Systems and methods for designing clinical trials

Abstract

To provide methods and systems for enrolling patient candidates in a clinical trial.SOLUTION: The method includes the steps of: generating first prediction data indicating predicted progression of a condition for a first group of patients that participated in a first clinical trial using a predictive model and clinical data associated with the first group of patients; grouping clinical trial data into subsets based on the first prediction data; analyzing each subset of clinical trial data to generate a measure of efficacy of the treatment; establishing screening criteria for a second clinical trial by identifying at least one subset that has a measure of efficacy that is higher than a measure of efficacy of the treatment for the full first group of patients; receiving clinical data of a candidate for the second clinical trial; generating second prediction data for the candidate; and enrolling the candidate in the second clinical trial when the second prediction data satisfies the screening criteria.SELECTED DRAWING: Figure 1A

Description

[Technical Field]

[0001] This application claims the benefit of U.S. Non-provisional Application No. 16 / 149,954, filed October 2, 2018, the entire contents of which are incorporated herein by reference.

[0002] This application relates generally to the design and conduct of human clinical trials. [Background technology]

[0003] Human clinical trials for diseases with highly heterogeneous disease progression are exceptionally challenging. Examples of such diseases include amyotrophic lateral sclerosis (ALS), Alzheimer's disease, Huntington's disease, Parkinson's disease, cancer, many chronic diseases (such as diabetes and hepatitis C), and psychiatric disorders such as depression and PTSD. ALS is a classic example. The two most famous ALS patients are Lou Gehrig and Dr. Stephen Hawking. While Gehrig died approximately two years after his diagnosis, Dr. Stephen Hawking lived with ALS for over 50 years. Dr. Stephen Hawking's longevity is not due to superior ALS treatments, but simply to the different ways the disease manifests itself. Some patients progress rapidly to near-fatal death, while others progress very slowly over decades. Many individuals progress at varying rates over time, alternating between fast, average, and slow rates. Summary of the Invention [Problem to be solved by the invention]

[0004] This heterogeneity in disease progression poses extraordinary challenges for drug discovery and development companies seeking to conduct human clinical trials. For example, human clinical trials may unwittingly enroll patients with a wide range of disease progression, including many slow-progressing patients (like Dr. Hawking) and many fast-progressing patients (like Gehrig). Such trials can be expected to have a lot of variability or "noise" such that even small positive signals among treated patients cannot be detected above the noise. This can lead to the conclusion that a drug does not work, or does not work sufficiently well, when in fact the opposite is true. Such false negatives can halt drug development programs, resulting in the loss of tens or hundreds of millions of dollars already invested. Even more seriously, they can lead to the abandonment of drugs that have the potential to alleviate human suffering. [Means for solving the problem]

[0005] According to some embodiments, methods and systems screen candidates for enrollment in clinical trials based on data from previously conducted clinical trials. A predictive model is used to generate a disease progression prediction for patients who participated in the previous clinical trial. Subgroups of patients defined by the disease progression prediction are identified as exhibiting improved treatment outcomes. Candidates for enrollment in the clinical trial are screened by using the predictive model to generate a disease progression prediction for the candidate and comparing the candidate's prediction with the prediction for the patient subgroup. Candidates are enrolled, denied enrollment, or selected for various clinical trial subgroups based on the screening.

[0006] According to some embodiments, a method of enrolling a patient candidate in a clinical trial for a treatment comprises: generating first predictive data indicating a prediction of progression of a condition for a first group of patients participating in a first clinical trial using a predictive model and clinical data associated with the first group of patients, wherein the predictive model is constructed based on clinical data associated with a second group of patients having the condition; grouping clinical trial data related to the first group of patients into a plurality of subsets of clinical trial data based on the first prediction data, each subset of clinical trial data being associated with a corresponding subgroup of the first group of patients; analyzing each subset of clinical trial data to generate a measure of efficacy of treatment for a subgroup of patients corresponding to each subset; establishing at least one screening criterion for a second clinical trial of the treatment by identifying at least one subset of the plurality of subsets having a measure of efficacy greater than the measure of efficacy of the treatment for the first group of patients as a whole, wherein the at least one screening criterion is based on a range of predictive data values ​​associated with the identified subset; receiving clinical data relating to the candidate for the second clinical trial for the treatment; generating second predictive data indicative of a predicted progression of a condition of the candidate using the predictive model and the clinical data associated with the candidate; determining whether the second prediction data meets at least one screening criterion; enrolling the candidate in the second clinical trial in accordance with a determination that the second predictive data meets the at least one screening criterion; and rejecting the candidate for the second clinical trial in accordance with a determination that the second predictive data does not satisfy the at least one screening criterion. Includes.

