population data may comprise drug data for a plurality of patients. The drug data for each patient may include patient data, dosage data and a PKPD metric.
The patient data may comprise one or more patient data parameters that can influence the effect of the drug on a patient. The patient data may include factors that influence the absorption of the drug into the body and/or the clearance of the drug from the body. The patient data may include one or more input parameters used in a PKPD model for the drug.
The dosage data may include a dosage of the drug administered to a patient.
The PKPD metric may comprise one or more parameters indicative of the effect of the drug on the patient's body.
The PKPD metric may comprise one or more of:
The population data may comprise simulated population data comprising simulated patient data, simulated dosages and/or simulated PKPD metrics.
The simulated population data may comprise simulated population data for at least 10,000 simulated patients.
The method may comprise calculating the simulated population data using the PKPD model for the drug.
Calculating simulated population data using a PKPD model may comprise:
The simulated patient data may comprise one or more of: a kidney function metric; a liver function metric; a patient age; a patient ethnicity; a patient sex; a patient weight; a patient body mass index; a patient haemoglobin level; a patient left ventricular function; a pharmacogenomic profile; and a patient medication list.
Defining the simulated patient data may comprise:
Defining the simulated patient data may comprise:
Generating each simulated patient may comprise generating simulated patient data representative of all combinations of one discrete value from each parameter type.
The PKPD model may comprise a plasma level prediction model and the PKPD metric may comprise a plasma level metric.
The PKPD model may comprise a time-based differential equation model for modelling a time dependence of a concentration of the drug in an effective area of the body as a function of patient data.
The PKPD model may comprise a two-compartment model for modelling a time dependence of a central compartment drug concentration as a function of patient data.
Training the machine learning dosage calculator may comprise training the machine learning dosage calculator using the simulated population data and the real population data.
The real population data may include real patient data comprising one or more of: a kidney function metric; a liver function metric; a patient age; a patient ethnicity; a patient sex; a patient weight; a patient body mass index; a patient haemoglobin level; a patient left ventricular function; a pharmacogenomic profile; and a patient medication list. The real population data may include real dosage data and a measured PKPD metric for each patient.
Training the machine learning dosage calculator using the population data may comprise:
The method may further comprise validating the machine learning model using further population data.
The further population data may be different to the population data.
The further population data may comprise further real population data and/or further simulated population data.
The method may further comprise locking the machine learning dosage calculator to prevent further adjustment to the machine learning dosage calculator.
The method may further comprise calibrating the machine learning algorithm for a patient based on a measured PKPD metric obtained from a physiological test on the patient.
The physiological test may comprise a blood test, a urine test, a cerebrospinal fluid analysis, or a biopsy. The measured PKPD parameter may comprise a drug concentration from the physiological test.
The machine learning dosage calculator may be configured to directly calculate the dosage for administering to the patient by processing patient data and a target PKPD metric.
The machine learning dosage calculator may directly calculate the dosage by:
The patient data may comprise one or more of: a kidney function metric; a liver function metric a patient age; a patient ethnicity; a patient sex; a patient weight; a patient body mass index; a patient haemoglobin level; a patient left ventricular function; a pharmacogenomic profile; and a patient medication list.
The drug may comprise an anticoagulant, preferably a direct oral anticoagulant, preferably dabigatran.
According to a second aspect of the present disclosure there is provided a computer implemented method for determining a dosage of a drug for administering to a patient, the method comprising:
The method may further comprise:
The updated patient data may comprise a measured PKPD metric obtained from a physiological test on the patient.
The PKPD metric may include a drug concentration.
The method may further comprise calibrating the dosage calculator by adjusting the dosage calculator and/or the PKPD model using the drug concentration.
The PKPD model may comprise a time-based differential equation model for modelling a time dependence of a concentration of the drug in an effective area of the body as a function of the patient data. The PKPD model may comprise a two compartment model for modelling a time dependence of a central compartment drug concentration as a function of the patient data.
Processing the patient data with a dosage calculator to determine the dosage of dabigatran for administering to the patient may comprise:
The method may comprise determining the target PKPD metric as a personalised target PKPD metric based on the patient data.
The method may comprise:
The PKPD metric may comprise a drug concentration in an effective location of the patient's body.
The patient data may comprise one or more dosage times at which the patient received a dose of the drug. The PKPD metric may comprise a time-dependent drug concentration in an effective location of the patient's body based on the one or more dosage times.
The PKPD metric may comprise one or more of:
According to a third aspect of the present disclosure there is provided a computer implemented method for determining a dosage of dabigatran for administering to a patient, the method comprising:
The patient data may further comprise one or more of: a patient age; a patient ethnicity; a patient gender; a patient weight; a patient haemoglobin level; a patient left ventricular function; and a patient medication list.
The patient medication list may comprise an indication of whether the patient is consuming one or medications comprising: a proton-pump inhibitor; a calcium channel blocker; a nonsteroidal anti-inflammatory drug; a H2 receptor antagonists, verapamil, amiodarone, clopidogrel, aspirin, and diltiazem.
The patient data may further comprise one or more of: reported side-effects; alcohol intake; smoking history, a patient clotting metric; a treatment purpose; patient genetic determinants; patient co-conditions; a patient activity level; a patient dosage compliance; a patient liver function; a patient thrombosis history; a patient haemorrhage history; a patient cancer history; a family thrombosis history; familial stroke history, familial bleeding history; a patient cardiovascular history; a patient metabolic history; a patient blood pressure history; a patient platelet count; a patient heart rate; and a patient haematocrit.
The treatment purpose may comprise: prevention of thrombosis, embolism and/or stroke, optionally including patients with non-valvular atrial fibrillation and one or more risk factors including: a previous stroke or transient ischaemic attach, heart failure, diabetes or hypertension; active thrombosis treatment and/or active pulmonary embolism treatment; prevention of venous thromboembolism in people who have undergone surgery, optionally including hip or knee replacement therapy.
The method may further comprise:
The updated patient data may include a patient clotting metric and/or a drug concentration, from a blood test result.
The method may further comprise calibrating the dosage calculator by adjusting the dosage calculator and/or the plasma level prediction model using the patient clotting metric and/or drug concentration.
The plasma level prediction model may comprise a time-based differential equation model for modelling a time dependence of a plasma concentration of dabigatran as a function of the patient data.
Processing the patient data with a dosage calculator to determine the dosage of dabigatran for administering to the patient may comprise:
Processing the patient data with a dosage calculator to determine the dosage of dabigatran for administering to the patient may comprise:
Refining the dose estimate may comprise:
The target plasma level metric may comprise one or more of:
The patient data may comprise one or more target dependent patient parameters comprising: reported side effects; a patient thrombosis history; a patient haemorrhage history; a patient cancer history; a patient stroke history; a patient liver function metric; a patient heart function metric; a patient brain state; a patient smoking history; a patient alcohol history; a patient blood pressure; a patient activity level; a patient dosage compliance; a patient mobility state; a patient menstruation state; a patient inflammation state; a patient infection state; a patient co-medication; a patient co-condition; a blood clotting metric; a patient genetic profile; familial stroke history, familial bleeding history; familial hypertension history; a patient cardiovascular history; a patient metabolic history; a patient blood pressure history; a patient blood pressure; a patient heart rate, a patient platelet count; a patient haematocrit; and a patient hydration state, wherein the method comprises:
The method may comprise:
The method may comprise:
The personalised target plasma level metric may comprise:
The ideal therapeutic level may comprise:
The method may comprise:
The plasma level metric may comprise a plasma level time profile.
Indicating one or more of: the plasma level metric; the patient thrombosis risk; or the patient haemorrhage risk, may comprise indicating to the patient or a health care professional. Indicating may be via a user interface.
The patient data may comprises one or more dosage times at which the patient received a dose of dabigatran. The plasma level metric may comprise a time-dependent plasma level metric based on the one or more dosage times.
The plasma level metric may comprise one or more of:
The dosage calculator may comprise a machine learning algorithm trained using the plasma level prediction model.
The dosage calculator may comprise a machine learning algorithm trained using simulated population data obtained from the plasma level prediction model.
The dosage calculator may comprise a machine learning algorithm trained using:
The machine learning algorithm may be locked to prevent further adjustment to the machine learning algorithm.
The machine learning algorithm may comprise an adjustable machine learning algorithm. The method may further comprise:
The dosage calculator may comprise the plasma level prediction model.
The dosage calculator may comprise one or more look-up tables defined according to simulated population data obtained from the plasma level prediction model.
Processing the patient data with the dosage calculator to determine the dosage of dabigatran for administering to the patient may comprise:
The selection of available dosage regimes may comprise dosage amounts comprising: 75 mg, 110 mg or 150 mg of dabigatran.
The selection of available dosage regimes may comprise dosage amounts comprising:
The selection of available dosage regimes may comprise selection of a microgranular or liquid formulation for titrating an ideal dosage amount of the ideal dosage regimen.
Processing the patient data with the dosage calculator to determine the dosage of dabigatran for administering to the patient may comprise processing the patient data with the dosage calculator to determine one or more of: a dosage amount; a dosage time; a dosage frequency; and/or a dosage type.
The dosage type may comprise a dabigatran slow-release formulation with a specific release time.
The specific release time may be at least 6 hours.
Indicating the dosage may comprise indicating the dosage to a health care professional and/or to the patient.
Indicating the dosage may comprise indicating the dosage to a health care professional and/or the patient via a user interface. The user interface may comprise a digital app. the method may comprise performing one or more of the steps within the digital app.
Any method disclosed herein for use in stroke prevention, thrombosis treatment, blood clot prevention or dosage management for an invasive procedure on the patient.
The patient may be a cancer patient.
According to a fourth aspect of the present disclosure, there is provided a method of generating a dosage calculator for determining a dosage of dabigatran for administering to a patient, the method comprising:
The simulated population data may comprise simulated population data for at least 100,000 simulated patients.
The simulated population data may comprise simulated patient data, simulated dosages and/or simulated plasma level metrics.
The method may comprise calculating the simulated population data using the plasma level prediction model.
Calculating simulated population data using a plasma level prediction model may comprise:
The simulated patient data may comprise one or more of: a kidney function metric; a patient age; a patient ethnicity; a patient gender; a patient weight; a patient haemoglobin level; a patient left ventricular function; and a patient medication list.
Defining the simulated patient data may comprise: receiving a population distribution for each parameter type of the simulated patient data; and generating each simulated patient by probabilistic selection of each parameter type according to the respective population distribution. Defining the simulated patient data may comprise: defining a plurality of discrete values for each parameter type; and generating each simulated patient data as a different combinations of one discrete value from each parameter type. Generating each simulated patient may comprise generating simulated patient data for every possible combination of one discrete value from each parameter type.
The plasma level prediction model may comprise a time-based differential equation model for modelling a time dependence of a plasma concentration of dabigatran as a function of the patient data.
The plasma level prediction model may comprise a compartment model for modelling a time dependence of a plasma concentration of dabigatran as a function of the patient data.
Training the machine learning dosage calculator may comprise training the machine learning dosage calculator using the simulated population data and real patient data.
The method may comprise validating the machine learning model using further simulated population data. The further simulated population data may be different to the simulated population data.
The method may comprise locking the machine learning dosage calculator to prevent further adjustment to the machine learning dosage calculator.
The method may further comprise calibrating the machine learning algorithm for a patient based on a measured drug plasma level or coagulation measure obtained from a blood test on the patient.
The machine learning dosage calculator may be configured to directly calculate the dosage for administering to the patient by processing patient data and a target plasma level metric.
