Systems and methods for providing patient-specific dosing as a function of a mathematical model - Patents.com

Abstract

Providing a system and method for providing patient-specific dosing as a function of a mathematical model. [Solution] A system and method for predicting, suggesting, and / or evaluating a suitable drug dosing regimen for a particular individual as a function of the individual's unique characteristics and the particular individual's observed response. A mathematical model of the observed patient response is used in determining the initial dose. The system and method uses the patient's observed response to the initial dose to refine the model, which is used to more accurately predict the expected response to the proposed dosing regimen for the particular patient.

Description

[Technical field]

[0001] The present invention relates generally to the administration of drugs to living patients. More specifically, the present invention relates to computerized systems and methods for processing drug-specific mathematical models, taking into account observed patient-specific responses, to predict, suggest, and / or evaluate suitable drug administration regimens for a particular patient. [Background technology]

[0002] A physician's decision to start a patient on a drug-based treatment regimen involves the creation of a dosing schedule for the prescribed drug. Different dosing schedules will be appropriate for different patients with different patient factors. As an example, dosage, dosing interval, duration of treatment, and other variables may be varied. A proper dosing schedule may be highly beneficial and therapeutically effective, whereas an inappropriate dosing schedule may be ineffective or harmful to the patient's health. Moreover, both underdosing and overdosing generally result in the loss of time, money, and / or other resources, and increase the risk of undesirable outcomes.

[0003] In current clinical practice, physicians typically prescribe dosing regimens based on dosing information contained in the package insert (PI) of a prescribed drug. In the United States, the content of a PI is regulated by the Food and Drug Administration (FDA). As recognized by those skilled in the art, a PI is typically a printed information leaflet that includes a text description of basic information, describing the appearance of the drug and the approved use of the drug. In addition, a PI typically describes how the drug works in the body and how it is metabolized. A PI also typically includes statistical details based on trials regarding the percentage of people with various types of side effects, interactions with other drugs, contraindications, special warnings, how to deal with overdosage, and special precautions. A PI also includes dosing information. Such dosing information typically includes information about dosages for different symptoms or for different populations, such as pediatric and adult populations. A typical PI provides dosing information as a function of certain limited patient factor information. Such dosing information is useful as a reference point for physicians in prescribing dosages for a particular patient.

[0004] FIG. 1 is a flow diagram 100 illustrating a method for creating a drug dosing plan for a patient, which is exemplary of the prior art. As shown in FIG. 1, a typical method involves identifying a patient having patient factors identified in a PI, as shown in steps 102 and 104 of the diagram, and then reviewing the PI dosing information for the selected drug that the physician has chosen to prescribe. As an example, the PI dosing information for Diovan® brand valsartan, an angiotensin receptor blocker used in adults to treat hypertension, heart failure, and the like, manufactured and / or distributed by Novartis Corporation (Summit, NJ, USA), provides, in part, the following dosing information: the type of patient for whom the drug is suitable (hypertensive patients), the type of patient for whom the drug should not be used (pregnant women, those who are volume or salt depleted), the recommended initial dose (80 mg), the recommended dosing interval (once daily), and the accepted range of doses (80-320 mg) that may be legally and / or ethically desirable.

[0005] Such dosing information is typically generated by the drug manufacturer after conducting clinical trials involving administration of the drug to a subject population, closely monitoring the patients, and recording clinical data associated with the clinical trial. The clinical trial data is later compiled and analyzed to generate dosing information for inclusion in the PI. Systems, methods, processes, and techniques for collecting, compiling, and analyzing data to generate dosing information for inclusion in the PI are well known in the art and will not be discussed in detail herein, as they are beyond the scope of the present invention.

[0006] It should be noted that typical dosing information is, in some sense, a generalized translation or composite from data collected in clinical trials of populations including individuals with various patient factors that are deemed appropriate for the "average" patient with "average" factors and / or "moderate" levels of disease, regardless of many of any particular patient's factors, including several patient factors, that may have been collected and tracked during the clinical trial. As an example, based on clinical trial data collected for abatacept, the associated PI provides prescribed dosing regimens at a very coarse level of detail, such as three weight ranges (<60kg, 60-100kg, and >100kg) and associated prescribed dosing regimens (500mg, 750mg, and 1000mg, respectively). The coarse gradations tied to limited patient factors (e.g., weight) ignore many patient-specific factors that may affect optimal or near-optimal dosing regimens. It is therefore well known that the dosing regimen recommended by the PI is unlikely to be optimal or near-optimal for any particular patient, but rather provides a safe starting point for treatment, leaving it to the physician to refine the dosing regimen for a particular patient, largely through a process of trial and error.

[0007] Nonetheless, the physician then determines a prescribed dosing regimen for the patient as a function of the PI information, as shown at step 106. For example, the prescribed dosing regimen may be determined to be 750 mg every four weeks for a patient having a weight falling within a weight range of 60-100 kg. The physician then administers the prescribed dosing regimen, as shown at 108. It will be appreciated that this may be done directly or indirectly, for example, by prescribing the medication, having the medication administered, and / or administering dosages to the patient consistent with the dosing regimen.

[0008] As referenced above, the prescribed dosing regimen may be a good starting point for treating a hypothetical "average" patient, and it is highly likely that the prescribed dosing regimen is not an optimal or near-optimal dosing regimen for the particular patient being treated. This may be due to, for example, individual factors of the particular patient being treated (e.g., age, concomitant medications, other diseases, renal function, etc.) that are not captured by the parameters described by the PI (e.g., weight). Furthermore, this may be due to roughly meeting the recommended dosing regimen (e.g., in 40 kg increments), but proper dosing is likely to be a continuously variable function of one or more patient factors.

[0009] Current clinical practice recognizes this difference. Thus, it is common clinical practice to follow up with the patient after the initial dosing regimen period so as to reassess the patient and the dosing regimen. Thus, as shown in FIG. 1, the physician then evaluates the patient's response to the prescribed dosing regimen, as shown at 110. By way of example, this may involve testing the patient, drawing blood or administering other tests to the patient, and / or soliciting patient feedback so that the patient's response to the previously administered dosing regimen may be observed by the treating physician. As a result of the evaluation and observed response, the physician determines whether a dosage adjustment is warranted, for example, because the patient response is inadequate, as shown at step 112. Such a determination may be made according to existing medical treatment practices and will not be discussed here as it is beyond the scope of the present invention.

[0010] If, at step 112, it is determined that a dose adjustment is not warranted, the physician may discontinue the dose adjustment, as indicated at steps 112 and 118, and the method may end.

[0011] However, if at step 112 it is determined that a dosage adjustment is warranted, the physician will adjust the dosing regimen on an ad-hoc basis, as shown at 114. Sometimes appropriate adjustments are made solely at the physician's discretion. Often, adjustments are made according to a protocol written in the PI or by educational practice. As an example, the PI may provide quantitative instructions for increasing or decreasing the dosage, or increasing or decreasing the dosing interval. In either case, adjustments are made largely on an ad-hoc basis, as part of a trial and error process, and largely based on data collected after observing the effect of the last administered dosing regimen on the patient.

[0012] After administering the adjusted dosing regimen, the patient's response to the adjusted dosing regimen is evaluated, as indicated at step 116. The physician then determines whether to adjust the dosing regimen, as indicated at 118, and the process is repeated.

[0013] Such a trial-and-error approach, which relies on a general prescribed dosing schedule and patient-specific observed responses, works reasonably well for drugs with rapid onset of response.However, this approach is not optimal and often unsatisfactory for drugs that take longer to develop desired clinical responses.Furthermore, the extended time to optimize dosing schedule exposes patients to the risk of undesirable outcomes.

[0014] What is needed are systems and methods for predicting, suggesting, and / or evaluating suitable drug dosing regimens for a particular individual as a function of the individual's unique characteristics that eliminate or reduce the trial and error aspect of traditional dosing regimen development, shorten the length of time to develop a satisfactory or optimal dosing regimen, and thus eliminate or reduce the associated waste of drugs, time, or other resources, and reduce the risk of undesirable outcomes. Summary of the Invention [Means for solving the problem]

[0015] The present invention provides systems and methods for providing patient-specific drug dosing as a function of a mathematical model that is updated to account for observed patient responses. More specifically, the present invention provides systems and methods for predicting, suggesting, and / or evaluating a suitable drug dosing regimen for a particular individual as a function of the individual-specific characteristics and the particular individual's observed response to drugs.

[0016] Conceptually, the present invention provides access to a mathematical model of observed patient response to a drug in a straightforward manner. In prescribing an initial dose, the present invention allows the use of published mathematical models to predict a particular patient's response as a function of the patient's specific characteristics, which are accounted for in the model as patient factor covariates. Thus, the prescribing physician can leverage the model in creating a rationally adjusted initial dose for a particular patient as a function of the particular patient's characteristics, with much greater accuracy than a PI can provide.

[0017] To account for this uniqueness (BSV) of any particular patient, the present invention allows the further use of the observed response of a particular patient to an initial dosing regimen to adjust the dosing regimen. Specifically, the systems and methods of the present invention use the observed response of a patient in conjunction with published mathematical models and patient-specific characteristics to account for BSV that cannot be accounted for by mathematical models alone. Thus, the present invention allows the observed response of a particular patient to be used to refine the model and associated predictions to effectively personalize the model so that it can be used to more accurately predict the expected response to a proposed dosing regimen for a particular patient. By using observed response data to personalize the model, the model is modified to account for inter-subject variability (BSV) that is not accounted for by traditional mathematical models that represent only typical responses for a patient population, or "covariate-typical" responses for a typical patient with certain characteristics that are accounted for as covariates in the model.

[0018] Conceptually, the present invention enables prescribing physicians to create personalized dosing regimens using one or more published mathematical models that reflect actual clinical data, without the loss of resolution of the data and / or models that result from distilling actual clinical data into a relatively crude set of recommendations for the "average" or "typical" patient, as is the case with PIs.

[0019] Such model-based generation of patient-specific drug dosing regimens eliminates or reduces the trial and error aspect of traditional dosing regimen generation, and further reduces the amount of time to generate a satisfactory or optimal dosing regimen, thus eliminating or reducing the associated waste of drugs, time, or other resources, as well as reducing the amount of time that a patient is at risk for an undesirable outcome.