[0007] In any of these embodiments, the method includes selecting a predictive model from a plurality of different predictive models based on a goal of treatment.

[0008] In any of these embodiments, the method includes building a predictive model by training a learning machine on clinical data of a second group of patients. In any of these embodiments, the method includes standardizing the clinical data of the second group of patients before training the learning machine.

[0009] In any of these embodiments, the predictive model may predict at least one condition progression indicator, and the clinical data for the second group of patients includes data for at least one condition progression indicator.

[0010] In any of these embodiments, the clinical data relating to the first group of patients may be clinical data generated prior to the first trial.

[0011] In any of these embodiments, the number of patients associated with one subset may be equal to the number of patients associated with at least one other subset.

[0012] In any of these embodiments, the number of patients associated with one subset may differ from the number of patients associated with at least one other subset.

[0013] In any of these embodiments, the predictive model may predict at least one condition progression indicator, and the measure of treatment efficacy is based on the at least one condition progression indicator.

[0014] In any of these embodiments, a measure of the effectiveness of treatment for the first group of patients as a whole may show that there is no statistically significant difference in treatment outcome between the control group and the treatment group.

[0015] In any of these embodiments, the measure of efficacy may be generated by comparing the treatment results of control patients with the treatment results of treated patients.

[0016] In any of these embodiments, identifying a subset of the plurality of subsets may include identifying a subset having a measure of efficacy that satisfies a threshold test, which may be a p-value not exceeding 0.05.

[0017] In any of these embodiments, identifying at least one subset of the plurality of subsets may include plotting a representation of a measure of efficacy as a function of subgroup size and predictive data value. In any of these embodiments, plotting the representation of the measure of efficacy may include generating a heat map, where a change in appearance indicates a change in the measure of efficacy. In any of these embodiments, the heat map may include a representation of a measure of treatment efficacy for the entire group of patients. In any of these embodiments, identifying at least one subset of the plurality of subsets may include identifying a location on the heat map that is surrounded by locations having the same appearance.

[0018] In any of these embodiments, at least one screening criterion may include a maximum value of a range of expected data values ​​and a minimum value of a range of expected data values.

[0019] In any of these embodiments, the method may include receiving the candidate's clinical data from a clinician via a clinical trial candidate portal, and sending instructions to the clinician to enroll or decline the candidate in the second clinical trial.

[0020] In any of these embodiments, the range of expected data values ​​may include an upper threshold expected data value and a lower threshold expected data value.

[0021] In any of these embodiments, the first predictive data may include predicted patient regression or predicted time to an event.

[0022] In any of these embodiments, the patient may be a human and the condition may be a disease. In any of these embodiments, the disease may be a neurodegenerative disease or cancer.

[0023] In any of these embodiments, the first clinical trial may be configured to determine at least one of the effectiveness of the agent, the effectiveness of different dosages of the agent, or the effectiveness of different combinations of agents.

[0024] In any of these embodiments, the method may include treating the registered candidate.