The patient data may comprise one or more of: a kidney function metric; a patient age; a patient ethnicity; a patient gender; a patient weight; a patient haemoglobin level; a patient left ventricular function; and a patient medication list.
According to a fifth aspect of the present disclosure there is provided a dosage calculator for determining a dosage of dabigatran for administering to a patient, the dosage calculator comprising one or more processors configured to:
According to a sixth aspect of the present disclosure there is provided a computer readable medium comprising instructions which, when executed by one or more processors, cause the one or more processors to carry out any method disclosed herein.
According to a seventh aspect of the present disclosure there is provided a method for administering a dosage of a drug to a patient, the method comprising:
There may be provided a computer program, which when run on a computer, causes the computer to configure any apparatus, including a circuit, controller, converter, or device disclosed herein or perform any method disclosed herein. The computer program may be a software implementation, and the computer may be considered as any appropriate hardware, including a digital signal processor, a microcontroller, and an implementation in read only memory (ROM), erasable programmable read only memory (EPROM) or electronically erasable programmable read only memory (EEPROM), as non-limiting examples. The software may be an assembly program.
The computer program may be provided on a computer readable medium, which may be a physical computer readable medium such as a disc or a memory device, or may be embodied as a transient signal. Such a transient signal may be a network download, including an internet download. There may be provided one or more non-transitory computer-readable storage media storing computer-executable instructions that, when executed by a computing system, causes the computing system to perform any method disclosed herein.
One or more embodiments will now be described by way of example only with reference to the accompanying drawings in which:
The present disclosure relates to a method of generating a drug dosage calculator comprising a machine learning (ML) model (also referred to as a ML algorithm) trained using population data. The population data may comprise drug data for a plurality of patients and can include real population data from one or more real patient studies, and/or simulated data output from a pharmacokinetic-pharmacodynamic (PKPD) model for the drug.
A first step 638 comprises receiving population data comprising real population data, and/or simulated population data calculated using a PKPD model.
A second step 640 comprises training the ML dosage calculator using the population data.
The population data may comprise drug data for a plurality of patients. For each patient, the drug data may include patient data, dosage data and a PKPD metric.
The patient data may comprise one or more patient data parameters that can influence the effect of the drug on a patient. For example, the patient data may include factors that influence the absorption and/or clearance of the drug into and from the body. The patient data may include one or more input parameters used in a PKPD model for the drug. The patient data may include one or more of: a kidney function metric; a liver function metric; a patient age; a patient ethnicity; a patient sex; a patient weight; a patient body mass index; a patient haemoglobin level; a patient left ventricular function; a pharmacogenomic profile (e.g. CYP enzyme profile and/or drug transporter profile); and a patient medication list, among others.
The dosage data may include a dosage of the drug administered to the patient. The dosage may include a nominal dosage and a period of administration, for example a dosage amount administered daily for ten days.
The PKPD metric may comprise any parameter indicative of the effect of the drug on the patient's body. For example, the PKPD metric may comprise a drug concentration in a particular compartment of the body, for example, a plasma concentration, a tissue concentration, a urine concentration etc. The PKPD metric may relate to a physiological measurement, for example a blood pressure or electrocardiogram. The physiological measurement may be a metric for monitoring a disease or condition treated by the drug.
Real population data may comprise a plurality of drug data collected from one or more patient studies or from a drug monitoring program. Each patient record of the drug data may include real patient data (e.g. age, weight, kidney function metric (such as creatinine clearance), BMI, co-medications etc.); real dosage data comprising a drug dosage administered to the patient; and a real (measured) PKPD metric (such as a blood drug concentration or other physiological measurement).
Simulated population data may be obtained from a PKPD model for the drug. The simulated population data may comprise simulated drug data for a plurality of simulated patients. The simulated drug data for each simulated patient may comprise simulated patient data (e.g. age, weight, kidney function metric (such as creatinine clearance), BMI, co-medications etc.); simulated dosage data comprising a simulated drug dosage administered to the patient; and a simulated PKPD metric (such as a blood drug concentration) calculated by processing the simulated patient data and the simulated dosage data with the PKPD model.
The PKPD model may comprise any time dependent model that can output a dependence of a PKPD metric based on received inputs comprising a dosage of the drug and patient data (real or simulated). For example the PKPD model may comprise a compartmental model, such as a two-compartment model, for modelling a time dependence of a central compartment drug concentration as a function of the dosage and the patient data. The PKPD metric may comprise a concentration of the drug in a central compartment. The central compartment may relate to one or more areas of the body where the drug takes effect such as the blood, liver, tissue, brain etc. As discussed further below, the PKPD metric may comprise one or more other metrics such as a physiological measurement (such as a blood pressure, electrocardiogram etc.), maximum drug concentration, a trough drug concentration, an average drug concentration, a time profile of the drug concentration, an area under the time profile, or one or more combinations or ratios thereof.
The patient data may comprise any input variables used in the PKPD model. For example, the patient data may comprise one or more variables that can affect the absorption and/or clearance of the drug from the body. For example, the patient data may comprise one or more of: a kidney function metric; a liver function metric; a patient age; a patient ethnicity; a patient sex; a patient weight; a patient body mass index; a patient haemoglobin level; a patient left ventricular function; a pharmacogenomic profile (e.g. CYP enzyme profile and/or drug transporter profile); and a patient medication list, among others.
The data output from the PKPD model may comprise simulated population data. The simulated population data may comprise data output from the PKPD model following processing of the simulated patient data which may represent a known population variation in the population. For example, a Monte Carlo type approach may be used to generate the simulated patient data using known distributions of each parameter type of the patient data (see below). In some examples, the simulated population data may comprise PKPD metrics output by the PKPD model. In some examples, the simulated population data may comprise dosages for administering to each patient of the population set to achieve a target PKPD metric. The ML model can be trained using the simulated population data to provide the ML dosage calculator.
where Ka is a first order drug absorption rate constant: F is a relative bioavailability of the drug; and D is the drug dosage.
Drug clearance from the central compartment 301, for example via the kidneys, may be defined as:
where M is the drug (mass) cleared from central compartment and Cl is the total body clearance of the drug from the central compartment.
A peripheral compartment 303 may represent locations in the body in which the drug may reside and not necessarily provide a therapeutic effect. The peripheral compartment 303 has a peripheral compartment drug concentration, Cp, and a peripheral compartment volume, Vp. A rate of change of the peripheral compartment drug concentration, Cp, resulting from the flow of drug between the two compartments may be defined as:
where Q is an intercompartmental clearance between the central and peripheral compartments.
Combining everything together, the rate of change of the central compartment drug concentration, Cc, can be written as:
Equations 1 to 4 illustrate that using a PKPD model directly as a dosage calculator results in a computationally intensive and time intensive dosage calculator resulting from the iterative time dependent differential equations. As a result, such a dosage calculator is not suitable for deployment at scale on personal mobile devices or the like. The present disclosure addresses this time and resource intensive issue by training a ML dosage calculator based on simulated data from the underlying PKPD model. The resulting ML dosage calculator is less time and resource intensive and can be advantageously deployed at scale to patient personal electronic devices.
The method may comprise calculating the simulated population data using the PKPD model. For example, the method may comprise: (i) defining simulated patient data for a simulated patient population comprising a plurality of simulated patients; and (ii) calculating PKPD metrics for each of the plurality of simulated patients using the PKPD model to define the simulated population data. The simulated population data may comprise the simulated patient data, simulated dosages and the calculated PKPD metrics.
The simulated patient population may comprise at least 10,000 simulated patients, at least 100,000 simulated patients, at least 1,000,000 simulated patients or at least 10,000,000 simulated patients. Defining the simulated patient data may comprise: (i) receiving population distributions for each parameter type of patient data (e.g. age, weight, other medications, liver function, kidney function, pharmacogenomic profile etc); and (ii) generating the plurality of simulated patients by probabilistic based selection of each parameter type according to the respective population distribution. For example, a Monte Carlo type analysis may be used to define the simulated patient data. In other examples, the simulated patient data may comprise a more methodical sweep of the parameter space defined by the patient data. In some examples, the simulated patient data may also include a simulated dosage amount. The simulated patient data (including the simulated dosage) can be processed using the PKPD model to obtain the simulated population data comprising a simulated PKPD metric.
In some examples, defining the simulated patient data may comprise: (i) defining a plurality of discrete values for each parameter type; and (ii) generating each simulated patient data as different combinations of one discrete value from each parameter type. In some examples, the different combinations may generate simulated patient data representative of all combinations of one discrete variable from each parameter type. The parameter values for a particular parameter type may comprise a discrete step between each parameter value that is sufficiently fine to represent a continuous variable according to the biological range of the parameter important to determine drug level.
Following receipt of the population data, the ML dosage calculator can be trained using the population data. The ML dosage calculator may comprise any known ML architecture such as an artificial neural network or a generative model. In some examples, the population data may include both real population data and simulated population data as the ML training data. The ML training data may comprise weightings for each population data set, with a higher weighting assigned to real population data than to simulated population data.
In some examples, training the ML dosage calculator may be performed in two steps. In a first pre-training step, the ML dosage calculator may be trained using simulated population data. A second refinement step may comprise training the ML dosage calculator further using refined training data. In some examples, the refined training data may comprise real population data such as clinical data from real patients comprising patient data, associated drug dosages and resulting PKPD metrics. In some examples, the refined training data may relate to a different drug that has a similar PKPD pathway to the drug used in the first pre-training step. In this way, ML dosage calculators can be pre-trained in a generic way and then refined using real world patient data and/or for a related drug with a similar underlying structure.
The present disclosure also provides a method for determining a dosage of a drug for administering to a patient that utilizes the ML dosage calculator trained using the population data. The method may determine the dosage without requiring physiological measurements. The ML dosage calculator can process patient data for the individual patient to indicate a dosage for administering to the patient. The ML dosage calculator may receive a target PKPD metric and calculate the dosage for administering to the patient by processing the patient data and the target PKPD metric. The PKPD metric may comprise an ideal therapeutic level (ITL) or ideal drug concentration.
A first step 212 comprises receiving patient data relating to a patient for the individual patient. The patient data may be of the same type used to train the ML dosage calculator.
A second step 214 comprises processing the patient data with the ML dosage calculator to determine the dosage of the drug for administering to the patient.
A third step 216 comprises indicating the dosage.
In a first set of examples, the ML dosage calculator may perform the second step 214, by replicating a PKPD model and determining PKPD metrics for a particular dose and patient data. The dosage calculator may perform the calculation iteratively and revise the dosage estimate to obtain a dosage for administering to a patient such that the calculated PKPD metric matches a target PKPD metric, such as an ideal therapeutic level (ITL).
A first step 526 comprises setting an initial value of a dosage estimate, for example based on the received patient data.
A second step 528 comprises processing the dosage estimate with the dosage calculator to determine a PKPD metric. The PKPD metric may comprise a trough drug concentration, Ctrough, a maximum drug concentration, Cmax, a ratio of the maximum to trough drug concentrations, an average drug concentration over a dosing interval at steady state, Caverage, (i.e. once the drug has accumulated and stabilised over a number of days) or any other drug concentration metric described herein.
A third step 530 comprises comparing the PKPD metric to the target PKPD metric. The target PKPD metric may comprise the same metric (Ctrough, Cmax, Caverage, etc) as the PKPD metric.
A fourth decision step 532 comprises determining if a difference in the values of the PKPD metric and the target PKPD metric is within a difference threshold. If not the method proceeds to a fifth step 534 and refines the dosage estimate, before returning to the second step 528.