[0020] In general, the systems and methods involve collecting mathematical models created from clinical data collected from patients administered a particular drug, processing the models to create composite models enriched with patient data, and determining patient-specific dosing regimens as a function of patient-specific observed response data that is processed in conjunction with data from the mathematical models. More specifically, the systems and methods use Bayesian averaging, Bayesian updating, and Bayesian prediction techniques to create patient-specific dosing regimens as a function of not only the general mathematical model and patient-specific characteristics accounted for in the model as covariate patient factors, but also the observed patient-specific responses that are not accounted for within the model itself and reflect BSVs that distinguish a particular patient from the typical patient reflected by the model. In an embodiment of the present invention, for example, the following items are provided: (Item 1) 1. A method for providing a patient-specific dosing regimen for treatment with a drug using a computerized drug dosing regimen recommendation system, the system comprising a microprocessor, a memory, and a plurality of mathematical models, each model representing a response profile for a patient population treated with the drug, each model further representing a typical patient response as a function of covariate patient factors, the method comprising: creating a composite model as a function of the plurality of mathematical models and specific patient characteristics corresponding to the covariate patient factors of the models, the composite model representing a response profile for a typical patient having the specific patient characteristics as covariate patient factors; predicting a typical patient response for each of a first plurality of proposed dosing regimens as a function of the composite model; Selecting a recommended exemplary dosing regimen suitable for a typical patient having said particular patient characteristics; providing, as an output from the system, the recommended exemplary dosing regimen; receiving data reflecting an observed response of said particular patient to the administered dosing regimen; updating each of the plurality of mathematical models as a function of the observed patient response data to generate a corresponding plurality of updated patient-specific mathematical models; processing each updated patient-specific mathematical model to produce an updated patient-specific composite model; predicting a patient-specific response for each of a second plurality of proposed dosing regimens as a function of the updated patient-specific composite model; Selecting a recommended patient-specific dosing regimen that is appropriate for said particular patient; providing, as an output from said system, said recommended patient-specific dosing regimen; A method comprising operating the system. (Item 2) 2. The method of claim 1, wherein the composite model is generated by processing the multiple mathematical models using a Bayesian model averaging technique. (Item 3) 2. The method of claim 1, wherein the typical patient response is predicted by processing the composite model using Bayesian prediction techniques. (Item 4) Predicting the typical patient response comprises: receiving as input a plurality of proposed dosing regimens; predicting a corresponding typical patient response for each of said plurality of proposed dosing regimens; The method according to item 1, comprising: (Item 5) 5. The method of claim 4, wherein selecting a recommended typical dosing regimen comprises the system selecting from the predicted typical patient response a recommended typical dosing regimen that is suitable for a typical patient having the particular patient characteristics, the recommended typical dosing regimen being selected for meeting predetermined criteria. (Item 6) 5. The method of claim 4, wherein providing the recommended exemplary dosing regimen as an output comprises outputting a list including a plurality of alternative recommended exemplary dosing regimens. (Item 7) Predicting a typical patient response may include: automatically selecting a plurality of suggested dosing regimens according to a predetermined logic; and predicting a typical patient response for each of the plurality of suggested dosing regimens. The method according to item 1, comprising: (Item 8) 8. The method of claim 7, wherein automatically selecting the plurality of suggested dosing regimens according to predetermined logic comprises selecting a next suggested dosing regimen as a function of a predicted typical patient response to the suggested dosing regimen. (Item 9) 9. The method of claim 8, wherein the predetermined logic provides for a selection of the plurality of suggested dosing regimens to provide a corresponding predicted typical patient response that best meets a treatment objective. (Item 10) 9. The method of claim 8, wherein the predetermined logic provides for selection of the plurality of suggested dosing regimens to find a minimum of an objective function. (Item 11) 9. The method of claim 8, wherein the predetermined logic provides for selection of the plurality of suggested dosing regimens to find a global minimum of an objective function. (Item 12) 8. The method of claim 7, further comprising selecting, from the predicted typical patient response, a recommended typical dosing regimen suitable for a typical patient having the particular patient characteristics, wherein the recommended typical dosing regimen is selected for meeting predetermined criteria. (Item 13) 13. The method of claim 12, further comprising the system providing, as an output, the recommended exemplary dosing regimen. (Item 14) 13. The method of claim 12, further comprising the system providing as output a list comprising a plurality of recommended exemplary dosing regimens. (Item 15) Predicting the patient-specific response comprises: receiving as input a plurality of proposed dosing regimens; predicting a corresponding patient-specific response for each of said plurality of proposed dosing regimens; The method according to item 1, comprising: (Item 16) 2. The method of claim 1, wherein selecting a recommended patient-specific dosing regimen comprises the system selecting a recommended patient-specific dosing regimen from the predicted patient-specific response that is suitable for the particular patient, wherein the recommended patient-specific dosing regimen is selected for meeting predetermined criteria. (Item 17) 13. The method of claim 1, wherein providing as an output the recommended patient-specific dosing regimen comprises outputting a list including a plurality of alternative recommended patient-specific dosing regimens. (Item 18) Predicting a patient-specific response may include: automatically selecting a plurality of suggested dosing regimens according to a predetermined logic; predicting a patient-specific response to each of said plurality of proposed dosing regimens; The method according to item 1, comprising: (Item 19) 20. The method of claim 18, wherein automatically selecting the plurality of suggested dosing regimens according to predetermined logic comprises selecting a next suggested dosing regimen as a function of a predicted patient-specific response to the suggested dosing regimen. (Item 20) 20. The method of claim 18, wherein the predetermined logic provides for a selection of the plurality of suggested dosing regimens to provide a corresponding predicted patient-specific response that best meets a treatment objective. (Item 21) 20. The method of claim 18, wherein the predetermined logic provides for selection of the plurality of suggested dosing regimens to find a minimum of an objective function. (Item 22) 20. The method of claim 18, wherein the predetermined logic provides for selection of the plurality of suggested dosing regimens to find a global minimum of an objective function. (Item 23) 20. The method of claim 17, wherein the system further comprises selecting a recommended patient-specific dosing regimen suitable for the particular patient from the predicted patient-specific response, the recommended exemplary dosing regimen being selected for meeting predetermined criteria. (Item 24) 24. The method of claim 23, further comprising the system providing as an output the recommended patient-specific dosing regimen. (Item 25) 24. The method of claim 23, further comprising the system providing as output a list comprising a plurality of recommended patient-specific dosing regimens. (Item 26) Item 10. The method of item 1, wherein each of the plurality of mathematical models is updated by processing each of the plurality of mathematical models using a Bayesian updating technique. (Item 27) 2. The method of claim 1, wherein each updated patient-specific mathematical model is processed using a Bayesian model averaging technique. (Item 28) 2. The method of claim 1, wherein the patient-specific response is predicted by processing the updated patient-specific mathematical model using Bayesian prediction techniques. (Item 29) 2. The method of claim 1, wherein the recommended exemplary dosing regimen is selected from among a set of exemplary dosing regimens, the recommended exemplary dosing regimen being selected to provide a relevant predicted patient response that closely matches a predetermined criterion. (Item 30) 30. The method of claim 29, wherein the predetermined criteria is a target patient response. (Item 31) 2. The method of claim 1, wherein the recommended patient-specific dosing regimen is selected from among a set of patient-specific dosing regimens, the recommended patient-specific dosing regimen being selected to provide a relevant predicted patient response that closely matches predetermined criteria. (Item 32) 32. The method of claim 31, wherein the predetermined criteria is a target patient response. (Item 33) 2. The method of claim 1, wherein the covariate patient factor comprises at least one covariate patient factor selected from the group consisting of blood concentration levels, blood pressure readings, and hematocrit levels. (Item 34) 1. A method for providing a patient-specific dosing regimen for treatment with a drug using a computerized drug dosing regimen recommendation system, the system comprising a microprocessor, a memory, and a plurality of mathematical models, each of the plurality of mathematical models representing a response profile for a patient population treated with the drug, each model further representing a typical patient response as a function of covariate patient factors, the method comprising: performing Bayesian model averaging to produce a composite model as a function of the plurality of mathematical models and specific patient characteristics corresponding to the covariate patient factors of the models, the composite model representing a response profile for a typical patient having the specific patient characteristics as covariate patient factors; processing the composite model to predict a typical patient response for each of a first plurality of proposed dosing regimens as a function of the composite model by performing a Bayesian prediction; selecting, from a set of predicted typical patient responses, a recommended typical dosing regimen suitable for a typical patient having said particular patient characteristics, said recommended typical dosing regimen being selected to achieve a predetermined target response; providing, as an output from the system, the recommended exemplary dosing regimen; receiving data reflecting an observed response of said particular patient to the administered dosing regimen; performing a Bayesian updating step to update each of the plurality of mathematical models as a function of the observed patient response data to generate a corresponding plurality of updated patient-specific mathematical models; performing Bayesian model averaging of the plurality of updated patient-specific mathematical models to generate an updated patient-specific composite model; performing a Bayesian prediction to predict a patient-specific response for each of a second plurality of proposed dosing regimens as a function of the updated patient-specific composite model; selecting a recommended patient-specific dosing regimen suitable for said particular patient from a set of predicted patient-specific responses, said recommended patient-specific dosing regimen being selected to achieve said predetermined target response; providing, as an output from said system, said recommended patient-specific dosing regimen; A method comprising operating the system. (Item 35) 1. A method for providing a patient-specific dosing regimen for treatment with a drug using a computerized drug dosing regimen recommendation system, the system comprising a microprocessor, a memory, and a plurality of mathematical models, each of the plurality of mathematical models representing a response profile for a patient population treated with the drug, each model further representing a typical patient response as a function of covariate patient factors, the method comprising: performing Bayesian model averaging to produce a composite model as a function of the plurality of mathematical models and specific patient characteristics corresponding to the covariate patient factors of the models, the composite model representing a response profile for a typical patient having the specific patient characteristics as covariate patient factors; processing the composite model to predict a typical patient response for each of a first plurality of proposed dosing regimens as a function of the composite model by performing a Bayesian prediction; selecting a recommended exemplary dosing regimen from the set of predicted typical patient responses that is suitable for a typical patient having said particular patient characteristics, said recommended exemplary dosing regimen being selected to achieve a predetermined target response; operating the system to provide, as an output from the system, the recommended exemplary dosing regimen. (Item 36) 1. A method for providing a patient-specific dosing regimen for treatment with a drug using a computerized drug dosing regimen recommendation system, the system comprising a microprocessor, a memory, and a plurality of mathematical models, each of the plurality of mathematical models representing a response profile for a patient population treated with the drug, each model further representing a typical patient response as a function of covariate patient factors, the method comprising: performing Bayesian model averaging to produce a composite model as a function of the plurality of mathematical models and specific patient characteristics corresponding to the covariate patient factors of the models, the composite model representing a response profile for a typical patient having the specific patient characteristics as covariate patient factors; receiving data reflecting an observed response of said particular patient to the administered dosing regimen; performing a Bayesian update on each of the plurality of mathematical models as a function of the observed patient response data to generate a corresponding plurality of updated patient-specific mathematical models; performing Bayesian model averaging of the plurality of updated patient-specific mathematical models to generate an updated patient-specific composite model; providing, as an output from the system, the updated patient-specific composite model; A method comprising operating the system. (Item 37) 1. A method for providing a patient-specific dosing regimen for treatment with a drug using a computerized drug dosing regimen recommendation system, the system comprising a microprocessor, a memory, and at least one mathematical model, each mathematical model representing a response profile for a patient population treated with the drug, each model further representing a typical patient response as a function of covariate patient factors, the method comprising: processing the model to predict a typical patient response for each of a first plurality of proposed dosing regimens as a function of the model by performing a Bayesian prediction; selecting a recommended exemplary dosing regimen from the set of predicted typical patient responses that is suitable for a typical patient having said particular patient characteristics, said recommended exemplary dosing regimen being selected to achieve a predetermined target response; providing, as an output from the system, the recommended exemplary dosing regimen; receiving data reflecting an observed response of said particular patient to the administered dosing regimen; performing a Bayesian update on each mathematical model as a function of the observed patient response data to generate a corresponding updated patient-specific mathematical model; performing a Bayesian prediction to predict a patient-specific response for each of the second plurality of proposed dosing regimens as a function of each updated patient-specific model; selecting a recommended patient-specific dosing regimen suitable for said particular patient from a set of predicted patient-specific responses, said recommended patient-specific dosing regimen being selected to achieve said predetermined target response; providing, as an output from said system, said recommended patient-specific dosing regimen; A method comprising operating the system. (Item 38) 1. A method for providing a patient-specific dosing regimen for treatment with a drug using a computerized drug dosing regimen recommendation system, the system comprising a microprocessor, a memory, and at least one mathematical model, each mathematical model representing a response profile for a patient population treated with the drug, each model further representing a typical patient response as a function of covariate patient factors, the method comprising: receiving data reflecting an observed response of said particular patient to the administered dosing regimen; performing a Bayesian update on each mathematical model as a function of the observed patient response data to generate a corresponding updated patient-specific mathematical model; performing a Bayesian prediction to predict a patient-specific response for each of the second plurality of proposed dosing regimens as a function of each updated patient-specific model; selecting a recommended patient-specific dosing regimen suitable for said particular patient from a set of predicted patient-specific patient responses, said recommended patient-specific dosing regimen being selected to achieve said predetermined target response; providing, as an output from said system, said recommended patient-specific dosing regimen; A method comprising operating the system. (Item 39) 1. A method for providing a patient-specific dosing regimen for treatment with a drug using a computerized drug dosing regimen recommendation system, the system comprising a microprocessor, a memory, and at least one mathematical model, each mathematical model representing a response profile for a patient population treated with the drug, each model further representing a typical patient response as a function of covariate patient factors, the method comprising: processing the model to predict a typical patient response for each of a first plurality of proposed dosing regimens as a function of the model by performing a Bayesian prediction; selecting a recommended exemplary dosing regimen from the set of predicted patient-specific patient responses that is suitable for a typical patient having said particular patient characteristics, said recommended exemplary dosing regimen being selected to achieve a predetermined target response; providing, as an output from the system, the recommended exemplary dosing regimen; receiving data reflecting an observed response of said particular patient to the administered dosing regimen; performing a Bayesian update on each mathematical model as a function of the observed patient response data to generate a corresponding updated patient-specific mathematical model; providing the updated patient-specific mathematical model as an output from the system; A method comprising operating the system. (Item 40) 1. A system for providing a patient specific dosing regimen for treatment with a drug, the system comprising: A microprocessor; a memory operatively connected to said microprocessor; a plurality of mathematical models, each of the plurality of mathematical models representing a response profile for a patient population treated with the agent, each model further representing a typical patient response as a function of covariate patient factors; microprocessor-executable instructions stored in the memory for operating the system to perform the method of claim 1; A system comprising: (Item 41) 1. A system for providing a patient specific dosing regimen for treatment with a drug, the system comprising: A microprocessor; a memory operatively connected to said microprocessor; a plurality of mathematical models, each of the plurality of mathematical models representing a response profile for a patient population treated with the agent, each model further representing a typical patient response as a function of covariate patient factors; 35. Microprocessor-executable instructions stored in the memory for operating the system to perform the method of claim 34. A system comprising: (Item 42) 1. A system for providing a patient specific dosing regimen for treatment with a drug, the system comprising: A microprocessor; a memory operatively connected to said microprocessor; a plurality of mathematical models, each of the plurality of mathematical models representing a response profile for a patient population treated with the agent, each model further representing a typical patient response as a function of covariate patient factors; 36. Microprocessor-executable instructions stored in the memory for operating the system to perform the method of claim 35. A system comprising: (Item 43) 1. A system for providing a patient specific dosing regimen for treatment with a drug, the system comprising: A microprocessor; a memory operatively connected to said microprocessor; a plurality of mathematical models, each of the plurality of mathematical models representing a response profile for a patient population treated with the agent, each model further representing a typical patient response as a function of covariate patient factors; 37. Microprocessor-executable instructions stored in the memory for operating the system to perform the method of claim 36. A system comprising: (Item 44) 1. A system for providing a patient specific dosing regimen for treatment with a drug, the system comprising: A microprocessor; a memory operatively connected to said microprocessor; a plurality of mathematical models, each of the plurality of mathematical models representing a response profile for a patient population treated with the agent, each model further representing a typical patient response as a function of covariate patient factors; 38. Microprocessor-executable instructions stored in the memory for operating the system to perform the method of claim 37. A system comprising: (Item 45) 1. A system for providing a patient specific dosing regimen for treatment with a drug, the system comprising: A microprocessor; a memory operatively connected to said microprocessor; a plurality of mathematical models, each of the plurality of mathematical models representing a response profile for a patient population treated with the agent, each model further representing a typical patient response as a function of covariate patient factors; 39. Microprocessor-executable instructions stored in the memory for operating the system to perform the method of claim 38. A system comprising: (Item 46) 1. A system for providing a patient specific dosing regimen for treatment with a drug, the system comprising: A microprocessor; a memory operatively connected to said microprocessor; a plurality of mathematical models, each of the plurality of mathematical models representing a response profile for a patient population treated with the agent, each model further representing a typical patient response as a function of covariate patient factors; 39. Microprocessor-executable instructions stored in the memory for operating the system to perform the method of claim 39. A system comprising: [Brief description of the drawings]