[0025] According to some embodiments, there is provided a system for enrolling candidates in clinical trials for a treatment, the system comprising one or more processors, a memory, and one or more programs stored in the memory and executable by the one or more processors; The one or more programs stored in the memory by the one or more processors, generating first predictive data indicating a prediction of progression of a condition for a first group of patients participating in a first clinical trial using a predictive model and clinical data associated with the first group of patients, wherein the predictive model is constructed based on clinical data associated with a second group of patients having the condition; grouping clinical trial data related to the first group of patients into a plurality of subsets of clinical trial data based on the first prediction data, each subset of clinical trial data being associated with a corresponding subgroup of the first group of patients; analyzing each subset of clinical trial data to generate a measure of efficacy of treatment for a subgroup of patients corresponding to each subset; establishing at least one screening criterion for a second clinical trial of the treatment by identifying at least one subset of the plurality of subsets having a measure of efficacy greater than the measure of efficacy of the treatment for the first group of patients as a whole, wherein the at least one screening criterion is based on a range of predictive data values ​​associated with the identified subset; receiving clinical data relating to the candidate for the second clinical trial for the treatment; generating second predictive data indicative of a predicted progression of a condition of the candidate using the predictive model and the clinical data associated with the candidate; determining whether the second prediction data meets at least one screening criterion; sending a notification to enroll the candidate in the second clinical trial in accordance with a determination that the second predictive data meets the at least one screening criterion; and sending a notification rejecting the candidate for the second clinical trial in accordance with a determination that the second predictive data does not meet the at least one screening criterion; This is what is carried out.

[0026] In any of these embodiments, the one or more programs stored in the memory by the one or more processors: selecting a predictive model from a plurality of different predictive models based on a goal of treatment; It is possible to do this.

[0027] In any of these embodiments, the one or more programs stored in the memory by the one or more processors: building a predictive model by training a learning machine on clinical data of a second group of patients; In any of these embodiments, the one or more programs stored in the memory may be executed by the one or more processors: standardizing clinical data of the second group of patients before training the learning machine; It is possible to do this.

[0028] In any of these embodiments, the predictive model may predict at least one condition progression indicator, and the clinical data for the second group of patients includes data for at least one condition progression indicator.

[0029] In any of these embodiments, the clinical data relating to the first group of patients may be clinical data generated prior to the first trial.

[0030] In any of these embodiments, the number of patients associated with one subset may be equal to the number of patients associated with at least one other subset.

[0031] In any of these embodiments, the number of patients associated with one subset may differ from the number of patients associated with at least one other subset.

[0032] In any of these embodiments, the predictive model may predict at least one condition progression indicator, and the measure of treatment efficacy is based on the at least one condition progression indicator.

[0033] In any of these embodiments, a measure of the effectiveness of treatment for the first group of patients as a whole may show that there is no statistically significant difference in treatment outcome between the control group and the treatment group.

[0034] In any of these embodiments, the measure of efficacy may be generated by comparing the treatment results of control patients with the treatment results of treated patients.

[0035] In any of these embodiments, identifying a subset of the plurality of subsets may include identifying a subset having a measure of efficacy that satisfies a threshold test, which may be a p-value not exceeding 0.05.

[0036] In any of these embodiments, identifying at least one subset of the plurality of subsets may include plotting a representation of a measure of efficacy as a function of subgroup size and predictive data value. In any of these embodiments, plotting the representation of the measure of efficacy may include generating a heat map, where a change in appearance indicates a change in the measure of efficacy. In any of these embodiments, the heat map may include a representation of a measure of treatment efficacy for the entire group of patients. In any of these embodiments, identifying at least one subset of the plurality of subsets may include identifying a location on the heat map that is surrounded by locations having the same appearance.

[0037] In any of these embodiments, at least one screening criterion may include a maximum value of a range of expected data values ​​and a minimum value of a range of expected data values.

[0038] In any of these embodiments, candidate clinical data may be received from a clinician via a clinical trial candidate portal. In any of these embodiments, the one or more programs stored in the memory by the one or more processors: sending instructions to the clinician to enroll or decline the candidate in the second clinical trial; It is possible to do this.

[0039] In any of these embodiments, the range of expected data values ​​may include an upper threshold expected data value and a lower threshold expected data value.

[0040] In any of these embodiments, the first predictive data may include predicted patient regression or predicted time to an event.

[0041] In any of these embodiments, the patient may be a human and the condition may be a disease. In any of these embodiments, the disease may be a neurodegenerative disease or cancer.