If the difference in the two values is within the difference threshold, the method proceeds to a sixth step 536 and outputs or indicates the dosage estimate as the dosage for administering to a patient.
In some examples, the loop around the second to fifth steps 528-534 may be performed in an iterative fashion until the values are within the difference threshold. In some examples, the loop may be performed at least two times and a dosage value corresponding to the target PKPD metric may be interpolated. In some examples, the loop may correspond to an optimisation routine.
In a second set of examples, the ML dosage calculator may receive a target PKPD metric and patient data, and directly determine a dosage for administering to the patient.
A major advantage of the ML dosage calculator is that it can operate at significant speed relative to a two-compartment differential equation model. The efficient processing means the ML dosage calculator requires much less processing power than a calculator attempting to implement the differential equations of a two-compartmental model directly. Such lower processing power requirements enables deployment of the ML dosage calculator at scale to HCPs and/or patients, e.g. via a cloud platform and/or personal computing devices. Other advantages of the ML dosage calculator include: (i) a wealth of data available at speed including optimised dosages and PKPD metrics, such as trough levels, Cmax, ratios or time profiles (similar to the profiles of
In some examples, a ML dosage calculator may be locked following training and validation, in that the ML dosage calculator will not “learn” from any future data. Such an approach can improve safety and enable easier regulatory approval by ensuring that the ML dosage calculator does not evolve into an unsafe regime based on an error in further training data. Alternatively, in some examples, such as applications where safety requirements may be more relaxed, for example where the drug is used in terminal illnesses, a highly monitored environment, or a clinical situation in which the HCP has no alternative, the ML dosage calculator may be free to evolve and use live data as further training data to provide a more accurate dosage calculator.
An advantage of training a ML dosage calculator with simulated population data from a verified PKPD model relates to the safety, confidence and regulatory approval of ML models. As illustrated by the above examples, a large number of patients may be required to successfully train and maximise the advantages of a ML dosage calculator. Obtaining patient data for the equivalent number of real patients for a particular drug can be time consuming and expensive. As a result, ML models can struggle to obtain regulatory approval, particularly if part of the training phase is proposed to be conducted as part of deployment because such an approach lacks predictability. Embodiments of the present disclosure relate to training and verifying a ML model with a large simulated patient data set that is based on a verified and accepted analytical model developed from real (and attainable) patient data. The resulting ML dosage calculator is safe, predictable and can be locked to avoid any further learning (with a possible exception for defined and regulatory approved update points).
A yet further advantage of training a ML dosage calculator with simulated data is that it can provide outputs for combinations of co-variants that are not well represented in original patient data set. In this way, the ML dosage calculator can artificially enrich specific cases of interest and improve the weighting and consideration of less common combinations of characteristics thereby reducing bias and inferior predictions.
Furthermore, deploying the safe, predictable, efficient ML dosage algorithm at scale for patients and HCPs can enable the collection of real patient data (including measured PKPD metrics such as drug plasma levels (see “Ongoing Treatment Management” section below)). Such deployment would be infeasible with a two-compartment differential equation based PKPD model due to the processing constraints. The collected real patient data can then be used to evolve the ML dosage calculator and/or the PKPD model to further improve the accuracy and precision of the dosage calculator.
The following description often refers to the generation, use and advantages of a ML dosage calculator in relation to the specific example drug of dabigatran. However, the described generation, use and advantages of the ML model are applicable to any drug whose absorption and clearance can be modelled by pharmaco-kinetic modelling and/or pharmaco-dynamic modelling such as two-compartment, differential equation based models. For example, drugs including vancomycin, phenobarbital, butalbital, etomidate can be modelled using two-compartment PKPD models. Other drugs may be modelled using similar PKPD models to those of equations 1-4 described above. Specific refinements for other drugs, similar to equations 5 to 7 described below, can be provided by corresponding real patient studies and PKPD analysis for the corresponding drug and contributing individual patient data (which may differ from the patient data listed for dabigatran). For dabigatran the central compartment comprises the blood plasma. For other drugs, the central compartment may represent locations other than blood plasma, such as the liver, muscle tissue, the brain etc. As a result, the dosage calculator may be generated from real population data and/or simulated population data output from a PKPD model for the drug that predicts a time dependence of an appropriate PKPD metric. For dabigatran the PKPD metric is a plasma level metric, however the PKPD metric could include other metrics such as drug concentration in the central compartment (e.g. liver, muscle tissue etc).
The body's clotting system has evolved to a particular equilibrium. This represents a trade-off. Clotting is a protective factor to repair internal blood vessel breakdown or external wounds. However, too much tendency to clot and blood vessels become blocked when this is not desired. This clot can then break off and embolise to distant blood vessels with catastrophic consequences. There is a constant turnover of the multiple molecular components involved in clotting.
Clots can form in any part of the vasculature, in particular the veins of the leg, arteries of the thorax and neck and within the heart. Clots that either embolise from the heart or neck arteries or form directly within the cerebral vasculature can cause a stroke.
Clot is more likely to form in the heart if the heart chambers are enlarged or do not contract normally. Atrial fibrillation (AF), characterized by disorganized atrial electrical activation and contraction in the heart muscle, accounts for ˜30% of all hospitalisations for heart rhythm irregularities, occurring in almost 10% of people over the age of 80 with prevalence increasing as the population get older. The clinical consequences of uncontrolled AF results in a 5-fold increase in stroke and blood clots requiring hospitalisation with consequential increasing health costs which in the USA alone is currently estimated to be $8 billion per year.
Management of AF includes methods to restore normal sinus rhythm, control heart rate, and where possible prevent recurrence. When these methods have not succeeded anticoagulants are used to inhibit the formation of the clots. In the past, warfarin has been successfully used, significantly reducing stroke by about 60%, but can lead to severe risks of bleeding due to a narrow therapeutic range, differences in metabolism between individuals, and multiple interactions with a number of co-administered drugs and food stuffs. Warfarin is difficult to prescribe at the correct dose and demands frequent measurements of blood clotting times (international normalised ratio, INR levels) to allow for regular dose adjustment.
Another disease in which anticoagulants are used is in the treatment of thromboembolism where clots are formed in the veins (VTE) and if untreated can lead to disability (pain, scaling ulcers, oedema in the legs), deep vein thrombosis (DVT) and if clots break off can cause pulmonary embolism and death. As many as 600,00 VTEs occur each year in the USA.
New drugs have been developed which allow simplified dose management and have been shown to reduce the chance of developing a major bleeding event when compared to warfarin. These direct acting oral anticoagulants (DOACs) act through different more targeted mechanisms. Whereas warfarin and other similar anticoagulants are indirect inhibitors of Vitamin K through both intrinsic and extrinsic pathways, DOACs work directly on the common pathways lower in the clotting cascade. For example, one DOAC, dabigatran, acts directly by inhibiting the thrombin molecule both in free and bound forms. Thrombin is central to the formation of blood clots. These newer modes of action lead to fewer monitoring requirements, less frequent follow-up, more immediate drug onset and offset effects, particularly important in relating plasma drug levels to activity and fewer drug and food interactions
Dabigatran etexilate is a DOAC licenced for stroke prevention in non-valvular atrial fibrillation, as well as the treatment and prevention of venous thromboembolism; it has been studied in many thousands of patients. The phase III clinical trial RELY study has shown that it is superior to warfarin in patients with nonvalvular AF and resulted in similar or lower rates of both ischemic stroke and major bleeding compared with adjusted-dose warfarin (INR of 2.0 to 3.0) with the advantage of a small reduction in the risk of intracranial bleeds. Dabigatran is also used to reduce the tendency to form clots and thus the incidence of stroke in patients from other systemic embolisms.
There are disadvantages to the DOACs such as dabigatran, including lack of efficacy and safety data in patients with severe chronic kidney or hepatic disease, or those with significant valvular disease, lack of easily available monitoring of blood levels and compliance, and higher patient cost in some health care areas. Additionally, whilst reversal agents are now available for some DOACs, they are expensive and do not cover all forms of bleeding.
There are three doses of dabigatran available in Europe (75, 110 and 150 mg) to be taken twice a day, but only two in the USA (75 and 150 mg). Prescriptions may mix tablet doses to achieve recommended posology for individuals who may not meet the normal criteria.
The limitations in dosage quanta and the prescribing guidelines can result in very limited dosing flexibility for a health care professional (HCP). As a result, patients can be prescribed an inappropriate starting dose and some patients may be excluded from treatment with dabigatran, for example those with kidney impairment. FDA Real World Evidence studies have provided evidence that currently DOACs may be both under dosed leading to an excess of thrombotic events and overdosed, particularly when there is renal impairment, leading to an excess of haemorrhage.
The issue of incorrect starting dosage (or even a correct starting dosage) can be further exacerbated by insufficient monitoring. Unlike warfarin, routine monitoring of blood coagulation in patients taking DOACs is not currently recommended, except in certain patients, particularly those with cryptic thromboses, renal failure, the elderly, or those taking certain co-administered drugs. In these latter groups, monitoring is needed but not often undertaken. Recommendations are that such patients should be reviewed at least once per year. This recommendation is often poorly adhered to and often no review is performed at all. Reasons include delegation to general practitioners and general physicians who may be too busy and/or do not have sufficient information and understanding. Furthermore, even when haematology experts give clear instructions to primary care, the instructions are often not followed properly.
A further challenge resulting from insufficient monitoring is managing risk around the times of invasive procedures for an operation such as hip and knee replacements or a lumbar puncture. Management of patients in the perioperative period involves a careful assessment of the relative risk of bleeding or the possibility of a thromboembolic event. Current guidelines are a one size fits all with the result that some patients have their anticoagulation stopped too soon and are thus rendered at high risk of clots, whereas others may have their anticoagulation stopped too late and have higher bleeding risks. As well as risks to the patient, there is also secondary harm from bed blocking from excessive stay in hospital whilst waiting for anticoagulation to wear off. Furthermore, there is also delay in any investigations. For example, an unplanned lumbar puncture which may be required at short notice to diagnose a neurological condition may be delayed for an unnecessarily long period because of concerns around ongoing anticoagulation.
A further area of particular need relates to thrombosis risk in cancer. Management is particularly difficult in cancer because as well as an increased thrombosis risk, there is also an increased bleeding risk. The problems with cancer are increasing with the shift to more home based chemotherapy and thus more chemotherapy lines being inserted, further increasing the thrombosis risk. The requirement for particularly precise control and knowledge of actual risks of haemorrhage would be extremely useful.
Dabigatran is formed by the hydrolysis of dabigatran etexilate ester in the GI tract but only ˜5% of the drug is absorbed with the remainder going into the faeces (˜90%). This may be a consequence of a saturation of the Pgp transporter system by which the drug is absorbed. Peak absorption occurs after ˜1 hour and although food does not affect the total absorption it can delay the peak by as much as 2 hours. Interestingly, if the pellets are removed from the capsule the availability is increased by 75%. After absorption the drug is slowly eliminated with a half-life of approximately 12-17 hours and thus to maintain reasonably constant levels, the drug is often administered twice a day. The total and peak systemic exposure are dose-proportional in the range 50-400 mg twice daily.
The primary route of elimination of dabigatran is via the kidney (80% of total clearance) but the remainder is metabolized, going to equally active acyl glucuronides representing a mean of ˜10% of parent drug in the plasma. In some cases this may contribute to the activity of the drug. 80% of that absorbed is eliminated into urine. There is large variability in the plasma concentrations achieved with any given dose, depending on absorption, renal function, and other patient factors. However the patient insert leaflet does not provide guidance on dosage change, except for with renal impairment, as follows: “No dose adjustment of PRADAXA is recommended in patients with mild or moderate renal impairment [see Clinical Pharmacology (12.3)]. Reduce the dose of PRADAXA in patients with severe renal impairment (CrCl 15-30 mL/min) [see Dosage and Administration (2.1) and Clinical Pharmacology (12.3)]. Dosing recommendations for patients with CrCl<15 mL/min or on dialysis cannot be provided.”