[0021] The following description will be better understood with reference to the accompanying drawings, in which:

[0022] [Figure 1] FIG. 1 is a flow diagram illustrating an exemplary prior art method for creating a drug regimen for a patient.

[0023] [Diagram 2] FIG. 2 is a flow diagram illustrating an exemplary method for providing patient-specific drug dosing as a function of a mathematical model that is updated to account for observed patient responses, according to an exemplary embodiment of the present invention.

[0024] [Diagram 3] FIG. 3 is a schematic diagram of an exemplary system for providing patient-specific drug dosing as a function of a mathematical model that is updated to account for observed patient responses, in accordance with an exemplary embodiment of the present invention.

[0025] [Figure 4] FIG. 4 is a series of graphs showing the relationship between time since last dose (TSLD) and observed / predicted patient response for 16 exemplary patients.

[0026] [Figure 5A]5A-5D are exemplary graphs showing infliximab blood level concentrations as a function of time. [Figure 5B] 5A-5D are exemplary graphs showing infliximab blood level concentrations as a function of time. [Figure 5C] 5A-5D are exemplary graphs showing infliximab blood level concentrations as a function of time. [Figure 5D] 5A-5D are exemplary graphs showing infliximab blood level concentrations as a function of time.

[0027] [Figure 6] 6A-6D are exemplary graphs illustrating the iterative Bayesian updating process.

[0028] [Figure 7] 7A and 7B are exemplary graphs of blood level concentration as a function of time illustrating the effects of Bayesian model averaging and updating.

[0029] [Figure 8A] 8A-8D and 9A-9E are exemplary graphs showing infliximab blood level concentrations as a function of time. [Figure 8B] 8A-8D and 9A-9E are exemplary graphs showing infliximab blood level concentrations as a function of time. [Figure 8C] 8A-8D and 9A-9E are exemplary graphs showing infliximab blood level concentrations as a function of time. [Figure 8D] 8A-8D and 9A-9E are exemplary graphs showing infliximab blood level concentrations as a function of time. [Figure 9A] 8A-8D and 9A-9E are exemplary graphs showing infliximab blood level concentrations as a function of time. [Figure 9B]8A-8D and 9A-9E are exemplary graphs showing infliximab blood level concentrations as a function of time. [Figure 9C] 8A-8D and 9A-9E are exemplary graphs showing infliximab blood level concentrations as a function of time. [Figure 9D] 8A-8D and 9A-9E are exemplary graphs showing infliximab blood level concentrations as a function of time. [Figure 9E] 8A-8D and 9A-9E are exemplary graphs showing infliximab blood level concentrations as a function of time.

[0030] [Figure 10A] 10A-10F are exemplary graphs showing covariate typical and patient-specific predictions of infliximab blood level concentrations as a function of time illustrating the impact of various dosing intervals. [Figure 10B] 10A-10F are exemplary graphs showing covariate typical and patient-specific predictions of infliximab blood level concentrations as a function of time illustrating the impact of various dosing intervals. [Figure 10C] 10A-10F are exemplary graphs showing covariate typical and patient-specific predictions of infliximab blood level concentrations as a function of time illustrating the impact of various dosing intervals. [Figure 10D] 10A-10F are exemplary graphs showing covariate typical and patient-specific predictions of infliximab blood level concentrations as a function of time illustrating the impact of various dosing intervals. [Figure 10E] 10A-10F are exemplary graphs showing covariate typical and patient-specific predictions of infliximab blood level concentrations as a function of time illustrating the impact of various dosing intervals. [Figure 10F]10A-10F are exemplary graphs showing covariate typical and patient-specific predictions of infliximab blood level concentrations as a function of time illustrating the impact of various dosing intervals.

[0031] [Figure 11] FIG. 11 is a system diagram illustrating an exemplary networked computing environment in which the present invention may be employed. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0032] The present invention provides systems and methods for providing patient-specific drug dosing as a function of a mathematical model that is updated to account for observed patient responses, such as blood concentration levels, or measurements, such as blood pressure or hematocrit. More specifically, the present invention provides systems and methods for predicting, suggesting, and / or evaluating a suitable drug dosing regimen for a particular individual as a function of the individual-specific characteristics and the particular individual's observed response to the drug. Conceptually, the present invention provides the prescribing physician with access to a mathematical model of the observed patient response to the drug when prescribing the drug to a particular patient in a straightforward manner. In prescribing an initial dose, the present invention enables the use of a published mathematical model to predict the particular patient's response as a function of the patient-specific characteristics that are accounted for in the model as patient factor covariates. Thus, the prescribing physician can leverage the model in creating a rationally adjusted initial dose for a particular patient as a function of the particular patient's characteristics with much better accuracy than a PI can provide.