[0042] In any of these embodiments, the first clinical trial may be configured to determine at least one of the effectiveness of the agent, the effectiveness of different dosages of the agent, or the effectiveness of different combinations of agents. [Brief explanation of the drawings]

[0043] The patent or application file contains at least one drawing executed in color. Copies of this patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee. The invention will now be described, by way of example only, with reference to the accompanying drawings, in which: [Figure 1A] 1A and 1B are flow diagrams of methods for enrolling candidates in clinical trials based on patient-level predictions, according to some embodiments. [Figure 1B] 1A and 1B are flow diagrams of methods for enrolling candidates in clinical trials based on patient-level predictions, according to some embodiments. [Figure 2] FIG. 2 is an exemplary list of predictive models for generating disease progression metrics for ALS patients, according to some embodiments. [Figure 3A] 3A-3D are plots for identifying patient subgroups associated with improved treatment outcomes, according to one embodiment. [Figure 3B] 3A-3D are plots for identifying patient subgroups associated with improved treatment outcomes, according to one embodiment. [Figure 3C]3A-3D are plots for identifying patient subgroups associated with improved treatment outcomes, according to one embodiment. [Figure 3D] 3A-3D are plots for identifying patient subgroups associated with improved treatment outcomes, according to one embodiment. [Figure 4A] 4A and 4B are plots for identifying patient subgroups associated with improved therapeutic outcomes, according to another embodiment. [Figure 4B] 4A and 4B are plots for identifying patient subgroups associated with improved therapeutic outcomes, according to another embodiment. [Figure 5] FIG. 5 is a functional block diagram of a system for enrolling candidates in clinical trials based on patient-level predictions, according to some embodiments. [Figure 6] FIG. 6 illustrates an exemplary user interface that a user can use to screen candidates and enroll them in a clinical trial, according to some embodiments. [Figure 7] FIG. 7 illustrates a computing device according to some embodiments. DETAILED DESCRIPTION OF THE INVENTION

[0044] Described below are systems and methods for using patient-level predictive algorithms to rescue or improve previously clinically tested treatment programs. According to some embodiments, the systems and methods utilize a database of historical clinical data collected from individual patients with a disease or other condition to build a predictive model that can be used to predict future disease progression for other patients. The predictive model is used to predict disease progression for patients using clinical data obtained from patients who participated in previously conducted clinical trials before they received a therapeutic drug or placebo. Based on the predicted results, patient subgroups are defined. For each subgroup, the clinical trial outcome data for the observed treatment group is compared with the clinical trial outcome data for the observed control group to determine whether there is an observable treatment effect in that subgroup. In other words, subgroups are defined and analyzed to determine whether there are subgroups of patients defined by predicted disease progression in which the treatment effect is detectable (i.e., statistically significant). In some embodiments, the subgroups are defined and analyzed to identify subgroups in which the combination of treatment effect and variance results in a statistically significant effect size.

[0045] The predictive data for the identified subgroup or subgroups are used to screen candidates for new clinical trials. Patients are screened at the start of the clinical trial, for example, to predict a mortality risk score for each candidate patient, and only patients whose risk score falls within a predetermined range are allowed to participate in the clinical trial. In another variation, all patients may participate in the clinical trial, but only patients who meet predefined predictive criteria are included in the primary analysis of the clinical trial.

[0046] According to some embodiments, previously conducted clinical trials are "failed" clinical trials, meaning that they failed to detect a therapeutic effect or were not strong enough to justify the continuation of a therapeutic development program. However, previously conducted clinical trials do not have to be failed clinical trials. Non-registration studies can serve as the basis for identifying subgroups to analyze in pivotal trials. For example, an "all comers" study can be conducted to identify subgroups that significantly respond to a treatment, and the results can be used to design clinical trials targeting patients with the same characteristics as the identified subgroups. Initial trials can be either placebo-controlled or uncontrolled. Uncontrolled trials generate predictions for each participant that assume no therapeutic effect, which can be used as virtual controls to identify patient subgroups for designing future clinical trials.

[0047] For purposes of this disclosure, the terms "clinical trial" and "patient" refer broadly to any study of a subject through experimentation. In some embodiments, the subject is a human, and the clinical trial is conducted to test the effectiveness of a treatment for a condition (e.g., a disease) afflicting the human subject. However, the subject may also be an animal or cells in tissue culture, and the clinical trial may refer to the testing of a treatment on the animal or cells, or more broadly, the study of animals or cells. This includes evaluation of veterinary research (e.g., determining whether a drug is effective in dogs) and testing new therapies in animals or cells intended for human use (e.g., mouse models for drug testing).