However, it has been shown that 43% of patients with atrial fibrillation and renal impairment were potentially overdosed and had an increased hazard ratio for major bleeding.
Interaction of dabigatran pharmacokinetics with other drugs known to effect Pgp results in the absorption of dabigatran being drug and time dependent. With ketoconazole the total exposure is increased ˜1.4 fold whilst with verapamil it is increased by 2.4 fold depending on when the drug is taken in comparison with dabigatran. With Quinidine the total absorption was only increased by 53% and the maximum concentration, Cmax, by 56%. These increases can increase the chance of haemorrhaging, depending on the individual.
Because of the large effect of renal function on dabigatran PK and increased bleeding, guidelines in the USA for use of dabigatran are complex and exclude patients that might benefit from dabigatran if prescribed and monitored in a more sophisticated way. The guidelines are:
However there is still large inter-subject variability in the resulting measures of coagulation and outcomes between underdosed (increased morbidity and mortality) and overdosed (increased haemorrhagic incidences and GI disturbances).
Further, potentially large inter-subject variation arises during conversion from the prodrug dabigatran etexilate and low GI absorption with 3-7% of dabigatran actually absorbed from the GI tract contributing to variability in circulating plasma levels (up to 50%).
Several studies have shown relationships between plasma drug levels and clinical outcome, suggesting that trough levels between 75-150 ng/ml provide the best balance between reduction in stroke and minimization of bleeding. However, measuring drug levels in general practice is costly and is infrequently undertaken. A trough level is the plasma level before a next scheduled dose (typically the morning dose). There appears to be relationships between drug levels and particular measures of the activity of dabigatran on coagulation, namely dilute Thrombin Time (dTT) and ecarin thrombin time (ECT), and these could form a way of monitoring patient outcomes.
In summary, a large variability in drug plasma levels between patients is driven by: the influence of renal function; the effect of other drugs; low absorption; and other factors (discussed further below).
The method of
For dabigatran, the patient data includes a kidney function metric, which may be a measured creatinine clearance. In some examples the kidney function metric may comprise other renal function metrics such as Inulin cystatin C, beta trace proteins, 51Cr-EDTA (radioactive chromium complexed with ethylene diamine tetracetic acid), 99Tc-EDTA (radioactive technetium complexed with ethylene diamine tetracetic acid). As described above and below, kidney function is the most significant driver of dabigatran drug plasma level variability between patients. By including kidney function metric, the method advantageously accounts for the biggest driver of variability in dabigatran drug plasma levels—kidney function.
The term dabigatran drug plasma level may also be referred to herein as dabigatran blood plasma level, blood plasma level, drug plasma level or simply plasma level. The term level may be referred to as a concentration. As described below, embodiments of the method can process patient data comprising a plurality of parameters that contribute to plasma level variability, such as demographic data, medication data and patient condition data, to advantageously provide more accurate dosages for patients.
For any drug, the method of
As also described below, for any drug, the method of
In relation to the second step 214, the PKPD model for dabigatran may comprise a plasma level prediction model comprising a theoretical model that can predict a dabigatran drug plasma level based on a dosage and the patient data. In some examples, the plasma level prediction model may be a time-dependent differential equation based (or rate equation based) model. The plasma level prediction model may comprise a compartment model such as a two-compartment model, or any other PKPD analytical model for predicting the drug plasma level as a function of patient data.
Estimated values of the specific parameters for the two-compartment model of Equations 1 to 4 described above, can be obtained from a patient study. For example Liesenfeld [2] produced a two compartmental model for Dabigatran. Their model was based on a population pharmacokinetic analysis of all the data from a large prospective randomised open trial (RE-LY) using two doses of dabigatran (110 and 150 mg) compared to warfarin over a period of 2 years where plasma levels were taken from subjects 4 weeks after the start of treatment, before and ˜2 hours after taking the drug and additionally in a subset of patients after 3, 6 and 12 hours. Various covariates (patient data) thought to relate to the drug's PK were collected, such as age, weight, ethnicity, renal function, Creatinine clearance (CrCl), Left ventricular function, heart failure, Haemoglobin, PgP inhibitors including verapamil, amiodarone, clopidogrel, diltiazem, proton pump inhibitors, and H2 receptor antagonists. The analysis showed individual influences on the total exposure of the drug as measured by the area under the plasma curve (AUC), as listed in Table 1 below.
Where Vd=drugs volume of distribution; AUC=Area under the plasma drug level time curve to infinity; Cmax=Peak plasma drug levels; CrCl=creatinine clearance; PPI=proton pump inhibitors.
The following modifiers for the total body clearance, Cl, the central compartment volume, Vc, and the relative bioavailability of the drug, F, based on individual patient characteristics were as follows:
According to one or more embodiments of the present disclosure, a two-compartment plasma level prediction model may be defined by equations 1 to 4, with expressions for Cl, Vc and F adjusted according to equations 5 to 7 and parameter estimates taken from Liesenfeld.
It can be seen from the above equations for the model and from
As discussed below, the provision of a ML dosage calculator based on a PKPD model can enable: (i) the provision of accurate and precise dosages; (ii) the provision of personalised dosages; (iii) the utilisation of a broad range of patient data, including data never before considered in pharmacometrics modelling; (iv) the provision of dosage monitoring and adjustment; (v) the definition of novel dosage regimens; and (vi) the provision of clot and bleeding risk according to time since last medication taken. It is noted that such advantages are applicable to all drugs and not just the specific example of dabigatran.
Embodiments of the present disclosure may comprise ML dosage calculators trained using two-compartment models based on equations 1 to 4 and modified by other expressions for F, Vc and/or Cl derived from other patient studies of dabigatran, including studies conducted using the apparatus and methods disclosed herein. Other embodiments may include ML dosage calculators trained using simpler single compartment models that do not include the peripheral compartment 303. However, two compartment models can provide more accurate results than single compartment models. Yet further embodiments include ML dosage calculators trained using single or two compartment (e.g. equations 1 to 4) based models for drugs other than dabigatran.
The plasma level prediction model can predict plasma levels throughout the day including peak or maximum plasma levels (Cmax) and trough levels (Ctrough) based on the individual patient data. Returning to the method of
In some examples, the dosage calculator may receive a target plasma level metric and calculate the dosage for administering to the patient by processing the patient data and the target plasma level metric. The target plasma level metric may comprise an ideal therapeutic level (ITL) of the trough plasma level (e.g. 75-150 ng/ml or a more targeted level such as 112.5 ng/ml). The ITL is intended to provide the best balance between the reduction in risk of thrombosis and embolism balanced against increased risks from haemorrhage. The target plasma level metric may relate to other blood plasma metrics such as maximum concentration and may be personalised based on the patient data, as discussed under the “personalised target PKPD metrics” section below.
As described above, the patient data may comprise any variables used as inputs to the PKPD model used to train the ML dosage calculator. The patient data may comprise one or more variables that can affect the absorption and/or clearance of the drug from the body and may include one or more of: a kidney function metric; a liver function metric; a patient age; a patient ethnicity; a patient sex; a patient weight; a patient body mass index; a patient haemoglobin level; a patient left ventricular function; a pharmacogenomic profile (e.g. CYP enzyme profile and/or drug transporter profile); and a patient medication list, among others.
For dabigatran, current regulatory guidelines for dosing only account for a limited amount of patient parameters. In the EU, the starting dose is based on weight, age, mode of treatment (preventative or reactive) and co-medication. Dosing guidelines can be particularly complex for pediatrics. In the USA, dosing guidelines are based on the renal function of the individual since the majority of the drug is cleared through the kidneys but doesn't take into account patient demographics nor the large inter-subject variability in drug absorption described above. However because only two doses are available, dosing flexibility is limited.
The above described plasma level prediction models indicate that a kidney function metric, specifically creatinine clearance, CrCl, is the largest contributing factor to drug plasma level variability among the patient population. Age, weight, haemoglobin, sex, ethnicity, the presence of heart failure and other medications (specifically, proton pump inhibitors, amiodarone and verapamil) can also all modulate the circulating drug level in the blood plasma.
Returning to the first step 212 of
Patient weight can present a particular issue with present guidelines as international guidelines for dabigatran only extend up to 140 kilograms. There is also limited evidence for patient body mass index (BMI) exceeding 40. Increasingly, patients weigh more than this limit and there is no knowledge as to what to do for them in terms of DOACs, with many then receiving warfarin with consequent higher risk. By including weight as a parameter of the patient data, the dosage calculator can advantageously account for patients with excessive weight.
Additionally, endothelial changes, with disease or just age, can affect absorption of dabigatran. By including age as a parameter of the patient data, the dosage calculator can advantageously account for the age dependency on plasma drug levels.
The above described plasma level prediction models and calculators can require values for age, sex, CrCl, hemoglobin, weight, ethnicity, heart failure, and co-medication drugs. While the use of specific interfering drugs is likely to be known, along with weight, age, sex and the likely presence/absence of heart failure, in some circumstances, CrCl and/or hemoglobin level may not be known without prior testing.
In some examples, the values of CrCl or hemoglobin may be provided by population based surrogate estimates. For example, an estimated creatinine clearance rate (eCCR) can be provided by the Cockcroft-Gault formula. Creatinine clearance can be estimated using serum creatinine levels. The Cockcroft-Gault (C-G) formula uses a patient's weight (kg) and gender to predict CrCl (mg/dL). The resulting CrCl is multiplied by 0.85 if the patient is female to correct for lower CrCl in females. The C-G formula is dependent on age as its main predictor for CrCl. The C-G formula can be written as:
Serum creatinine can be measured, derived from age related change and potentially further revised according to other diseases or conditions. In the absence of a direct measurement of CrCl, the eCCR can be substituted for CrCl in the dosage calculations described herein.
In some examples, similar substitutions can be made for haemoglobin level. Haemoglobin values in men between 20 and 60 years of age are typically 14.5 to 15 Gm. per 100 ml., the higher values tending to occur among younger men. After the fifth decade there are progressive and marked decreases to an average of 12.4 in men between 80 and 90 years of age. In women from 20 years of age onward the average haemoglobin values remained near 13 Gm. per 100 ml (Hawkins [3]). In the absence of a direct measurement of haemoglobin, these relationships can be used to define a surrogate estimate to substitute for haemoglobin in the dosage calculations described herein.
The present disclosure may encompass other PKPD models and resulting trained ML dosage calculators that are available now or may be available in the future following appropriate patient studies.
Therefore, according to embodiments of the present disclosure, the patient data for dabigatran may include other patient parameters that may be endothelial function drivers. These include one or more of: smoking history; alcohol intake; a patient thrombosis history; a patient haemorrhage history; a patient cancer history; a patient cardiovascular history; a patient metabolic history; a patient platelet count; patient genetic determinants; a patient haematocrit; a patient liver function; a patient blood pressure; patient co-conditions; patient activity level; patient dosage compliance; family history of thrombosis; family history of strokes; family history of bleeding (e.g. anaemia); and family history of hypertension.