[0033] However, model-based predictions and resulting dosing regimens are somewhat limited in that they are based on the response of a hypothetical typical patient with the same patient characteristics. Thus, predictions and resulting dosing regimens are likely to be suboptimal due to the fact that published mathematical models cannot account for the uniqueness of patients, i.e., patients' responses are likely to be unique and not typical. To account for this uniqueness or inter-subject variability (BSV), the present invention allows the additional use of a specific patient's observed response to an initial dosing regimen to adjust the dosing regimen. Specifically, the systems and methods of the present invention use the patient's observed response in conjunction with published mathematical models and patient-specific characteristics to account for BSV that cannot be accounted for by mathematical models alone. Thus, the present invention allows the specific patient's observed response to be used to refine the model and associated predictions, to effectively personalize the model so that it can be used to more accurately predict the expected response to a proposed dosing regimen for a specific patient. By using observed response data to personalize the model, the model is modified to account for between-subject variability (BSV) that is not accounted for in traditional mathematical models, which represent only typical responses for a patient population, or "covariate-typical" responses for a typical patient with certain characteristics, which are accounted for as covariates in the model.

[0034] Conceptually, the present invention enables prescribing physicians to create personalized dosing regimens using one or more published mathematical models that reflect actual clinical data, without the loss of resolution of the data and / or models that result from distilling actual clinical data into a relatively crude set of recommendations for the "average" or "typical" patient, as is the case with PIs.

[0035] Such model-based generation of patient-specific drug dosing regimens eliminates or reduces the trial and error aspect of traditional dosing regimen generation, and further reduces the amount of time to generate a satisfactory or optimal dosing regimen, thus eliminating or reducing the associated waste of drugs, time, or other resources, as well as reducing the amount of time that a patient is at risk for an undesirable outcome.

[0036] In general, the systems and methods involve collecting mathematical models created from clinical data collected from patients administered a particular drug, processing the models to create composite models enriched with patient data, and determining patient-specific dosing regimens as a function of patient-specific observed response data that is processed in conjunction with data from the mathematical models. More specifically, the systems and methods use Bayesian averaging, Bayesian updating, and Bayesian prediction techniques to create patient-specific dosing regimens as a function of not only the general mathematical model and patient-specific characteristics accounted for in the model as covariate patient factors, but also the observed patient-specific responses that are not accounted for within the model itself and reflect BSVs that distinguish a particular patient from the typical patient reflected by the model.

[0037] In other words, the present invention provides a system and method that takes into account variability among individual patients that is not accounted for and / or explained by traditional mathematical models (e.g., patient responses that would not have been predicted based on dosing regimens and patient factors alone). Furthermore, the present invention allows patient factors accounted for by models, such as weight, age, race, laboratory test results, etc., to be treated as continuous functions rather than as categorical (cut-off) values. By doing so, the present system and method adapts a known model to a particular patient such that patient-specific predictions and analyses can be made to predict, suggest, and / or evaluate a dosing regimen that is personalized for the particular patient. Notably, the present system and method can be used not only to retrospectively assess dosing regimens previously administered to a patient, but also to proactively assess a proposed dosing regimen before administering the proposed regimen to the patient, or to identify a dosing regimen (dosage, dosing interval, and administration route) for a patient that will achieve a desired outcome.

[0038] Conceptually, observed patient-specific response data is effectively used as "feedback" to fit a generic model, representing typical patient responses, to a patient-specific model capable of accurately predicting patient-specific responses so that patient-specific dosing regimens can be predicted, proposed, and / or evaluated on a patient-specific basis.

[0039] Moreover, even before incorporating observed response data to fit a general model, the present invention allows for the identification of an initial dosing regimen that is better than would be suggested by the PI or any one mathematical model. More specifically, in one embodiment, the initial dosing regimen is based on a composite model that is comprised of multiple patient data-rich mathematical models. In such an embodiment, the initial dosing regimen is more closely matched to the needs of a particular patient than the general recommendations of a PI for the "average" patient.

[0040] By taking into account the complex mathematical model and refining the initial dosing regimen for a particular patient as a function of observed patient-specific data, a personalized patient-specific dosing regimen is created, and is created rapidly. It will be appreciated that the exemplary method is implemented and executed by the computerized model-based patient-specific drug regimen recommendation system 200 with input provided by a human operator, such as a physician, and thus acts as a recommendation engine and / or physician expert system, providing information for consideration by the prescribing physician.

[0041] An exemplary method of the present invention is discussed below with reference to FIG. 2. FIG. 2 shows a flow diagram 150 illustrating an exemplary method for providing patient-specific dosing based on observed response data for the particular patient being treated as well as a mathematical model representing responses for a patient population. Referring now to the flow diagram 150 of FIG. 2, the exemplary method begins with providing a mathematical model-based patient-specific drug regimen recommendation system 200, as shown at 152. The exemplary system 200 is shown in FIG. 3 and discussed in further detail below. By way of example, the system 200 may include conventional hardware and software representative of a general-purpose desktop, laptop / notebook, or tablet computer. Furthermore, in accordance with the present invention, the system 200 is further specially configured with hardware and / or software comprising microprocessor-executable instructions for specially configuring the system 200 to perform the methods described below.

[0042] Referring again to FIG. 2, the method then involves providing at least one mathematical model to the system, as indicated at 154. In one embodiment, the mathematical models are provided as part of a software module or library operated by a software program that is modular in nature and comprises microprocessor-executable instructions for specifically configuring the system 200 to perform the method described below. By way of example, the models may be pre-stored on the system, added to the system periodically, e.g., as part of distributed updates that are periodically stored on the system, or downloaded on demand from the Internet or another network, or other electronic media. Alternatively, the mathematical models may not be part of a software module or library, but rather may be hard-coded into or otherwise an integral part of a unitary software program. In either case, the models are provided to the system as input, as indicated at step 154 ​​in FIG. 2, and stored in the system's memory, e.g., in the mathematical model data store 218a in FIG. 3.

[0043] Any suitable mathematical model may be used. A suitable mathematical model is a mathematical function (or set of functions) that represents the relationship between the dosing regimen and the observed patient exposure and / or observed patient response (collectively "response") to a particular drug. Thus, the mathematical model represents the response profile for a patient population. In general, creating a mathematical model involves creating a mathematical function / equation that defines a curve that best "fits" or represents the observed clinical data, as understood by those skilled in the art.

[0044] A typical model also represents the expected effect of certain patient characteristics on response, and quantifies the amount of unexplained variability that cannot be explained by patient characteristics alone. In such models, patient characteristics are reflected as patient factor covariates in the mathematical model. Thus, mathematical models are typically mathematical functions that represent the underlying clinical data and the associated variability found in patient populations. These mathematical functions include terms that represent the variation of individual patients from the "average" or typical patient, allowing the model to represent or predict various outcomes for a given dose, making the model not only a mathematical function but also a statistical function, but the models and functions are generally and non-limitingly referred to herein as "mathematical" models and functions.

[0045] It will be appreciated that many suitable mathematical models already exist and are used for purposes such as drug development. Examples of suitable mathematical models that represent response profiles for patient populations and account for patient factor covariates include pharmacokinetic (PK) models, pharmacodynamic (PD) models, and exposure / response models, which are well known to those skilled in the art. Such mathematical models are typically published or otherwise available from drug manufacturers, peer-reviewed literature, and the FDA or other regulatory agencies. Alternatively, suitable mathematical models may be prepared by original research.

[0046] It should be noted that in a preferred embodiment, multiple mathematical models for a single drug, to the extent available, are provided as inputs to the system 200. Each mathematical model may be stored in the memory 218 of the system 200 as a mathematical function and / or in the form of a compiled library module.

[0047] Next, the method involves identifying a particular patient having patient-specific characteristics, as shown at step 156. This step may be performed by a physician or other human and may involve, for example, examining the patient and collecting and / or measuring patient-specific factors such as gender, age, weight, race, stage of disease, medical condition, previous treatments, other comorbidities, and / or demographic and / or laboratory test result information. More specifically, this involves identifying patient characteristics that are reflected as patient factor covariates in the mathematical model. For example, if a model is constructed to represent typical patient response as a function of weight and gender covariates, this step would include identifying the patient's weight and gender characteristics (e.g., 175 pounds, male). Any other characteristics that have been shown to predict response and thus be reflected as patient factor covariates in the mathematical model may be identified. By way of example, such patient factor covariates may include weight, gender, race, laboratory results, stage of disease, and other objective and subjective information.

[0048] Next, the method involves providing the patient-specific characteristics to the system 200 as input, as shown in step 158. By way of example, this may be done by a physician or other human operator of the system by providing input via a keyboard 208, mouse 210, or touch screen, barcode / scanner, or other interface device 212 of the system 200, as shown in FIG. 3. The patient-specific characteristics may be stored in the memory 218 of the system 200 in the form of a database record, for example in the patient factor data storage portion 218b of FIG. 3. Thus, the inputted patient characteristics can be used in conjunction with the model to identify a dosing regimen closely suited to the needs of a particular patient, as discussed below. More specifically, the inputted patient characteristics can be used in conjunction with the model to identify a dosing regimen that is "typical for the covariates," i.e., the model predicts that a dosing regimen is suitable for a typical patient (e.g., a typical 175 pound male patient) with a particular patient's covariate characteristics that is likely to provide a better dosing regimen than a typical PI. The use of such models to generate "covariate-representative" dosing regimens is discussed below.

[0049] In this example, multiple different mathematical models representing patient response to a single drug are provided as input to the system 200. Accordingly, the system then performs Bayesian model averaging to create a composite mathematical model as a function of the mathematical models provided to the system as input and the patient-specific characteristics accounted for by the models as patient factor covariates, as shown in step 160. More specifically, the multiple mathematical models for the single drug are then processed by the system 200 to create a composite mathematical model for the drug using Bayesian model averaging. The composite mathematical model may be stored in the memory 218 of the system 200, for example in RAM or in the composite model data store 218c, as shown in FIG.

[0050] In general, Bayesian model averaging provides a systematic method to analyze specification uncertainty and check the robustness of single patient results against alternative model specifications. Bayesian model averaging accounts for model uncertainty that may result from poor parameter precision, model misspecification, or other defects that result from using a single model. The standard practice of selecting a single model from a class of models and then inferring based on this model ignores model uncertainty, which may impair predictive performance and overestimate the strength of evidence. Bayesian model averaging allows the incorporation of model uncertainty into inference. The basic idea of ​​Bayesian model averaging is to infer based on weighted averages over a model space that includes several models. This approach accounts for model uncertainty in both prediction and parameter estimation. The resulting estimates incorporate model uncertainty and therefore may better reflect the true uncertainty in the estimates. When several mathematical models are published, the Bayesian averaging of these models produces a weighted estimate of the patient response that is weighted for the amount of data used to create the original individual models, the accuracy and performance of the models, and the number of models incorporated into the Bayesian averaging step, or alternatively, the weighting may be set by the user or incorporated into the system using a Markov Chain Monte Carlo (MCMC) approach. Thus, the creation of a composite mathematical model from multiple mathematical models for a single drug compensates for limitations that may exist in any single model and provides a more complete and robust mathematical model typically used in the preparation of PIs.