[0048] The scope of the research extends beyond pharmaceutical treatments and can apply to any intervention, including but not limited to the use of medical devices, biological therapies, exercise and physical therapy, and psychological counseling.

[0049] In the following description of the present disclosure and embodiments, reference is made to the accompanying drawings, in which is shown, by way of illustration, specific embodiments which may be practiced. It is to be understood that other embodiments and examples may be practiced and changes may be made without departing from the scope of the present disclosure.

[0050] It should further be understood that, as used in the following description, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that, as used herein, the term "and / or" means and encompasses one or more possible combinations of the associated listed items. It should also be understood that, as used herein, the terms "includes," "including," "comprises," and / or "comprising" specify the presence of stated features, integers, steps, operations, elements, components, and / or units, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, units, and / or groups thereof.

[0051] Certain aspects of the present invention include process steps and instructions described herein in the form of an algorithm. It should be noted that the process steps and instructions of the present invention may be embodied in software, firmware, or hardware, and, if embodied in software, may be downloaded to reside on and be operated from different platforms for use by various operating systems. As will become apparent from the discussion that follows, unless otherwise stated, throughout this specification, descriptions utilizing terms such as "processing," "computing," "calculating," "determining," "displaying," and the like will be understood to refer to the operations and processes of a computer system, or similar electronic computing device, that manipulate and transform data represented as physical (electronic) quantities within the computer system's memory or registers or other such information storage, transmission, or display devices.

[0052] The present invention also relates to devices for performing the operations herein. The devices may be specially constructed for the required purposes, or they may comprise general-purpose computers selectively activated or reconfigured by a computer program stored in the computer. Such computer programs may be stored on non-transitory computer-readable storage media, such as floppy disks, USB flash drives, external hard drives, optical disks, CD-ROMs, magneto-optical disks, read-only memory (ROM), random access memory (RAM), EPROMs, EEPROMs, magnetic or optical cards, application-specific integrated circuits (ASICs), or any other type of disk or medium suitable for storing electronic instructions, each coupled to a computer system bus. Furthermore, the computers referred to herein may include a single processor or may be architectures employing multiple processor designs to increase computing power.

[0053] The methods, apparatus, and systems described herein are not inherently related to any particular computer or other apparatus. Various general-purpose systems may be used with programs in accordance with the teachings herein, or it may prove convenient to construct more specialized apparatus to perform the required method steps. The required structure for a variety of such systems will appear from the description below. Further, the present invention is not described with reference to any particular programming language. It will be understood that a variety of programming languages ​​may be used to implement the teachings of the present invention as described herein.

[0054] 1 illustrates a method 100 for enrolling candidates in a new clinical trial based on data from a previous clinical trial. Both the previous clinical trial and the new clinical trial may be to determine the effectiveness of a treatment for a condition, such as a new drug for treating a disease. In some embodiments, the previous clinical trial is a failed trial, in the sense that there was no statistically significant difference in results between the treatment group and the control (placebo) group. However, as described below, this need not necessarily be the case.

[0055] Method 100 identifies one or more subgroups of patients who show improved outcomes compared to the outcomes of the overall study population in a previous clinical trial and uses one or more characteristics of the subgroups to screen candidates for a new clinical trial. Candidates with characteristics matching those defining the subgroups generated from the previous clinical trial may be selected to participate in the new trial and / or assigned to specific roles within the new trial.

[0056] As noted above, the conditions for which method 100 can be used may be diseases, disorders, or other conditions for which an effective treatment is desired. Examples of such diseases include coronary artery disease, cardiovascular disease, stroke, diabetes, dementia, neurodegenerative diseases including Parkinson's disease (PD) and amyotrophic lateral sclerosis (ALS), cirrhosis of the liver, musculoskeletal diseases, skin diseases, endocrine system diseases, eye diseases, intestinal diseases, infectious diseases such as amoebic, viral, and bacterial diseases, prion diseases, respiratory diseases, asthma, urinary system diseases, chronic obstructive pulmonary disease (COPD), cancers such as breast cancer, prostate cancer, colon cancer, and lung cancer, and psychiatric disorders.