The patient data may also include medications other than those specifically mentioned above, including: a calcium channel blocker; a nonsteroidal anti-inflammatory drug; a H2 receptor antagonist, clopidogrel, aspirin, and diltiazem. For any drug, the patient data may also comprise Cytochrome P450 data. Cytochrome P450 are a suite of enzymes which mainly occur in the liver, but can also be found throughout the body, which metabolise drugs. In some examples, an administered drug can reduce or increase the activity of these enzymes which may result in other co-administered drugs having higher or lower plasma levels. For dabigatran, other drugs can have effect on dabigatran blood levels by modifying how it is metabolised. Therefore, models and calculators utilising Cytochrome P450 data can account for drug interactions.
The patient haemorrhage history may include haemorrhage events indicating a severity and time of event. The patient haemorrhage history may relate to the endothelial state or bleeding propensity such as propensity to superficial bruises on arms and legs and other measurements of skin elasticity. Other methods for assessing blood vessel data and/or endothelial state include image analysis of retinal vessels from fundal photography. The flexibility of the model facilitates incorporation of surrogate measures of endothelial state that may be incorporated into future plasma level prediction models, for example by assessment and training using Machine Learning capabilities.
The patient thrombosis history may include thrombosis events indicating a severity and time of event. The family history of thrombosis may include first degree relatives and may include those with an age less than 50 years.
The patient history of at risk events may include recordings of intermittent atrial fibrillation events, including their duration, time of occurrence and rate. Such recordings may be provided from cardiac data from a smart watch.
In relation to cancer history, cancer has a particularly high thrombotic risk because the endothelium becomes very activated, which then releases various coagulation factors. Even with treatment with DOACs, thrombosis can commonly occur in cancer settings, with a high recurrence rate of approximately 17 to 18%, yet also a risk of major haemorrhage of 20%. This is particularly so with certain cancers such as lung, pancreas and colon, demanding even more patient monitoring and precise dosing. Overall, it is estimated that even in the best centres, only one third of patients with thrombosis get well managed and that elsewhere where management is by a general physician or respiratory physician, optimal management is received in less than 10% of cases. As described below, embodiments of the disclosure can include a personalised target plasma level metric for cancer patients to account for their unique PK-PD. As also described below, embodiments of the disclosure may include enhanced monitoring to calibrate and/or personalise the plasma level prediction model and dosage calculator.
Patient genetic determinants may supplant, refine or improve the model or calculator classification of ethnicity (for example).
The patient activity level may represent personal risk exposure, for example an indication that the patient partakes in hazardous activities or other lifestyle factors, such as recreational drugs, physical activity, etc.
The patient co-conditions may include an indication of whether the patient suffers from AF and an extent of the AF. If the extent of the AF is particularly bad with dizzy spells fainting and tachycardia then higher doses would be warranted.
If the patient has a family history of stokes or hypertension then higher dosing with some bleeding may be more tolerated.
Patient dosage compliance may indicate a patient's propensity to take their medication in accordance with a prescribed regimen. The plasma level prediction model and dosage calculator may be refined to account for patient compliance (e.g. via future patient trial). The method may include monitoring a patient compliance and adjusting the dosage accordingly.
The patient metabolic history may be indicative of diabetes which can modify coagulation and blood circulation as a comorbidity.
One or more of the above patient parameters may be incorporated into the PKPD model and/or ML dosage calculator via an appropriate patient trial. Patient data, including the one or more (new) patient parameters, together with patient dosage (amount and time) and directly measured PKPD metrics can be recorded to evolve the PKPD model and/or ML dosage calculator to account for the additional dependency of the one or more new patient parameters accordingly.
The patient data may comprise yet further patient data parameters. For example patient data parameters may include those that can influence personalised target PKPD metrics as discussed in the relevant section below. Further patient data parameters may include side effect reporting (see section “side effect monitoring” below), a time a dose was taken and other examples described herein.
Patient data may be received by one or more of: manual data entry, by either a patient, HCP or third party at a computing device such as a personal computing device; data from medical records stored on a database or similar; and receiving physiological measurements, for example from medical devices or clinical databases. For example, patient data may include cardiac data from a smart watch that can indicate periods of AF.
The method of
In the illustrated embodiment, patient device 1054 is a smartphone. However, the invention is not limited in this respect and the patient device 1054 can take many other forms, including but not limited to a mobile telephone, a tablet computer, a desktop computer, a voice-activated computing system, a laptop, a gaming system, a vehicular computing system, a wearable device, a smart watch, a smart television, an internet of things device and a medicament-dispensing device.
The patient device 1054 may be configured to gather one or more parameters of the patient data. The patient data can be obtained via manual data entry using a human interface device of patient device 1054 and/or from a remote source via the network 1058.
The patient device 1054 may comprise a memory (not illustrated) for storing the patient data and/or outputs of the ML dosage calculator. Such data may also be stored on a database 1062 as networked or cloud-based data storage.
The patient device 1054 may have one or more applications (or apps) installed on a storage medium associated with the patient device (not shown). The one or more apps may be configured to perform any of the computer implemented methods disclosed herein. The one or more applications may be configured to assist the patient in providing the patient data and/or may include the dosage calculator for processing the patient data. The one or more applications may be downloaded from a network, for example from a website or an online application store.
In this example, the system 1054 further comprises a data processing device 1060 that is communicatively coupled to the patient device 1054 via the network 1058. In the illustrated embodiment network 1058 is the internet, but the invention is not limited in this respect and network 1058 could be any network that enables communication between patient device 1054 and data processing device 1060, such as a cellular network or a combination of the internet and a cellular network.
The data processing device 1060 may supplement the patient device 1054 and perform one or more steps of any of the computer implemented methods disclosed herein. For example, in some embodiments, the patient device 1054 may receive the patient data and provide the patient data to the data processing device 1060. The data processing device 1060 may then process the patient data with the ML dosage calculator to determine the dosage for administering to the patient. The data processing device 1060 may then provide the dosage to the patient device 1054 or a HCP device 1064 (also referred to as clinician device 1064) either or both of which may indicate the dosage. In this way, the data processing device 1060 provides networked, server based or cloud based, processing capability to the system for performing the computer implemented methods.
The data processing device 1060 may be coupled to the database 1062 that can store the patient data and/or the outputs of the dosage calculator.
In this example, the system 1052 includes a clinician data processing device 1064 that is communicatively coupled via network 1058 to the patient device 1054 and the data processing device 1060. The clinician data processing device 1064 may be broadly similar to patient device 1054, offering a similar set of functionality. Specifically, the clinician data processing device 1064 enables patient data to be collated or received. Clinician data processing device 1064 is contemplated as being physically located at a HCP's premises during its use, such as a clinic, a doctor's surgery, a pharmacy or any other healthcare institution, e.g. a hospital. Clinician data processing device 1064 may include one or more sensors, and/or be configured to control one or more separate sensors, which sensors are capable of gathering information about the patient, e.g. a blood pressure sensor.
It is also contemplated that clinician data processing device 1064 is typically used by a medically trained person with appropriate data security clearance, such that more advanced functionality may be available than via the patient device 1054. For example, the clinician data processing device 1064 may be able to access a medical history of the patient, generate a drug prescription for the patient, place an order for medication, etc. Access to functionality may be controlled by a security policy implemented by a local processor or data processing device 1060.
The data processing device 1060 and/or the clinician device 1064 may have an application installed that is compatible with or the same as the application installed on the patient device 1054.
It will be appreciated that the various steps of the computer implemented methods disclosed herein may be performed in any combination by any of the one or more processors in the patient device 1054, the data processing device 1060 and the clinician device 1064. For example, all steps may be performed by the clinician device 1064 which receives one or more parameters of the patient data locally, from the patient device 1054 or another remote device via the network 1058 and optionally via the data processing device 1060. In a further example, all steps may be performed in a networked back-end on data processing device 1060, with patient device 1054 and clinician device 1064 acting as human interfaces for gathering and indicating data. In a yet further example, all steps may be performed on patient device 1054 with clinician device 1064 merely gathering relevant data from the patient device 1054 for informing or directing the HCP.
It will also be appreciated that one or more of the components of the system 1052 could be omitted depending upon the application. For example, at a clinic setting, the disclosed computer implemented methods could be performed solely on the clinician device 1064. Alternatively, the methods could be performed solely on the patient device 1054 in a domestic setting.
The dosage calculator may be deployed to any of patient device 1054, data processing device 1060, and clinician device 1064 as a digital app. The digital app may indicate to the HCP a dose (starting dose, continuing dose or revised dose, timing of dose) for administering to a particular patient based on computed changes in drug levels to achieve the target PKPD metric.
For the specific example of dabigatran, the digital app may also summarise important factors (other patient data) that affect thrombosis and bleeding and present this to the HCP. These factors can include thrombosis and haemorrhage history, cancer history, pattern of any falls, pattern of renal function and liver function, pattern of platelet count, haematocrit, platelet and blood transfusions and age of patient. The app may also recommend a follow up pattern and prompt for any missing information.
In relation to the second and third steps of
The indicated dosage may comprise a starting dosage for a patient beginning a drug treatment, a continuing dosage confirming that a patient taking the drug is dosed at a correct level, or a revised dosage suggesting that a patient's dosage should be adjusted in view of updated patient data (discussed further below under “Ongoing Treatment Management” section).
Wherever dosage is used, this refers to both amount administered, form of medicament (e.g. immediate or controlled release), time taken in relation to food and other events, and periodicity.
In some examples, determining the dosage may comprise: (i) processing the patient data with the ML dosage calculator to determine an ideal dosage regimen; and (ii) selecting the dosage for administering to the patient from a selection of available dosage regimes, based on the ideal dosage regimen. For the specific example of dabigatran, the second step of the method of
As described below, the present disclosure encompasses novel dosage forms that increase the flexibility in selecting the dosage amount and selecting the dosage amount may comprise selecting dosage amounts that are: increments of 1 mg, 5 mg, 10 mg or 25 mg. Selecting the dosage regimen may also comprise selecting dosage forms such as liquid formulations or micro-granule formulations as described below.
Determining and indicating the dosage regimen may comprise determining and indicating one or more of: a dosage amount; a dosage time; a dosage frequency; and/or a dosage type. For the specific example of dabigatran, the dosage frequency may comprise twice daily as is the current standard for DOACs. However, the disclosure also encompasses other dosage frequencies including once daily or less than daily which can improve patient compliance and multiple times daily which can reduce the maximum plasma level, Cmax. The method can determine and indicate that for certain patients the medication should only be taken every other day (for example if creatinine clearance is very low), or three times per day (for example if creatinine clearance very high). Or for other patients the method can determine and indicate that twice daily dosing will result in unacceptably risky Cmaxs to achieve a desired Ctrough, and therefore either a slow release formulation should be given or an increased frequency of dosing with a lower dosage form.
The dosage time may comprise a time for the patient to take their medication. The dosage time may be provided as simple reminder prompts to a patient to take their medication. The dosage time may also be specific to an event such as upcoming or recently performed surgery. The dosage type may relate to different dosage forms such as solid dose, liquid dose or micro-granule formulation.
The disclosed ML dosage calculators, digital apps and associated methods can encompass dosage amounts, formulations and regimens beyond those currently approved.
For dabigatran, as noted above, there are only three doses of immediate release in the EU 75, 110 and 150 mg and no explanation is available for this uneven distribution particularly as the PK is linear. In addition the recommended dosages are complex depending on the patient's weight and for children's age so that a juvenile less than 18 who weighs 30 kg needs a dose of 150 mg BID whilst an individual who is aged the same but weighs 60 kg needs 260 mg BID made up of a 150 and a 110 mg tablet BID. In the USA only 2 doses are licensed, 75 and 150 mg, which makes dosage titration even more difficult
Thus the need for other doses is apparent such as 25, 50, 100, 200, 250 mg in addition to those already available which can provide the right dose for the right patient using the app algorithms described above.