[0051] Bayesian model averaging techniques are well known in the art. The methods implemented for Bayesian model averaging vary depending on the type of data and models being averaged, but are most commonly averaged using the Markov Chain Monte Carlo Model Composition (MC3) method. A description of exemplary methods used for Bayesian model averaging can be found in Hoeting JA, Madigan D, Raftery AE, Volinsky CT. Bayesian Model Averaging: A Tutorial. Statistical Model Averaging, the entire disclosure of which is incorporated herein by reference. Available in Science 1999, 14(4)382-417.

[0052] Techniques for applying Bayesian averaging to mathematical models are well known in the art. As one example, Bayesian averaging as a mathematical technique has been used in political science (e.g., Bartels, Larry M. “Specification Uncertainty and Model Averaging.” American Journal of Political Science 41:641-674; 1997), traffic flow and accident rate assessment (see, for example, Demirhan H Hamurkaroglu C "An Application of Bayesian Model Averaging Approach to Traffic Accidents Data Over Hierarchical Log-Linear Models" Journal of Data Science 7:497-511; 2009), and in the context of toxicological assessment (see, e.g., Morales KH, Ibrahim JG, Chen CH Ryan LM "Bayesian Model Averaging With Applications to Benchmark Dose Estimation for Arsenic in Drinking Water" JASA 101(473):9-17, 2006).

[0053] As a further example, Bayesian model averaging is a method of using multiple mathematical models to make forecasts, as is well known in the field of weather forecasting, for example to predict the most likely path of a tropical cyclone as a function of the different paths predicted by the different models. See, e.g., Raftery AE, Gneiting T, Balabdaoui F, and Polakowski M “Using Bayesian Model Averaging to Calibrate Forecasts” See, “Ensembles” Mon. Wea. Rev., 133, 1155-1174; 2005.

[0054] According to the present invention, a Bayesian model averaging approach is used to predict patient response when multiple patient response models are available. An example involving Bayesian model averaging of multiple mathematical patient response models is shown in FIG. 7A. Referring now to FIG. 7A, the red (A) and green (B) dashed lines represent expected concentration-time profiles from two different mathematical models. The red dashed line (A) is derived from the first mathematical model as published by Xu Z, Mould DR, Hu C, Ford J, Keen M, Davis HM, Zhou H, “A Population-Based Pharmacokinetic Pooled Analysis of Infliximab in Pediatrics”. The green dashed line (B) is derived from a second mathematical model as published by Fasanmade AA, Adedokun OJ, Ford J, Hernandez D, Johanns J, Hu C, Davis HM, Zhou H, “Population pharmacokinetic analysis of infliximab in patients with ulcerative colitis.” Eur J Clin Pharmacol. 2009; 65(12):1211-28. In this example, the concentration-time profiles resulting from both models are utilized and a combined model (shown as the blue dashed line (C)) representing a weighted average is generated by Bayesian model averaging, as shown in FIG. 7A.

[0055] However, it should be noted that the methods and systems of the present invention do not require multiple models as illustrated in the example of Figure 2. In alternative embodiments, a single model may be used in place of the composite model, and the corresponding Bayesian model averaging step may be omitted.

[0056] Further, for clarity, it should be noted that one or more mathematical models may be provided as inputs to and stored within system 200 for each of multiple drugs, although in the example of FIG. 2, only a single drug is discussed for illustrative purposes.

[0057] Referring again to FIG. 2, the system 200 then performs Bayesian prediction to predict a typical patient response for each of the multiple proposed dosing regimens, as shown in step 162. This step involves using the Bayesian prediction techniques of the system to test dosing regimens for a particular patient as a function of the model and patient-specific characteristics, described as patient factor covariates in the composite mathematical model. The predicted patient responses may be stored in the system's memory 218, for example in the dosing regimen prediction data storage 218e. This prediction, based on the composite model, involves evaluating dosing regimens based on predicted responses for a typical patient with patient-specific characteristics, which may be referred to as "typical for covariates" values. Techniques for applying Bayesian prediction to mathematical models are well known in the art.

[0058] Generally, Bayesian prediction involves using mathematical model parameters to predict the likely response that a particular patient will exhibit with various dosing regimens. Notably, in this step, prediction allows for the determination of likely patient response to a proposed dosing regimen prior to the actual administration of the proposed dosing regimen. Thus, prediction can be used to test multiple different proposed dosing regimens (e.g., different dosages, dosing intervals, and / or administration routes) to determine how each dosing regimen is likely to affect the patient as predicted by patient-specific factors and / or data in the model / composite model.

[0059] In other words, this prediction step involves the use of published models to evaluate dosing regimens to the extent that the published models allow for the evaluation of dosing regimens that are limited to analysis for a typical patient with the patient factor covariates of the particular patient. The resulting dosing regimen (corresponding to a satisfactory or best predicted patient response) is likely to be more accurate than that which can be provided by the PI, but is not precisely personalized for the particular patient being treated, since it does not account for the unique characteristics of the patient being treated. For example, a particular 175 pound male patient may respond to a particular dosing regimen differently than a typical 175 pound male. Thus, the composite model is used to predict a dosing regimen that would be expected to be suitable and / or optimal for a typical patient with patient-specific characteristics.

[0060] More specifically, the system performs multiple predictions of patient response by referring to and / or processing a composite model to evaluate multiple proposed dosing regimens based on patient characteristics. The system may determine that each dosing regimen is sufficient or insufficient to meet a therapeutic objective or target profile. For example, the target profile may involve maintaining trough blood concentration levels above a therapeutic threshold. Furthermore, the system may compare predictions of patient response to various dosing regimens and generate a set of satisfactory or best dosing regimens to achieve the therapeutic objective or target profile.

[0061] In one embodiment, a plurality of proposed dosing regimens are provided to the system as inputs by a user, either manually and / or arbitrarily. For example, the user may provide typed inputs to suggest dosing regimens to be evaluated, and in response, the system will perform Bayesian predictions to predict the patient's response to the proposed dosing regimens. The user may provide other inputs to suggest alternative dosing regimens to be evaluated at a later time. For example, dosages, dosing intervals, and / or administration routes may be varied between the proposed dosing regimens to be evaluated. This is essentially a trial and error approach, but is sophisticated based on a continuous function mathematical model that accounts for patient factor covariates, allowing the physician to test the proposed dosing regimens against a composite model.

[0062] In another embodiment, the system automatically proposes and tests proposed dosing regimens and follows an automated search algorithm to optimize the dosing regimen within the bounds of published clinical experience. In a preferred embodiment, the optimization algorithm uses Bayesian prediction but systematically suggests doses, dosing intervals, and / or administration routes. Plans that achieve the desired goal are marked as feasible and stored by the system. The resulting output is a set of possible initial dosing regimens that are expected to achieve the desired clinical outcomes as predicted by the complex mathematical model and patient factor covariates.

[0063] Illustrative examples are provided below with reference to Figures 4A-4P. Referring now to Figures 4A-4P, patient responses for each of 16 different patients (ID:1-ID:16) are shown as a function of time since last dose (TSLD, hours). More specifically, the blue line (Y) is generated using the above-referenced published mathematical model for infliximab (Xu et al.). More specifically, each blue line (Y) is a plot of the average ("typical") patient response predicted using the mathematical model and patient-specific factors (such as body weight) provided as covariates within the published model. This blue line (Y) thus represents clinical data compiled for multiple patients, and is therefore a somewhat "average" or "typical" representation. Thus, in Figure 4, it can be seen that some individual patient responses (open circle symbols) are higher than the blue line (for ID:1 and ID:5), while other individual patient responses are lower than the blue line (for ID:3, ID:7, and ID:9).

[0064] Thus, the blue line ("typical of covariates") recommendation is likely to be better than dosing instructions in PIs, which often do not consider all of the patient factors, or important patient factors, or may be based on roughly meeting expected patient responses. Thus, the blue line (Y) does not necessarily provide an optimal dosing regimen for a particular patient, as it does not consider the variability in responses seen between patients with the same patient factors, sometimes referred to as intersubject variability or "BSV". Thus, two patients with the same patient factors may have different observed responses to a particular dose of a therapeutic agent as a result of variability not accounted for by individual patient factors in the published model.

[0065] According to the teachings herein, the system of the present invention will predict the patient response to the proposed dosing regimen (step 162 in FIG. 2B) based on the blue line (Y) shown in FIG. 4A. Specifically, for ID:3 in FIG. 4, the initial dosing regimen provided by the system based solely on patient factors (covariates) would be expected to produce a response that reflects the blue line.

[0066] Referring again to FIG. 4, the red line (Z) discussed below is created according to the teachings of the present invention to account not only for patient-specific factors that are covariates in the model, but also for observed patient-specific responses consistent with the present invention. More specifically, the red line (Z) is based on a mathematical model that has been Bayesian updated to reflect observed patient-specific responses. Thus, the red line (Z) is likely to represent the correct dosing for the particular patient being treated. In accordance with the present invention, as described in more detail below, the subsequent dosing schedule recommendation (using Bayesian prediction) for ID:3 will be based on a Bayesian updated composite model (e.g., solid blue line (D) in FIG. 7B or red line Z in FIG. 4) that accounts for observed patient data (e.g., open circles), as discussed below.

[0067] Thus, in the example of Figure 4, the blue (Y) and red (Z) lines predicted as patient response for each patient ID:1-ID:16 differ based on patient-specific factors that are covariates in the model and the observed patient response data.

[0068] After performing one or more Bayesian predictions to provide one or more predicted typical patient responses, a physician and / or the system may compare the various predictions so that an appropriate dosing regimen may be identified.

[0069] Referring again to FIG. 2, the exemplary method then involves the system determining a recommended exemplary dosing regimen as a function of the predicted response, as shown in step 164. For example, the recommended exemplary dosing regimen may be selected by the system 200 as one of a number of Bayesian predictions tested to achieve the treatment objective, based on feasibility, practicability, etc. For example, the recommended exemplary dosing regimen may be selected as one of several tested / predicted proposed dosing regimens identified as capable of maintaining exposure (i.e., drug concentration) above therapeutic levels. For example, in FIG. 10B-10F, for the "difficult patient," multiple dosing intervals were tested using Bayesian predictions. As can be seen from FIG. 10B-10F, the predictions indicate that dosing regimens including dosing intervals of 1, 2, and 4 weeks can maintain drug concentration at the target level (represented by the black dashed line).

[0070] In one exemplary method, the system 200 displays a list of recommended typical dosing regimens to the physician via its display device 214, or causes the list of recommended dosing regimens to be printed via an associated printer, or transmitted by electronic data transmission via link 219 to a mobile computing device of the physician, a pharmacy, a hospital, a clinic, a patient, or the like, computer system. For example, a selected subset of the test dosing regimens (e.g., top 3, top 10, etc.) may be output by the system to the user as recommended or suggested dosing regimens. It will be understood by those skilled in the art that the characteristics of the "best" dosing regimen will vary according to drug characteristics and / or treatment objectives. Thus, in the example of Figures 10B-10F, dosing regimens for "difficult patients" identified as those expected to maintain concentrations (black circles) above the target concentration (dashed line) (i.e., dosing intervals of 1, 2, and 4 weeks) will be displayed, printed, or transmitted as the recommended typical dosing regimens of step 164.