[0057] Therapies being investigated in previous and new clinical trials can be any treatment designed to produce measurable results in patients, either in humans or animals. Therapies can be associated with pharmaceuticals. For example, the treatment can be the administration of a new drug, a new application of an existing drug, a new or different dosage of a drug, or a different combination of multiple drugs. The treatment can be a new or different therapy, including psychotherapy, physical therapy, behavioral therapy, or cognitive therapy, or the treatment can be or include a new or different device.

[0058] Step 102 of method 100 includes constructing one or more predictive models for predicting the progression of a condition in a patient. The predictive models are constructed to generate a prediction of the patient's future condition progression based on the patient's clinical data. The predictions can include various indicators commonly used as outcomes in clinical trials. Examples of predictive models include regression models and time-to-event models. Regression models can predict a patient's ability or condition at a future time point, while time-to-event models determine the log-likelihood that a patient will reach a certain ability (disability) or condition, including death.

[0059] Figure 2 shows examples of regression and time-to-event models for patients with ALS. Each model is associated with a clinical measure that clinicians can use to measure a patient's ALS progression. Each model is named according to the clinical measure it is designed to predict. For example, the ALSFRS-R total score, a functional assessment score for ALS. The test to create this score includes 12 questions related to the ability to perform tasks. Patients are rated on a 5-point scale (e.g., 0 = unable to perform, 4 = moderately well) for their ability to perform the tasks. The scores for the individual items are summed to form a total score (e.g., 0 = worst, 4 = best). Based on the patient's clinical data, the ALSFRS-R model can predict a patient's ALSFRS-R score at a defined time in the future (e.g., 6 months, 1 year, 5 years, etc.).

[0060] Another example of a predictive model for ALS is the "50% vital capacity" time-to-event model, which predicts the log-likelihood of a patient reaching 50% vital capacity. Vital capacity is a clinical measure of the amount of air a patient can expel. 50% vital capacity is a milestone often used in ALS to prescribe non-invasive mechanical ventilation. Similarly, the "% predicted vital capacity" regression model predicts % predicted vital capacity at a defined time point in the future.

[0061] Any of a number of predictive modeling techniques known to those skilled in the art can be used to create a predictive model, including regression and machine learning techniques. Regression techniques can include, but are not limited to, linear regression, logistic regression, Cox proportional hazards models, time series models, classification and regression trees, and multivariate adaptive regression splines. Machine learning techniques include k-nearest neighbors, deep learning algorithms, convolutional neural networks, support vector machines, or naive Bayes. These modeling techniques can be used to perform variable reduction to select the most predictive feature set and then build a predictive model. Models can also be combined using known techniques to create ensemble models, which may result in improved accuracy.

[0062] One or more predictive models may be constructed by training a learning machine with clinical data of patients having the condition for which the predictive model is being constructed. The clinical data may be obtained from one or more databases 140 of historical clinical data collected from individual patients. This data generally captures how each patient's disease has progressed over time. The data may include vitals such as height, weight, pulse rate, and forced vital capacity; genetic data; biomarker data; disease progression scores; event dates such as date of symptom onset, date of diagnosis, and date of death; gender; age; smoking and drinking habits; medication use; occupation; and other types of relevant information. An example of a database that stores historical clinical data is the PRO-ACT database of ALS patients, which contains data collected from over 10,700 ALS patients from completed ALS clinical trials.

[0063] Other examples of clinical information databases include the heart disease database and diabetes database maintained by the University of California, Irvine. Another example is the "All of Us" general patient database maintained by the NIH. Archived clinical research datasets, such as those provided by the NIH, can be used. Individual datasets can be used individually as "clinical trial datasets" for testing, or they can be combined into large disease datasets for model training. Alzheimer's disease data from the Critical Path Institute (CODR) and ADNI can be used to build models for Alzheimer's disease treatment trials. For Parkinson's disease treatment trials, models can be built using the PPMI Parkinson Disease Database provided by the MJ Fox Foundation. For cancer treatment clinical trials, models can be built using the Cancer Genome Atlas (TCAG). A type 1 diabetes treatment model can be built using the TEDDY type 1 diabetes dataset.