All of the available doses are immediate release producing peak levels after approximately 1 hour when taken without food and 3-4 hours after food. Since the mean half-life is only 12 hours, a second dose is required in the evening to achieve stable plasma levels. Multiple dosing can result in poor patient compliance. Therefore, there is a need for the development of sustained release formulations. Sustained release formulations would have the advantage of not only allowing once a day dosing but also reducing the maximum drug plasma level, Cmax, which is associated with bleeding episodes.
The disclosed methods and calculators may also determine a dosage regimen that increases the frequency of dosing to achieve optimal trough levels without exceeding a threshold value of Cmax. The inventors have realised such a capability is desirable because haemorrhage risk is more related to Cmax in contrast to thrombosis risk which is more related to overall drug exposure and hence Caverage, AUC or blood trough levels. Overall it may be desirable to minimise the Cmax/trough ratio, and avoid Cmax exceeding certain levels. Computing such profiles for particular patients has hitherto not been practically possible, but is enabled by the methods and apparatus disclosed herein.
The half life of dabigatran is approximately 12-14 hours and thus twice a day dosing is recommended. However, controlled release formulations may be developed for such purposes, for example using a microgranule formulation (or any other known formulation for controlling solubility and half life) which would only need to be taken once a day to improve compliance and reduce multiple peaks and troughs of plasma drug levels. Each microgranule may exhibit a controlled release profile and development of multiple doses may be facilitated by altering the number of microgranules in the capsule without having to reformulate and test for each individual dose. It is also envisaged that a reformulation can address the issue of hygroscopicity that prevents the licensed formulation of dabigatran from being placed in dosette boxes or similar in patients requiring such assistance devices.
More generally, for any drug, the disclosed PKPD models, ML calculators, apps and methods can accommodate any changes to dosage form, potentially following clinical trials to refine and validate the PKPD model and/or ML dosage calculator. The disclosed PKPD models, dosage calculators, digital apps and associated methods can account for other dosage amounts as described above. Indeed, the disclosed systems and methods can perform better when more precise dosage amounts are available.
Generally, for any drug, embodiments of the dosage calculation method, particularly those of the digital app implementation, may also process the patient data to determine and optionally output one or more PKPD metrics. The dosage calculation method may comprise determining one or more of: a maximum drug concentration (Cmax); a trough drug concentration; a ratio of the maximum drug concentration to the trough drug concentration; a drug concentration time profile; an area under the curve of the time profile; a ratio of the maximum drug concentration to the area under the curve of the time profile; and the average daily drug concentration at steady state, Caverage. The dosage calculation method may comprise outputting one or more of these PKPD metrics to the patient or HCP.
As discussed further below (under the “Personalised Target PKPD Metrics” section), the maximum drug concentration, the trough drug concentration and/or the ratio between these may be used to adjust the target PKPD metric to ensure an optimal balance between risks arising from too high a dose (e.g. side effects) and too low a dose (e.g. treatment ineffective). For the specific example of dabigatran, these risks manifest respectively as bleeding risk and thrombosis risk. Outputting these PKPD metrics to a HCP can enable the HCP to manually adjust the target PKPD metric.
In some embodiments, the dosage calculation method may comprise outputting the drug concentration time profile or risk levels derived from the drug concentration time profile. The drug concentration level time profile may comprise data similar to that represented in
For the specific example of dabigatran, the digital app may use the drug concentration time profile comprising a plasma level time profile to determine and indicate a patient thrombosis risk and/or a patient haemorrhage/bleeding risk according to the plasma time level profile and the dosage time. The digital app may translate circulating drug plasma level to risk based on predetermined plasma level—risk relationships such as those illustrated in
Indicating the bleeding risk may comprise indicating to the patient (e.g. via the digital app), times of day at which they are at increased or reduced risk of haemorrhage and thus able to adapt their behaviours accordingly. In some examples, a HCP portion of the app may output the bleeding risk, the plasma level time profile and/or a circulating drug plasma level (derived from the time profile) to a HCP in advance of an invasive procedure. Such information provided to the HCP can guide timing of procedures and how long a medication should be withheld before a particular invasive procedure, such as surgery, and how soon a procedure can be carried out. In some examples, the digital app may indicate a time that the patient or HCP should wait before undergoing an invasive procedure (or changing medication to a different anticoagulant). The output plasma level metrics can be used to determine if the dabigatran plasma levels are suitable for surgery, for example where a cut off of ˜50 ng/ml is thought to be acceptable, aiding the clinical decision. It is recognised that different surgeries carry different risks and the thresholds may be adapted accordingly.
In some examples, a patient facing portion of the digital app may indicate the thrombosis risk including advising isolated periods of thromboprophylaxis for events where they are at higher risk of a further thrombosis, for example, surgery or flights of longer than four hours duration, after the patient has completed regular anticoagulation for the original clot. The app could advise on when to increase back to therapeutic anticoagulation from prophylactic after pausing anticoagulation for a planned procedure. This latter aspect could be either presented in a patient facing app, and/or a clinician facing app.
In some examples, the digital app may receive patient data from other objective monitoring systems. The digital app may receive activity data, e.g. from wearable devices, to monitor activity levels and impute associated risk elevation or reduction. Increased activity confers lower risk of stroke, but must not exceed a level that bleeding risk is unacceptably elevated. The digital app may also receive blood pressure data from a blood pressure monitoring device. Blood pressure is linked with the probability of stroke, including haemorrhagic stroke. The digital app may adjust or scale the bleeding risk or thrombosis risk based on the blood pressure data and/or the activity data. Incorporating home blood pressure monitoring data into the patient data and providing it to the HCP may also be useful to help the HCP decide if additional treatment is necessary, something that is not routinely undertaken in patients receiving DOACs.
The tendency to clot can be influenced by a variety of patient parameters that vary with time and between individuals, such as those already described above and in addition, hydration state, infection, inflammation, menstruation, blood pressure and the rate of blood flow through blood vessels. Sluggish blood flow can be seen with prolonged immobility, and in particular after surgery. Increased risk of haemorrhage can occur if there is trauma or a medical procedure involving penetration of a needle, or surgery. As already described, blood vessel walls can become weaker as a result of ageing and in particular deposition of amyloid as occurs in the brain, or from inflammatory processes or drugs such as steroids or antidepressants. The beginnings of clot formation may occur as part of the normal range of physiology, but anticoagulant properties of the coagulation cascade prevent extension.
The inventors have realised that the process of clot formation, or thrombosis, takes place over a longer time course than the process of haemorrhage. The balance between thrombosis tendency and haemorrhage tendency will vary between individuals based on their particular body state. The tendency can be shifted by dosing of DOACs. The longer process of thrombus formation is more closely related to the total exposure to the DOAC, whereas the haemorrhage tendency is more related to the maximal drug concentration. Therefore, consideration of both of these parameters and the ratio thereof offers a route to a more personalised approach to anticoagulation to achieve the best risk benefit profile. For example, one approach may be to minimise the ratio of the maximum plasma level, Cmax, to the area under the plasma level time profile. As noted above, current formulations can only suppress Cmax by prescribing multiple doses per day which brings its own problems of compliance. The solution of a controlled release product that can be used once daily while desirable is not yet available.
Disrupting the above balance of risk, and exacerbating incorrect dosing of anticoagulants, is an asymmetry in prescribing tendency amongst HCPs reflecting the psychological desire for blame avoidance. HCPs may perceive the occurrence of a thrombosis which then may lead to an embolism as less blameworthy than occurrence of a haemorrhage (omission vs commission). As a consequence HCPs may tend towards under dosing to avoid the risk of a potential overdose, even though at a population level this leads to worse outcomes. This reflects a psychological flaw rather than rational prescribing.
Embodiments of the dosage calculation method for the specific example of dabigatran, may account for individual clotting and bleeding risk factors by determining the target plasma level metric as a personalised plasma level metric. In some examples, the target plasma level metric may comprise a trough plasma corresponding to the ITL. The personalised plasma level metric may comprise an adjustment to the ITL. As discussed below, the personalised plasma level metric may also include a ratio of the maximum plasma level, Cmax, to the area under the plasma level time profile being less than a bleeding risk threshold.
In examples employing personalised target plasma level metrics, the patient data may include one or more target dependent patient parameters. The target dependent patient parameters may comprise one or more of: reported side effects (see “side effect monitoring” below); a patient thrombosis history; a patient haemorrhage history; a patient cancer history; a patient stroke history; a patient liver function metric; a patient heart function metric; a patient brain state; a patient smoking history; a patient alcohol history; a patient blood pressure; a patient mobility state; a patient menstruation state; a patient inflammation state; a patient infection state; a patient co-medication; a blood clotting metric; and a patient hydration state. The dosage calculation method may determine the personalised plasma level metric based on one or more of the target dependent patient parameters. The target dependent patient parameters may be monitored on an ongoing or regular basis and personalised target plasma level metrics may be refined on an ongoing basis as part of the ongoing treatment management discussed below.
In some examples, the method may determine personalised adjustments to the target plasma level metric based on specific calculations relating to one or more specific target dependent patient parameters, particularly those indicating a time-limited risk resulting from a risk event (fall, surgery etc). Other target dependent patient parameters that result in bleeding or haemorrhage risk may be chronic in nature or lifestyle dependent, such as a genetic clotting disorder, alcoholism etc.
In some examples, the method may output the personalised adjustments to the patient or HCP. For example, the method may indicate a personalised ideal therapeutic trough level for an individual.
In some examples, the dosage calculation method may comprise determining the personalised plasma level metric based on a time since a bleeding risk event and/or a haemorrhage risk event. For example, the method may comprise applying a maximum adjustment to the nominal target plasma level metric immediately following the risk event and tapering the adjustment periodically back towards the nominal target plasma level metric as the time since the event increases. The method may indicate new dosage regimens with each variation of the adjustment.
For example, immediately following surgery, a patient may be at high risk of bleeding for the first 48 hours before the risk drips rapidly. Anticoagulant may be avoided during this period, except for high thrombosis risk patients. Following this initial period, the patient may be most at risk of thrombosis due to inflammation associated with healing and reduced mobility. At this stage, the method may determine a personalised target plasma level metric as a plasma trough level at the upper end of the ITL or above the ITL, e.g. 150 ng/ml. The dosage calculation method may comprise determining and indicating a dosage for administering to the patient that can provide the personalised target plasma level metric. Following a fixed period after the surgery, e.g. a number of days, the dosage calculation method may comprise reducing the personalised target trough level to 120-130 ng/ml and determining and outputting the dosage for administering accordingly. This may continue until the personalised target level metric is reduced to the ITL.
Other thrombosis risk events include: a thrombosis; one or more episodes of atrial fibrillation; dehydration; chemotherapy; an injury; a surgical procedure, particularly large joint orthopaedic procedures such as hip replacements; indwelling central venous catheters such as Hickman lines. Bleeding risk events include: a haemorrhagic stroke; an injury; a surgical procedure; needle penetration; or a patient fall.