[0071] In this exemplary method, the physician may then view the exemplary recommended dosing regimens provided as output by the system and then determine an initial dosing regimen for administering to the patient. In doing so, the physician may select a regimen from a list or may modify the recommended dosing regimen according to the physician's judgment.

[0072] Various considerations may be taken into account by the physician and / or the system in determining the recommended and / or initial dosing regimen. For example, a primary consideration may be meeting a particular treatment objective, such as maintaining a minimum blood level concentration, maintaining a target blood pressure, etc. However, other considerations, such as ease of compliance, scheduling considerations, drug / treatment costs, etc. may also be taken into account. The system may include a utility function to take such other considerations into account when determining a recommended exemplary dosing regimen.

[0073] The physician then directly or indirectly administers the initial dosing regimen, as shown in step 166. Compared to the prior art technique discussed above with reference to FIG. 1, this initial dosing regimen is better personalized for the specific patient to whom it is administered, since it is based on the interpretation of the underlying composite mathematical and statistical model as a function of the patient's individual characteristics. In other words, the initial dosing regimen reflects the dosing regimen typical of the covariates determined by the model, and is therefore suitable for a typical patient with specific patient characteristics. Thus, the initial dosing regimen is not based only on a rough interpretation of the underlying model as reflected in the general PI dosing information. Moreover, it is based on the predicted outcome following the evaluation of various doses, dosing intervals, and administration routes, as a function of the patient characteristics captured as patient factor covariates, as part of the Bayesian prediction process, and in a single model, or in a composite model generated by Bayesian model averaging of multiple mathematical models.

[0074] After the initial period, the physician, in this exemplary method, follows up with the patient to evaluate the patient's response to the initial dosing regimen, as shown in steps 168 and 170, and determine, for example, whether a dose adjustment is warranted due to inadequate patient response. These steps may be performed conventionally, as discussed above with reference to Figure 1. Thus, for example, this may involve obtaining laboratory test results that reflect the patient's response to the administered dosing regimen, examining the patient, and / or asking how the patient is feeling.

[0075] In this exemplary method, if at step 170 it is determined that a dose adjustment is not warranted, then the dose adjustment is discontinued, as indicated at 170, 186, and 188, and the method ends.

[0076] However, if at step 170 it is determined that a dosage adjustment is warranted, then the patient responses resulting from the evaluation are provided as inputs to the system 200, as shown at 170 and 172. For example, such inputs may include entering quantitative and / or qualitative laboratory result test data and / or physician assessments into the system 200. In addition, patient characteristics that may have changed may also be evaluated and / or provided as inputs to the system. Specifically, this step involves entering updated observed measurement patient responses, and updated patient-specific characteristics obtained from a particular patient.

[0077] Optionally, the system may be configured such that if a key individualized model parameter is more than ±3 standard deviations away from a key typical parameter value, the system indicates that further treatment with a therapeutic agent in this patient may not be warranted because the particular patient's response suggests that the patient will not respond adequately to the proposed treatment.

[0078] Referring again to FIG. 2, the system 200 performs Bayesian updating on each of the underlying mathematical models to update each model as a function of input patient-specific characteristics (tracked as patient factor covariates in the model) based on input patient response data, as shown in step 174. Preferably, this step is performed using iterative Bayesian updating, and is performed immediately prior to the creation and / or administration of the next dosing regimen. This update using observed patient-specific data takes into account the observed response of a particular patient, thus updating the model to account for inter-subject variability (BSV) not accounted for by the model itself. Thus, the application of Bayesian updating allows the software to account for changing patient conditions or patient factors, thus implicitly correcting the dosing regimen predictions and recommendations.

[0079] Techniques for applying Bayesian updating to mathematical models are well known in the art. See, e.g., Duffull SB, Kirkpatrick CMJ, and Begg EJ Comparison of two Bayesian approaches to dose-individualization for once-daily aminoglycoside regimens Br J Clin Pharmacol. 1997; 43(2): 125-135.

[0080] In general, Bayesian updating involves Bayesian inference, a method in which Bayes' rule is used to update probability estimates of hypotheses as additional evidence is acquired. Bayesian updating is particularly important in dynamic analysis of data collected over time (continuously). The method as applied here uses a model that not only represents the time course of exposure and / or response, but also includes terms that represent the unexplained (random) variability of exposure and response. It involves applying a "prior" to form a basic hypothesis. A "prior distribution" is the distribution of a parameter before any data is observed. In our example, the prior distribution is a set of underlying mathematical models that represent the expected exposure and / or response after administration of a drug, without the influence of observed individual patient data. In our example, this would be the initial estimates generated for the initial dosing regimen. A "sample distribution" is the distribution of the observed data, conditional on its parameters. This is also referred to as the "likelihood", especially when considered as a function of the parameters and observed response data. The marginal likelihood (sometimes also referred to as evidence or "posterior") is the distribution of the observed data marginalized over the parameters. In our example, this would be the model after it has been updated with the input observed response data. Thus, Bayes' rule can be applied iteratively. That is, after inputting the observed response data, the resulting posterior probability can then be treated as the prior probability of the next observed response, and new posterior probabilities can be calculated from the new evidence. This procedure is referred to as "Bayesian updating." The result of Bayesian updating is a set of parameters, conditional on the observed data. The process involves sampling parameters from the prior distributions (which in our example are all of the underlying models) and calculating expected responses based on the underlying models. For each underlying model, the difference between the model expectation and the observed data is compared. This difference is referred to as the "objective function." The parameters are then adjusted based on the objective function, and the new parameters are tested against the observed data by comparing the difference between the new model expectation and the observed data.This process runs iteratively until the objective function is as low as possible (minimizing the objective function), suggesting that the parameters are best to represent the current data. Thus, all the underlying models undergo a Bayesian update. Once all the models have been updated, a Bayesian averaging is performed to generate a new composite model.

[0081] To ensure that a global minimum of the objective function is obtained, the method may involve the use of a random function to interject some variation in the process to ensure that the objective function surface is fully explored and that a true global minimum is achieved (rather than a local minimum, which would not reflect the best explanation of the observed patient response data). The use of random functions to ensure that models achieve a true global minimum is well known in the field of stochastic approximate expectation methods. However, the use of such random functions is believed to be novel in the context of Bayesian updating.

[0082] For illustrative purposes, an example involving iterative Bayesian updating is discussed below with reference to Figures 6A-6D for a simple example model with three parameters (CL, V, and σ). Referring now to Figures 6A-6D, the numbers below each figure represent the parameter likelihoods (li), which are the likelihoods of the selected model parameters (Θ i σ i ), the observed data (y ij ) is the probability of {li=P(y ij |Θ i σ i)}. In an iterative process, the system first tests prior (typical) parameter values ​​(CL1, V1, and σ1, FIG. 6A) and uses these parameter values ​​to calculate the function. The system then compares the calculated values ​​(curve) with the observed data (filled circles) and calculates the likelihood (e.g., as shown in FIG. A). The system then selects a different set of parameters from the specified prior distribution to test (CL2, V2, and σ2, FIG. 6B) and again calculates the function using the second set of parameters. If the match between the observed data (filled circles) and the predicted values ​​(curve) is better (e.g., the likelihood is greater), the software will select a third set of parameters that follows the same trend as that used to select the second set of parameters. Thus, if the second set of parameters was smaller than the first set of parameters, the third set of parameters will still be smaller (lower) in value. If the match between the observed data and the predicted function is poor (e.g., the likelihood is smaller), the system will automatically choose a different path, in this case choosing a new estimate that is greater than the second parameter but less than the first parameter. This process is repeated iteratively (CL4, V4, and σ4, FIG. 6D) until the system cannot further improve the match between the observed and predicted values. Note that not all parameters necessarily behave in the same way, and some parameter values ​​may be greater than the initial values ​​and some may be smaller.

[0083] Referring again to FIG. 2, the system 200 then performs Bayesian model averaging to create an updated patient-specific composite model from the updated underlying model, as shown in step 176. This Bayesian averaging process is similar to that described above with reference to step 160, but the Bayesian averaging in step 176 is performed on the underlying updated model that is updated based on the observed patient response. The Bayesian averaging in step 176 generates an updated composite model. As an example, the updated model may be stored by the system in memory 218, such as in update model data storage 218d. At this point, the updated patient-specific composite model parameters obtained reflect the observed response of the particular patient being treated, and therefore more accurately reflect the individual patient and incorporate the variation in response that is not explained by patient factors such as age and weight. Thus, the updated patient-specific composite model better represents the expected patient response to subsequent doses.

[0084] This is reflected in the examples in Figures 4 and 7. Referring again to Figure 4, it will be seen that after administering the initial dose to ID:3 and obtaining the observed response from patient ID:3, the individual patient response (open circle) was derived from the average expected response (blue line, Y). The model for ID:3 will then be updated (using Bayesian updating) based on the observed response for patient ID:3 and incorporated into the composite model (using Bayesian averaging). In Figure 7B, the results of Bayesian updating and Bayesian model averaging are shown. The green dashed line (E) is the Fasanmade model before Bayesian updating, the green solid line (F) is the Fasanmade model after Bayesian updating, the red dashed line (G) is the Xu model before Bayesian updating, the red solid line (H) is the Xu model after Bayesian updating, the blue dashed line (I) is the Bayesian average model before Bayesian updating, the blue solid line (D) is the Bayesian average model after Bayesian updating, and the open circles are the observed data after the initial dosing regimen is administered. As can be seen in these figures, the blue solid line (D, updated composite model) reflects the observed data (open circles). Thus, the one used for the next prediction of the dosing regimen according to the present invention is the updated composite model reflected by the blue solid line (D).

[0085] Referring again to FIG. 2, the system 200 then performs Bayesian prediction, as shown at 178, to predict a patient-specific response for each of a plurality of proposed dosing regimens. The prediction is made as a function of the patient characteristics and the patient-specific composite model, which reflects the observed patient-specific response data. This generally involves the same process as described above with reference to step 162, but with the updated composite model used in step 178. As an example, this may involve testing a plurality of proposed doses prior to administration of the proposed doses to the patient, and using Bayesian prediction to predict the patient response to each of the proposed doses. More specifically, such Bayesian prediction uses the updated composite model to predict the patient response to each proposed dosing regimen.