A number of events result in both a bleeding risk and a thrombosis risk (e.g. falls and surgery). For such examples, the method may comprise determining the personalised plasma level metric as a ratio of Cmax to the area under the plasma level time profile being less than a bleeding risk threshold. This may be in combination with maintaining the plasma trough level within the ITL. In such examples, the method may comprise determining and indicating a dosage regimen comprising a controlled release formulation or a dosage frequency of more than twice a day, for example 8 hourly or 6 hourly, combined with a lower dosage amount, for example 75 mg or less if suitable dosage forms are available. Although, a higher dosage frequency can reduce compliance, it can be properly managed for carefully managed patients e.g. cancer patients, and in care settings for example post-surgery or in a nursing home. As mentioned above cancer patients can be at risk of both thrombosis and haemorrhage. Therefore, the method may also comprise determining similar personalised plasma level metrics (ratio and/or trough level) and dosage regimens for cancer patients. Furthermore, some patients may have a chronic bleeding or thrombosis risk and experience a thrombosis risk event or bleeding risk event. The method may also account for such dual risk by personalising the ratio and drug dosage regimen in a similar manner.
A patient fall may also provide an indication of a risk of further falls. There is a challenge in judging when the bleeding risk from a fall outweighs benefits of anticoagulation. A rough guide is if there are more than nine falls per year, anticoagulation may be contraindicated, but this depends on the nature of the fall. The dosage calculation method may reduce a target trough plasma level or the ratio of Cmax to the area under the time profile based on the frequency and/or timing of falls as indicated by the patient fall history. Data from devices that measure fall propensity may input into the system and the patient facing app may offer tailored advice as to risks from falls, or other activities, mitigating actions and when, in relation to the timing of medication taking, their risks are highest and lowest.
The risks of bleeding are in part related to age. Reasons are not fully understood, but may in particular relate to the alterations in endothelium with age, including through changes in collagen. The dosage calculation method may reduce a target trough plasma level or the ratio of Cmax to the area under the time profile based on patient age.
An additional problem for females relates to menstruation and contraception. Oestrogen increases the rate of thrombosis, so oestrogen based pills are usually stopped (As an aside, the menstrual cycle can also affect INR readings making Warfarin a difficult drug for menstruating patients). A patient may then have heavy periods and become anaemic and then have blood transfusions. Some may inadvertently fall pregnant. Optimal management is missed for the majority. Most do not know if their periods are abnormal. Data on menstruation patterns may input into the system and the patient facing app may offer tailored advice as to contraception and menstruation, e.g. advise appropriate treatment for heavy menstrual bleeding. The app could also alert the HCP to the issue.
Other target dependent patient parameters that can increase bleeding risk and therefore may require a reduction in target trough plasma level and/or the ratio of Cmax to the area under the time profile include: high patient blood pressure (haemorrhagic stroke risk), poor liver function, the presence of amyloid in the brain, poor renal function, high alcohol consumption and an indication of aspirin, clopidogrel, NSAIDS, steroids, anti-depressants or any other medication that can result in heightened bleeding risk. In some examples, the dosage calculation method may determine a bleeding risk score based on the presence, timing, and/or severity of a bleeding risk event and one or more of the aforementioned bleeding risk factors. The dosage calculation method may use the Hasbled score to determine the bleeding risk score.
Alcohol intake can also increase AF and dysthymias potentially increasing the chance of strokes and thrombosis.
The dosage calculation method may determine a personalised target plasma level metric by decreasing the target trough plasma level if the bleeding risk score exceeds a bleeding risk threshold.
Other target dependent patient parameters that can increase thrombosis risk and therefore may require an increase in target trough plasma level include: genetic profile; active cancer or inflammatory state; and patient hydration. In some examples, the dosage calculation method may determine a thrombosis risk score based on the presence, timing, and/or severity of a thrombosis risk event and one or more of the aforementioned thrombosis risk parameters.
The dosage calculation method may determine a personalised target plasma level metric by increasing the target trough plasma level if the thrombosis risk score exceeds a thrombosis risk threshold.
In some examples, the dosage calculation method may receive the bleeding risk score and/or the thrombosis risk score via data entry from a HCP.
The dosage calculation method may determine a personalised target plasma level metric as a ratio of Cmax to the area under the plasma level time profile being less than a bleeding risk threshold if the thrombosis risk event exceeds a thrombosis risk threshold and the bleeding risk score exceeds a bleeding risk threshold.
In some examples, the method may comprises assigning patients to one of a plurality of sub-groups. Each sub-group may correspond to one or more of the target dependent patient parameters described above. The method may comprise assigning the patient based on time since a risk event, and/or the presence of a particular patient risk factor. Each sub-group may have a corresponding personalised target plasma level metric.
More generally, for any drug, the digital app and associated methods may determine the target PKPD metric as a personalised PKPD metric based on the patient data. In some examples, the app and methods may determine: (i) a first risk associated with too high a dosage based on patient data; (ii) a second risk associated with too low a dosage based on patient data; and (iii) a personalised target PKPD metric based on the first risk and the second risk. For example, the patient data may indicate a susceptibility and/or tolerance to certain side effects and the first risk may be calculated accordingly. The patient data may also indicate a risk of under prescribing the drug to the patient (e.g. the thrombosis risk for the specific example of dabigatran, or a high blood pressure risk for an example of a blood pressure reduction drug) and the second risk may be calculated accordingly.
As the patient's state changes, so does their pharmacokinetics and pharmacodynamics, with the result that they can become under dosed or overdosed. The digital app and associated methods may: receive updated patient data comprising a measured PKPD metric obtained from a physiological test on the patient; process the updated patient data with the ML dosage calculator to determine an updated dosage; and indicate the updated dosage. The measured PKPD concentration may be a drug concentration. The physiological test may comprise any of: a blood test, a urine test, a cerebrospinal fluid analysis, a biopsy or any other physiological test. The measured PKPD metric/parameter may be a drug concentration from the physiological test.
For the specific example of dabigatran, ideally, the patient should be reviewed at least once per year and for high risk patients potentially every 3 to 6 months, or more frequent follow-ups for particularly high-risk patients (e.g. those with renal failure, where signs of bleeding are occurring in preparation for surgery, when the subject is taking drugs with interactions on absorption or where the chance of stroke is particularly high).
A comprehensive DOAC monitoring approach may include one or more of the following items assessed at each follow-up:
Embodiments of the digital app and associated methods may include providing a HCP with a summary of key patient data parameters that have been tracked during the treatment period. The digital app may track any of the patient data parameters described herein for presentation to a HCP. The patient data parameters may be tracked via manual data entry from a patient, HCP or clinic, from clinical records or any other input method. For the specific example of dabigatran, parameters which may be tracked include, amongst others, age, weight, renal function, liver function, bleeding tendencies, any new thrombotic events and falls risks.
Embodiments of the digital app and associated methods may include tracking the aforementioned patient data parameters as updated patient data (for example on a regular basis (daily, weekly, monthly)), processing the updated patient data parameters to determine an updated dosage for administering to the patient; and indicating the updated dosage. For example, changes to the patient data underpinning the PKPD model (such as kidney function or medication for the dabigatran plasma level model) may result in a different calculated dosage to obtain the target PKPD metric. As a further example, for dabigatran, changes to the patient data that indicate an increase in bleeding and/or thrombosis risk may result in a personalisation of the target plasma level metric (as described in the preceding section), thereby resulting in a change to the recommended dosage for administering to the patient. Indicating the updated dosage may comprise: indicating the updated dosage to a HCP at a regular review meeting; and/or indicating the updated dosage to a HCP or other healthcare professional (pharmacist, nurse etc) via an alert if a change in the updated dosage exceeds an alert threshold.
Embodiments of the digital app and associated methods may include tracking or monitoring patient data comprising side effects such as excessive bleeding, GI dysfunction, vomiting, skin rashes etc, as well as the behavioural pattern of the patient (and others listed in side effect monitoring below). The patient data may be provided to the HCP as part of the health care review or alerted if significant, e.g. haemorrhage.
More generally, by tracking the patient data, underlying changes to a patient's physiology with time, age, disease progression or through drug or lifestyle or other environmental factors can be periodically evaluated to maintain personal calibration of the dose yielding ideal therapeutic level for that patient.
For the specific example of dabigatran, ideally an ongoing monitoring program should include measuring drug trough plasma levels. However the measurement of plasma drug concentrations drug levels in both hospitals and the Home Care Provider setting is rare, costly and seldom undertaken.
Specific quantitative measures exist for dabigatran such as plasma dabigatran drug concentration, as well as coagulation measurements such as anti-factor Xa levels, diluted thrombin time (dTT) and Ecarin time (ECT) to directly assess anticoagulation effects. Currently the dilute prothrombin time and Ecarin clotting time (ECT) are used in high risk patients and in particular to indicate when the bleeding risk is sufficiently low to allow surgery, with a typical threshold being a blood level of 50 ng/ml. One issue is that these measures have not been shown to directly relate to clinical outcome and agreed standardized therapeutic ranges have not been established. In addition these more specific tests are not always available to every health care provider.
Other coagulation tests include activated partial thrombin time (aPTT), chromogenic assay (ECA) and the INR test, the latter being the standard test for warfarin and vitamin K antagonists (VKA) compounds although this is usually considered inappropriate for dabigatran.
Unlike with the VKA anticoagulants these tests can relate directly in time to the circulating dabigatran plasma levels due to the lack of PKPD hysteresis.
Several papers (Stangier [4], Jaffer [5]) have related these measures to circulating plasma levels of dabigatran (Cp) at steady state as shown in the following equations:
In addition, new assays are in development that may enable better computation of dabigatran drug plasma levels.
Embodiments of the digital app or associated methods for the specific example of dabigatran, may include receiving patient data comprising a patient coagulation metric (also referred to herein as a patient clotting metric) from a blood coagulation test result. The patient clotting metric may comprise a drug plasma level (also referred to as drug concentration). The patient clotting metric may comprise any of the above blood clotting metrics and the embodiments may comprise calculating a measured drug plasma level based on the blood clotting metric. In some examples, the embodiments include calibrating the dosage calculator and/or plasma level prediction model by comparing the measured drug plasma level to a corresponding plasma level metric calculated by the dosage calculator. Following calibration, embodiments may comprise processing the patient data with the calibrated dosage calculator to determine an updated dosage of dabigatran for administering to the patient.
More generally, for any drug, embodiments may comprise calibrating the ML dosage calculator by comparing measured PKPD metrics to a corresponding PKPD metric calculated by the ML dosage calculator.
A challenge with measuring drug or clotting level via a blood test is relating this to a time of administration and therefore a trough level is typically taken. This can be logistically challenging compared to taking a sample at any time point. Embodiments of the present disclosure allow for individualised calculation of clotting parameters and drug profile by recording the time of medication and the time of blood sample collection (which may be any time) as patient data. Embodiments of the digital app and associated methods may include: receiving a time of drug administration, a time of blood sample and a measured drug plasma level as patient data; and combining the measured drug plasma level, the time of drug administration, and the time of the blood sample with the plasma level time profile estimated by the dosage calculator to adapt the measured drug plasma level to a measurement-derived maximum plasma level, Cmax, or a measurement derived trough plasma level Ctrough. In other words, the digital app and associated methods can convert a blood test taken at any time point to a measurement-derived maximum plasma level or a measurement-derived trough plasma level. It will be appreciated that this time conversion can be applied more generally to a measured PKPD metric for any drug.
Embodiments of the digital app and associated methods may also calibrate the plasma level prediction model and/or dosage calculator based on the measured drug plasma level or clotting metric. Embodiments of the digital app and associated methods may include: receiving a time of drug administration and a time of blood sample as patient data; estimating a drug plasma concentration at the time of the blood sample using the dosage calculator based on the dosage amount and the time of drug administration; and calibrating the plasma level prediction model and/or the dosage calculator based on the difference between the estimated drug plasma concentration and the measured drug plasma concentration. In some examples, embodiments may comprise calibrating the plasma level prediction model and or dosage calculator based on a difference between the maximum plasma level estimated by the dosage calculator and the measurement-derived maximum plasma level and/or a difference between the trough plasma level estimated by the dosage calculator and the measurement-derived trough plasma level. In this way, the dosage can be adjusted to obtain a corrected ITL. Using both trough and maximum levels of coagulation, the best dose for a particular individual can be recommended.
In examples where the digital app and associated methods are used for patients having a high bleeding and clotting risks (e.g. surgery patients, cancer patients, elderly patients, patients prone to falls), regular blood test monitoring may be employed to ensure accuracy of the dosage calculator and/or underlying plasma level prediction model. The methods may comprise updating the dosage calculator, the plasma level prediction model, the personalised target plasma level metrics (e.g. ratio or trough level as described in previous section), and/or the indicated dosage for administering to the patient, based on the calculated or measured drug plasma level.
The availability of coagulation test data also allows for personal calibration of the dose prediction model. The plasma level prediction model and dosage calculator can predict circulating drug for an average patient at any time following drug intake. This would lead to a predicted coagulation test result. By comparing the actual coagulation test result to the predicted, the dose prediction model can be proportionately adjusted for the individual characteristics or vice versa.
If the dose response is personally calibrated with confidence then dosing may be adjusted to accommodate acute or long term changes in patient data, such as adjustment to accommodate temporary or long term anti platelet therapy, for example given following coronary artery stenting (e.g. via personalised target plasma level metrics as described above). Drugs like aspirin or NSAIDs may also be taken inadvertently by a patient for a headache or pain, the patient not realising it has an additive effect with dabigatran on the risk of a major bleed.
In some examples, the digital app and associated methods may be used in conjunction with a low cost home test for plasma dabigatran level. Combining the app and methods with such tests may calibrate the model and optimise/minimise the frequency of future test requirements. In some examples, it may be sufficient to improve prediction by undertaking just one such assessment. Measurement of the drug level at home may be easier and lower cost than clinic tests and potentially be undertaken at a convenient time during treatment (e.g. just prior to a dose, i.e. at trough level), rather than at an arbitrary time during the day or in an emergency period before surgery
Other patient data parameters that may be monitored by the app or associated methods may include signs and symptoms of bleeding, complete blood count, and a comprehensive metabolic panel specifically evaluating liver function tests, albumin, total bilirubin, and serum creatinine.
It will be appreciated that personal calibration via physiological tests is optional. In many examples (e.g. lower risk patients), the HCP may rely on the dosage recommendation and/or PKPD metrics provided by the app and associated methods disclosed herein.
As noted above, the app and associated methods may comprise receiving patient data as self-reported side effects. The digital app and associated methods may take a number of actions in response to detecting a side-effect above a respective sensitivity level, including: alerting a clinician; advising the patient to make a medical appointment; recommending a physiological test for calibration (as described above); adjusting a first risk level; adjusting a second risk level; adjusting a personalised target PKPD metric; recalculating the dosage for administering to the patient based on the adjusted first risk level, adjusted second risk level and/or adjusted personalised target PKPD metric; and indicating an updated dosage for administering to the patient to the clinician or patient. By monitoring and acting on side effects the app and methods can provide important feedback to the HCP on the need to obtain physiological measurements and advise on suitable dosage change for that particular patient.
For dabigatran, the digital app and associated methods may monitor side effects of dabigatran, including those that can occur when the dosage is too high (see table 2), using suitable questions. GI tract disturbances and bleeding in particular may be closely monitored. The first risk level may comprise a bleeding risk level and the second risk level may comprise a thrombosis risk level.
Following a diagnosis 1272 by a health care provider (HCP) 1274, for example a HCP, a first step of the method comprises receiving 1276, at the digital app 1278, patient data comprising demographic details entered by the patient 1280. The method may also comprise receiving 1282, at the digital app, patient data from the HCP 1274 (e.g. via manual data entry or clinical records etc).
Following receipt of the patient data, the method proceeds to processing the patient data 1284 using the dosage calculator to determine a starting dosage regimen for administering to the patient 1280. Following calculation of the starting dosage regimen, the method comprises indicating the dosage regimen 1286 to the HCP 1274. The HCP 1274 may consider the starting dosage regimen calculated by the app 1278 and provide an initial prescription to the patient 1280. The prescription may be provided 1287 to the digital app 1278 as patient data.
The patient 1280 begins treatment. During treatment the method comprises receiving updated patient data 1288 from the patient 1280, such as side effects or events, and receiving updated patient data 1290 from the HCP following physiological tests for PKPD metrics such as blood test results comprising blood clotting metrics, drug plasma levels and/or CrCl. The method proceeds to process the updated patient data 1292 to determine an updated dosage for administering to the patient. Processing the updated patient data 1292 may comprise calibrating the ML dosage calculator based on the physiological test as described above. Processing the updated patient data 1292 may comprise calculating an updated dosage regimen based on a significant change in a patient data parameter driving the calculator, such as a change in the PKPD metric. Processing the updated patient data 1292 may also comprise personalising the target PKPD metric and calculating an updated dosage regimen as described above. The method proceeds to indicating 1294 the updated dosage regimen to the HCP 1274.
The HCP may consider the updated dosage regimen calculated by the app 1278 and provide an updated prescription to the patient 1280. The updated prescription may be provided 1296 to the digital app 1278 as further updated patient data. The lower loop illustrated on the Figure may be repeated as treatment progresses and optionally as the patient makes further or regular visits to the HCP 1274 e.g. for further physiological tests.
The digital app and associated methods may comprise providing instruction, guidance and education to the patient to support them with their ongoing medication. For dabigatran, due to a lack of ongoing review at present, many patients can feel uncertain in relation to their medication which can lead to poor compliance. For example, the consequences of a thrombosis and a DOAC prescription for the patient extends well beyond the direct effects. Patients live with lots of fear and do not know what to do if a further bleed happens. There are two phases to management of venous thromboembolism, for example the active phase and the preventative phase. In the first phase, the aim is to stop clot extension and early recurrence. This lasts one month, perhaps extending in some to 3 months. After this, there is more flexibility in treatment. There is also limited patient understanding of how to take their medication, with many not realising that they need to request a repeat prescription, interpreting their hospital script as all that is required, like for a course of antibiotics. Some may overlook a requirement for certain medications to be taken with food.
Many patients do not know what to do if they develop a further clot and wrongly assume that they should carry out preventative actions, for example with a clot in their leg, moving the leg around, whereas they should do the opposite, keeping the leg elevated and immobile. Many patients have local complications from a thrombosed vein, with a post phlebitic limb very common. This is often badly managed. The nature and timing of how to wear compression stockings and bandages is not known by most. For patients who have had pulmonary emboli, many misinterpret subsequent chest pain and shortness of breath. If benign causes are misinterpreted, this leads to anxiety and exercise avoidance, with consequent deconditioning and increased future risk of cardiopulmonary events.
Similar concerns and uncertainties around a dabigatran prescription may be experienced by patients taking the drug for prophylaxis purposes, such as patients with AF.
The fact that DOACs aren't monitored in the way that warfarin is with a warfarin clinic means there is limited opportunity for education. Overall, it is estimated that 20 to 25% of patients are not taking their originally prescribed dose.
Therefore, as outlined below, the methods and apparatus of the disclosure are intended to provide patient instruction, guidance and education to support patients with their medication thereby improving trust and compliance.
Embodiments of the digital app and associated methods may comprise indicating the dosage to the patient via a patient device. The app may indicate the dosage amount to the patient and may at the same time provide information on how and when to take the medication in relation to food, how to address missed dosing and/or if there are issues with taking other drugs. For example, taking DOACs with food can slow the absorption of the drug by as much as three hours and the app may instruct the patient accordingly. Patient compliance can be an issue withsome drugs, for example dabigatran and anticoagulants, since failure to take a dose can lead to a reduction in the activity. The digital app can reinforce the need to take the medication on a regular basis according to HCP recommendations, through the use of visual, audio and/or tactile prompts, alerts, alarms, reminders and other alert mechanisms from the patient device. For missed doses, the app and associated methods may calculate a new schedule of dosing and/or adjusted dosage amounts (up or down) for a temporary period using the dosage calculator and the target PKPD metric. The app and associated methods may comprise: receiving a new candidate co-medication (from a HCP or patient), comparing the co-medication against a reference co-medication list; and indicating a co-medication instruction. The co-medication instruction may comprise: a revised dosage for the drug, an instruction not to take the co-medication; and/or an instruction to consult the HCP.
As noted above, the app and associated methods may comprise receiving patient data as self-reported side effects. The digital app and associated methods may provide guidance to the patient on the reported side-effects or any other side-effects to allay patient concern. As noted above, the app may alert the patient that they should seek medical care in the case of severe side-effects.
The digital app and associated methods may also incorporate and deliver a comprehensive education program with advice on the self-monitoring of side effects (as described above) and other lifestyle guidance. For dabigatran this may include messaging relating to precautions to avoid the chance of stroke and lifestyle modification improvements, such as smoking cessation, reduction in alcohol consumption, journey durations and habits, and hydration status. As noted above in the outputs section, the app and methods may determine periods of bleeding risk and/or thrombosis risk based on the plasma level time profile and suggest timing of risk activities (sport, long journeys etc) relative to the medication time and maximum plasma level, Cmax, as indicated by the plasma level time profile. In this way, the app and methods can minimise the risk of thrombosis and haemorrhage while enabling the patient to maintain a quality of life. As also noted above, the app and methods, may calculate and indicate an adjusted dose in advance and following high risk procedures (e.g. surgery and long journeys).
The education advice may include instructions on management of a phlebitic limb, including duration of elevation, type of bandage or stocking and application procedure, with tracking of any unwanted consequences and management thereof.
The app and methods may also advise on risks of various activities, in particular contact sports and those with higher risks of falls such as skiing. The app and methods may provide individualised risks for informed patient decisions. For example, the app may advise on strategies to decrease risk during long journeys.
The app and methods may provide advice on patients regarding decrease in haemorrhage risk from lifestyle changes, including improving balance to prevent falls and type of exercise, whilst still ensuring optimal quality of life which could be impacted if activities are curtailed too much.
The app and methods may track respiratory symptoms in patients who have had a pulmonary embolism to help identify whether chest pain and shortness of breath is merely residual effects of the original embolus or is a new cause for concern. The prospective tracking overcomes the problems of patient memory.
For females, the app and methods may advise on contraception and also management of heavier periods from the anticoagulant. The app and methods may incorporate bleeding scores to help determine whether periods are abnormal and warrant additional investigation, to pick up, for example, a new cancer.
For patients who have falls, the app and methods may track number of falls and their type to help with risk management.
The example, ML dosage calculators, PKPD models and methods for calculating personalised target PKPD metrics described herein are exemplary. Treatment algorithms, including computation of ideal and personalised PKPD metrics, may also be expressed by explicit rules (e.g. if . . . then . . . ), Bayesian or other statistical inference derived from population data, or via Machine Learning (e.g. Deep Learning). The type of algorithm may be selected for performance, suitability for context and governing regulatory framework. In some examples, dosage calculators and/or PKPD models may be refined continuously as new data is accumulated or revised periodically for regulatory approval according to governance requirements and patient risk.
Throughout the present specification, it will be appreciated that any reference to “close to”, “before”, “shortly before”, “after” “shortly after”, “higher than”, or “lower than”, etc, can refer to the parameter in question being less than or greater than a threshold value, or between two threshold values, depending upon the context.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2301936.7 | Feb 2023 | GB | national |