[0086] The system 200 then determines a recommended patient-specific dosing regimen as a function of the predicted patient response, as shown in step 180 of FIG. 2. Similar to the method described above with reference to step 164, the process of determining a recommended patient-specific dosing regimen may involve a manual Bayesian prediction process in which the system receives the proposed dosing regimen as input by a user and projects the associated expected response, or alternatively, an automated search algorithm that automatically selects and tests the proposed dosing regimen, taking into account commercially available products and optimizing the dosing regimen within the scope of published clinical experience by evaluating the dosage, dosing interval, and route of administration. For example, a list of suggested patient-specific dosing regimen recommendations may be selected by the system from a number of Bayesian predictions that have been tested to achieve a pre-specified clinical target (a specific target trough concentration in the examples of FIGS. 8 and 9). As a further example, a therapeutic objective may be predefined and stored by the system for each drug. For example, the therapeutic objective may be to maintain a blood level concentration above a pre-defined therapeutic threshold.

[0087] The physician then reviews the recommended patient-specific dosing regimen information obtained from system 200 and determines an adjusted patient-specific dosing regimen as a function of the predicted response. This may involve reviewing a comparison of multiple predicted responses of the system and selecting or modifying one of the recommended patient-specific dosing regimens as the adjusted dosing regimen.

[0088] The impact of Bayesian prediction as a tool to pre-assess dosing regimens is discussed below for illustrative purposes with reference to Figures 8A-8G. Figures 8A-8G show dosing regimen adjustments in patients with moderate disease (the "population patient" or "typical" patient on which the drug labeling is based) where all other relevant patient factors are average. In these figures, the shaded areas are the target concentrations and the black circles are the individual predicted concentrations. Thus, the observed data are fitted using a model (in this example, the Fasanmade et al. and Xu et al. models), the models are Bayesian updated and then averaged (using Bayesian averaging), and the resulting composite model is used to predict, for each proposed dosing regimen, the expected patient response trough for that patient (in this example, an important consideration for treatment efficacy). The crosses are predicted concentrations predicted based only on patient factors (typical predicted concentrations) (i.e., predicted by the model only as a function of model covariates, without Bayesian updating), the red triangles J are actual doses, and the light blue triangles K are predicted doses. 8A reflects the predicted trough concentration that would be achieved with the proposed nominal dose (5 mg / kg every 8 weeks), as predicted using Bayesian prediction in step 176. It can be seen that for this patient with moderate disease, the nominal concentration will not maintain concentrations above target.

[0089] Now referring to FIG. 8B, a proposed dosing regimen of 6 mg / kg given every 8 weeks is tested as part of step 176, which provides a concentration that is at a target level and may reflect a second trial that a physician would try on a patient if the patient failed to respond to the initial treatment.

[0090]

[0046] Referring now to Figure 8C, a proposed dosing regimen of 7 mg / kg given every 8 weeks is tested as part of step 176. As noted from Figure 8C, this proposed dosing regimen is predicted to provide a concentration that is appropriately higher than the target concentration and that, when administered to this patient, may reflect the dose to be tried for the third trial. Referring now to Figure 8D, a proposed dosing regimen of 5 mg / kg given every 7 weeks is tested and the predicted results are shown in Figure 8D to provide a sufficient concentration.

[0091] For the example of Figures 8A-8D, it may be noted that over a 56 week treatment interval, a 7 mg / kg dose given every 8 weeks (Figure 8C) would utilize a total dose of 49 mg / kg, while a 5 mg / kg dose given every 7 weeks would utilize a total dose of 40 mg / kg (Figure 8D), while providing satisfactory results (i.e., maintaining minimum concentrations), i.e., a 20% reduction in total dose reflecting additional cost savings for the patient and a higher margin of safety due to an overall reduction in drug exposure. Thus, step 178 may involve selection of the dosing regimen associated with Figure 8D.

[0092] Another illustrative example is described below with reference to Figures 5A-5D. Referring now to Figures 5A-5D, the black circles represent observed blood level concentrations as a result of laboratory tests or the like, the red line (W) is the target concentration, in this case as determined by peer-reviewed literature and current clinical practice, the solid line (X) is the individual predicted concentration using the updated composite model, as determined by the mathematical model as updated in step 174 and averaged in step 176, and the dashed line (V) is the typical predicted concentration (similar in concept to the blue line shown in Figure 4 and referenced in Figure 2 at step 160, calculated considering only covariate information without any updates). Figures 5A and 5B represent data from patients with moderate disease ("population" or "typical" patients), and Figures 5C and 5D represent data from patients with severe disease ("difficult patients").

[0093] In FIG. 5A, the solid and dashed lines X, V are superimposed because Bayesian updating has not yet occurred. In FIG. 5B, Bayesian updating has been completed and the updated composite model is based on the covariates and observed patient responses as discussed above. Thus, there is a difference between the typical concentrations for the covariates resulting from the composite model with only patient factor information and the patient-specific predicted concentrations resulting from the updated composite based on patient factor information and observed patient-specific patient response data (i.e., the model has undergone Bayesian updating in addition to Bayesian averaging).

[0094] Referring now to Figure 5C, the difference between observed values ​​(black circles) and typical predicted values ​​(solid line, X) is more pronounced with severe disease. In Figure 5D, the shaded area W is the target concentration, the black circles are the patient-specific predicted concentrations predicted by the Bayesian updated model, the crosses are the predicted concentrations based on patient factors alone (the typical predicted concentrations and therefore based only on the covariates in the model, not as a result of Bayesian updating), and the triangles are the actual doses.

[0095] In Figures 5C and 5D, it can be seen that the patient-specific predicted concentrations obtained immediately prior to dosing (i.e., the trough concentrations that are the metric for dosing in this example) are below the target concentration W that must be achieved for clinical response, as best seen in Figure 5D. Note that the typical ("mean") predicted concentrations for the covariates (crosses, based on predictions made without Bayesian updating) suggest that the concentrations resulting from the dose being evaluated should be appropriate in that the crosses are above the target levels (red band) and the typical concentrations are above the red line W, as shown in Figure 5D. However, following a Bayesian update of the model to update the model to account for patient-specific response data and inter-subject variability (black circles, Figure 5D), the model indicates that the dose being evaluated is not appropriate, as the blue dots are below the target concentrations (red band, W), and therefore the target response / treatment goal will not be met.

[0096] Additional examples of evaluating dosing regimens are discussed below with reference to Figures 9A-E. Referring now to Figures 9A-9C, the figures represent dose adjustments in patients with severe disease ("difficult patients" who deviate substantially from the typical "population patient" previously shown). In these figures, the shaded areas are target concentrations, the black circles are individual predicted concentrations (based on a Bayesian updated composite model), the crosses are predicted concentrations based on patient factors alone (typical predicted concentrations as reflected by a model accounting for patient factors, i.e., typical with covariates), the red triangles L are actual doses, and the light blue triangles M are predicted doses (i.e., Bayesian predicted outcomes based on a composite model accounting for patient factors and updated to reflect observed patient response / exposure data). Figure 9A reflects trough concentrations achieved with the indicated dose (5 mg / kg every 8 weeks). It is understood that for this patient with severe disease, because the labeled dosage does not maintain supra-target concentrations, a physician considering whether to prescribe this dosing regimen would likely reject it and attempt to find a dosing regimen predicted to provide a better outcome for the particular patient being treated. Thus, the traditional need for the patient to try this ineffective regimen is thus avoided.

[0097] In FIG. 9B, the predicted response is shown for a proposed dosing regimen of 10 mg / kg given every 8 weeks. As noted from FIG. 9B, this proposed dosing regimen predicts concentrations below the target level, but may reflect a second attempt that a physician would try on a patient with a prior art therapeutic approach if the patient fails to respond. In FIG. 9C, the predicted response is shown for a proposed dosing regimen of 20 mg / kg given every 8 weeks. This dosing regimen may reflect a third attempt that a physician would try to achieve a response in a patient with a prior art therapeutic approach. As shown in the figure, none of these dosing regimens utilizing doses given every 8 weeks would provide concentrations above the target level. As a result of the predictions provided by the present invention, these ineffective dosing regimens are avoided, and the associated adverse effects to the patient are avoided. In addition, patients may be immediately provided with an effective dosing regimen without the associated loss of time to test these dosing regimens on patients, which is typical of prior art therapeutic approaches.

[0098] 9D and 9E show predicted responses associated with more frequent dosing regimens. FIG. 9D is a predicted response for a proposed dosing regimen of 5 mg / kg given every 4 weeks (which is also a viable alternative dosing regimen for 5 mg / kg given every 8 weeks). From FIG. 9D, it can be seen that this dosing regimen does not result in concentrations above the target trough. Thus, this dosing regimen is not satisfactory and therefore not optimal. In FIG. 9E, a predicted response is shown for a proposed dosing regimen of 10 mg / kg given every 4 weeks. From FIG. 9E, it can be seen that this dosing regimen provides concentrations above the target level. Thus, the dosing regimen in FIG. 9E is satisfactory and would be identified by the system 200 as an acceptable dosing regimen during automated dosing regimen evaluation since the expected concentrations are at or above the target concentrations (shaded line).

[0099] 9A-9E, an appropriate dosing regimen would not be identified by a physician until at least three to four dosing regimens have been tested on a patient according to prior art treatment techniques. However, the present system allows a physician to test proposed dosing regimens prior to administration to a patient and identify dosing regimens that are likely to be successful before testing them on a patient.

[0100] After the system determines a recommended patient-specific dosing regimen in step 180 (which may be a list of recommended patient-specific dosing regimens), the physician may then view the recommended patient-specific dosing regimens provided as output by the system and determine an adjusted dosing regimen to administer to the patient. This may involve reviewing the system's comparison of multiple predicted responses and selecting a dosing regimen that is predicted to best meet the treatment objectives and that is also feasible and feasible. In doing so, the physician may select a dosing regimen from the list or may modify the recommended patient-specific dosing regimen according to the physician's judgment. As explained above, various considerations may be taken into account by the physician and / or the system in determining the recommended dosing regimen and / or the adjusted dosing regimen.

[0101] The physician then directly or indirectly implements the adjusted dosing regimen, as shown in step 182. For example, this may be done by amending the prescription to increase or decrease the amount of drug, to increase or decrease the dosing interval, to change the route of administration, etc., and may be done directly or indirectly by the physician, as discussed above.

[0102] In comparison to the prior art techniques discussed above with reference to FIG. 1, which is the initial dosing regimen discussed above with reference to step 168, this patient-specific adjusted dosing regimen is better personalized for the particular patient to whom it is administered, since it is based on interpretation of underlying complex mathematical and statistical models as a function of the patient's personal factors and the patient's own response to the initial treatment. Moreover, it is based on predicted outcomes following evaluation of various doses, dosing intervals, and routes of administration, as part of a Bayesian prediction process, and as a result of the updated model, accounts for observed patient-specific responses and inter-subject variability not accounted for in published models. Thus, the adjusted dosing regimen is highly personalized to the needs of the particular patient.

[0103] After a period of treatment using the adjusted patient-specific dosing regimen, the physician, in this exemplary method, follows up with the patient to evaluate the patient's response to the adjusted dosing regimen, as shown in steps 184 and 170, to determine, for example, whether a dose adjustment is warranted due to inadequate patient response. As discussed above, such evaluation may be performed by traditional observation and / or data collection. Thus, for example, this may involve obtaining laboratory test results that reflect the patient's response to the administered dosing regimen, examining the patient, and / or asking the patient how they are feeling.

[0104] If at step 170 it is determined that a dose adjustment is not warranted, for example because the patient is not responding satisfactorily, then the dose adjustment is discontinued and the exemplary method ends, as indicated at steps 186 and 188.

[0105] However, if step 170 determines that further dosage adjustment is warranted, e.g., because the patient response is inadequate or suboptimal, new patient response data may be entered into the system and the method may be repeated, as shown in steps 170-184.

[0106] Because patient conditions (disease stage or condition) and demographics may change over time, the system can be used to repeatedly update optimal dosing based on each patient's updated condition and factors, exposure and / or response behavior, so that throughout the course of treatment, the dose can be adjusted to an optimal level or treatment can be interrupted if sufficient exposure cannot be achieved. Furthermore, the system thus allows the physician to evaluate potential dosing strategies (increasing / decreasing the dose, shortening / extending the dosing interval, or both) before administering the next dose of drug to the patient. This approach allows the dosing regimen to be optimized (or nearly optimized) within one to two dosing regimen cycles, allowing for continuous monitoring and selection of the appropriate dosing regimen as the patient's condition changes during treatment. This approach is revolutionary because it allows the treating physician to provide an initial dosing regimen that is broadly adjusted for known influencing factors, subsequent dosing regimens are adjusted based on each patient's individual behavior, the dosing regimen can be rapidly optimized, allowing adequate drug coverage to be achieved more quickly than current practice, estimated doses can be tested in silico before administration to the patient, and, if necessary, the dosing regimen can be adjusted during treatment based on the patient's condition, avoiding over- and under-dosing.

[0107] FIG. 3 is a block diagram of an exemplary physician expert system (PES) 200 (logically shown as a single representative server for ease of illustration) in accordance with the present invention. The PES 200 includes conventional computer hardware that stores and / or executes specially configured computer software, configuring the hardware as a specific special-purpose machine having various specially configured functional subcomponents that collectively execute the method in accordance with the present invention. Thus, the PES 200 of FIG. 3 includes a general-purpose processor and a bus 204 that is employed to connect and enable communications between the processor 202 and the components of the PES 200 in accordance with known techniques. The PES 200 typically includes a user interface adapter 206 that connects the processor 202 to one or more interface devices, such as a keyboard 208, a mouse 210, and / or other interface devices 212, which may be any user interface device, such as a touch-sensitive screen, a digitized input pad, etc., via the bus 204. The bus 204 also connects a display device 214, such as an LCD screen or monitor, to the processor 202 via a display adapter 216. The bus 204 also connects the processor 202 to a memory 218, which may include a hard drive, a diskette drive, a tape drive, or the like.

[0108] The PES 200 may communicate with other computers or networks of computers, for example, via a communications channel, network card, or modem 219. The PES 200 may be associated with such other computers in a local area network (LAN) or wide area network (WAN), act as a server in a client / server arrangement with another computer, etc. Such configurations, as well as appropriate communications hardware and software, are known in the art.

[0109] The PES software is specially configured in accordance with the present invention. Thus, as shown in FIG. 3, the PES 200 includes computer-readable processor-executable instructions stored in memory for carrying out the methods described herein. Furthermore, the memory stores certain data, for example, in a database or other data store, which is logically shown in FIG. 3 for illustrative purposes, without regard to any particular embodiment in one or more hardware or software components. For example, FIG. 3 shows the storage in memory 218 of the PES 200 mathematical model data in mathematical model data store 218a as library modules, observed response and dosage information stored in patient factor data store 218b, composite data models stored in composite model data store 218c, updated models stored in updated model data store 218d, and dosage regimen prediction results stored in dosage regimen prediction data store 218e.

[0110] In an alternative embodiment, the methods and systems described herein are delivered as web services via a network computing model. In such an embodiment, the system may be implemented via a physician / user operable client device and a centralized server that performs much of the functionality described above. Such an embodiment is described below with reference to FIG. 11, which is a system diagram illustrating an exemplary network computing environment 10 in which the present invention may be employed. Referring now to FIG. 11, the network computing environment includes a PES server system 200a that is operatively connected to a plurality of client devices 20, 40, 60, 80 via a communications network 50, such as the Internet or a proprietary wireless cellular network. The PES server system 200a may include conventional hardware and software for a web / application server, but in accordance with the present invention is further configured to provide the processing functionality described above with reference to the PES system 200 and to interact with the client devices. By way of example, the client devices may be personal computers 20, mobile phones / smartphones 40, or tablet PCs, which may have substantially conventional hardware and software for communicating with the PES server system 200a via the communications network 50. For example, such a device may be configured to access a website or web interface maintained by the PES server system 200a such that a physician / user may operate the client device to provide input and / or receive output as described above and communicate with the PES server system 200a which performs the associated processing as described herein. In these embodiments, the client device may not require any special purpose software; rather, all special purpose software is incorporated into the PES server system 200a and the client device is used solely to communicate with the PES server system 200a of the present invention.

[0111] Alternatively, the client device may be a smartphone, tablet PC, or other computing device 80 configured with specially configured native software applications running on the client device 80 and communicating with the PES server system 200a. In such an embodiment, some or all of the structure and / or processing described above with reference to the PES system 200 may be provided on the client computing device 80, which may be operated by a user / physician and which may communicate with the PES server system 200a to provide the functionality described herein.

[0112] Thus, while several specific embodiments of the invention have been described, various changes, modifications, and improvements will readily occur to those skilled in the art. Such configurations, modifications, and improvements as made apparent by this disclosure, although not expressly described herein, are intended to be part of this description and are intended to be within the spirit and scope of the invention. Accordingly, the foregoing description is illustrative only and not limiting. The present invention is limited only as defined in the following claims and their equivalents.

Claims

1. 1. A drug administration regimen recommendation system for providing a personalized therapeutic drug administration regimen to a subject patient, comprising: The system includes a processor coupled to a memory; the memory comprises a database; The processor, (i) a first model stored in the memory, the first model configured to describe a response of a plurality of patients in a population to a therapeutic agent, the therapeutic agent being administered to the plurality of patients based on a time course of exposure of the therapeutic agent in blood of the plurality of patients in the population; and (ii) first data stored in the memory, the first data indicative of an amount of exposure of the therapeutic agent in the subject patient's blood following ingestion of a dose of the therapeutic agent, the amount of exposure of the therapeutic agent being determined from a blood sample taken from the subject patient; and (iii) second data stored in the memory, the second data indicative of a first plurality of characteristics specifically associated with the subject patient, the first plurality of characteristics being shared by a plurality of patients in the population of the first model, the first plurality of characteristics specifically associated with the subject patient including at least one of a body size indicator, a gender, an ethnicity, a laboratory result, a stage of a disease, a disease state, a previous treatment, a concomitantly administered therapeutic agent, a concomitant disease, and a demographic information of the subject patient; (iv) third data stored in the memory, the third data indicating a target exposure threshold amount of the therapeutic agent in the blood of the subject patient; and configured to receive The processor, (i) preparing a patient-specific model by performing a Bayesian update on the first model based on the first data and the second data, the patient-specific model providing a predicted time course of an exposure profile of the therapeutic agent in the blood of the subject patient taking into account a response of the subject patient to an administered dose of the therapeutic agent; (ii) providing one or more suggested dosing regimens to achieve the amount of the therapeutic agent in the blood of the subject patient of the target exposure threshold, the one or more suggested dosing regimens being provided based on the patient-specific model and the second data by selecting a recommended dosing regimen suitable for the subject patient having the predicted time course of the exposure profile and the first plurality of characteristics from the second data, the one or more suggested dosing regimens being stored in the database; (iii) the drug administration regimen recommendation system outputs at least one of the one or more suggested administration regimens stored in the database to the subject patient; and The drug administration regimen recommendation system, further configured to:

2. The system of claim 1 , wherein the time course of exposure of the therapeutic agent in the subject patient's blood is the concentration of the therapeutic agent in the subject patient's blood.

3. The system of claim 1 , wherein parameters of the patient-specific model are stored in the memory.

4. The system of claim 3 , wherein the processor is configured to output a subset of the one or more suggested dosing regimens upon updating a dosing regimen.

5. 4. The system of claim 3, wherein the therapeutic agent is administered to the subject patient on a first day, and the processor is further configured to receive fourth data stored in the memory, the fourth data indicating a concentration of the therapeutic agent in the subject patient's blood in a sample taken from the subject patient after administration of the therapeutic agent.

6. The system of claim 5 , wherein at least one sample providing the fourth data is taken from the subject patient's blood on the first day.

7. The processor, receiving an input signal indicative of the amount of the target exposure threshold of the therapeutic agent in the blood of the subject patient; periodically updating the patient-specific model stored in the memory based on exposure of the therapeutic agent in one or more additional patient blood samples until the target exposure threshold amount of the therapeutic agent in the subject patient's blood is achieved. The system of claim 5 configured to:

8. 8. The system of claim 7, wherein the processor is configured to determine a revised dose based on whether the fourth data exceeds a predicted exposure level of the therapeutic agent in the subject patient's blood.

9. 8. The system of claim 7, wherein the processor is configured to adjust the dosing regimen after updating the patient-specific model by one or more of increasing a dose and shortening a dose interval.

10. 8. The system of claim 7, wherein the processor is configured to adjust the dosing regimen after updating the patient-specific model by one or more of decreasing a dose and increasing a dose interval.

11. The system of claim 10 , wherein the first model and the patient-specific model comprise a PK model.

12. The system of claim 11 , wherein the first model is a composite model prepared from a plurality of mathematical models derived from data obtained from the plurality of patients in the population.

13. The system of claim 12 , wherein the processor is further configured to prepare the composite model using Bayesian model averaging.

14. The system of claim 13 , wherein the processor is further configured to perform the Bayesian updating for each of the plurality of mathematical models.

15. 15. The system of claim 14, wherein the processor is further configured to perform a Bayesian analysis by performing a Bayesian prediction to predict a typical patient response to each of the one or more proposed dosing regimens based on the patient-specific model and the second data.

16. 8. The system of claim 7, wherein the processor is further configured to calculate an updated patient-specific dosing regimen of the therapeutic agent by selecting the dosing regimen that corresponds to a minimum of an objective function.

17. 8. The system of claim 7, wherein the one or more proposed dosing regimens include an interval between therapeutic drug administrations that is less than four weeks.

18. 8. The system of claim 7, wherein the one or more proposed dosing regimens include an 8 week interval between therapeutic drug administrations.

19. 20. The system of claim 18, wherein the therapeutic agent administered to the subject patient is infliximab, and the administered dose of infliximab is determined by a Bayesian composite model.

20. 20. The method of claim 19, wherein the administered dose of infliximab is between 5 mg / kg and 7 mg / kg and has an 8 week therapeutic administration interval.

Images