[0064] Predictive models can be constructed to predict any outcome of interest to a user, including outcomes useful as primary or secondary endpoints for measuring the effectiveness of a drug candidate, including continuous and categorical outcomes. Useful outcomes include clinical measures such as survival, disease recurrence, and vitality; laboratory values ​​such as cholesterol and serum glucose levels; diagnostic test results such as angiograms; and biomarkers indicating disease progression or severity. Other models can predict outcomes considered to be broader indicators of disease burden, such as absenteeism from work, or more patient-centric outcomes, such as quality of life indicators and functional status.

[0065] A preferred embodiment would be predictive outcomes useful in ALS clinical trials, including time-to-event outcomes such as survival rate, time to 50% vital capacity, time to wheelchair use, time to loss of voice, and time to feeding tube use, or regressive outcomes such as ALSFRS-R functional score, or ALSFRS-R subscores (including bulbar score, respiratory score, fine motor score, and gross motor score), or percent predicted vital capacity, or vital capacity in liters.

[0066] Another preferred outcome is the prediction of ADAS-Cog score and / or activities of daily living, global severity, or global change assessment, or Mini-Mental State Examination (MMSE) by the Alzheimer's disease model.In Parkinson's disease, the primary outcome may be the prediction of the Unified Parkinson's Disease Rating Scale (UPDRS) or any of its subscores, or the Parkinson's Disease Cognitive Rating Scale (PD-CRS), the Montreal Cognitive Assessment (MoCA), the PD Cognitive Outcome Scale (SCOPA-COG), the Mini-Mental State Examination (MMSE), or the Mattis Dementia Rating Scale (MDRS) by the Parkinson's disease model.

[0067] In some embodiments, clinical data from disease databases is reformatted and / or screened before being used to train predictive models. The data in the database may have been collected by different parties at different times using different clinical techniques and may therefore not be appropriate in its raw form. Therefore, the data must be harmonized or screened to generate training data that has consistent content and format from one patient to the next. This may involve simple tasks such as standardizing date formats and units of measurement, or more complex tasks such as determining the version of a clinical test used to generate a test score that has changed over time. Some individual data points may have been entered using any of a variety of methods known to practitioners in the field. The reformatting and / or screening of clinical data used for training may be an automated task, or it may involve one or more individuals, such as individuals experienced in clinical disease evaluation processes, making decisions regarding the reformatting or screening of data within a given patient record or set of patient records.

[0068] In step 104, a predictive model 150 is selected to analyze data generated during a clinical trial for a treatment for a condition. In some embodiments, the predictive model is selected from a set of available predictive models. The predictive model 150 may be selected based on the type of condition and the desired outcome of the treatment. For example, for a clinical trial conducted to evaluate the effectiveness of a drug designed to extend the lives of ALS patients, a predictive model that predicts time to death in ALS patients may be selected, or for a clinical trial of a drug designed to slow the decline of functional impairment in ALS patients, a predictive model that predicts ALSFRS-R scores may be selected.

[0069] According to some embodiments, the method for enrolling candidates in a new clinical trial based on data from a previous clinical trial may not include step 102 and / or step 104. One or more predictive models may have already been constructed at a previous time. One or more existing models may be provided by a third party, such as through purchase or publication. In some embodiments, a user may select an existing predictive model from a set of available models (e.g., selecting an ALS predictive model for analyzing an ALS clinical trial from among predictive models for numerous diseases). In other embodiments, the user does not select a predictive model; instead, the predictive model may be a default or otherwise pre-selected model.

[0070] In step 106, patient baseline records for patients who participated in prior clinical trials are extracted from the clinical trial database 160. The patient baseline records may include any patient data generated before the patient received the clinical trial treatment or placebo. The baseline data may be data used to screen the patient for enrollment in the prior clinical trial or other patient information. The patient records may be received from a third party, such as one that conducted the clinical trial, or may be retrieved from local databases or public databases for individual clinical trials. The patient baseline records generally include records for patients in the control group(s) and the treatment group(s).

[0071] In step 108, patient baseline records from the previously conducted clinical trial are provided to predictive model 150 to generate predictions for patients in the previously conducted clinical trial. As described above, predictive model 150 was constructed by training a learning machine on historical clinical data of patients with the disease, so predictions for patients in the previously conducted clinical trial will be generated based on clinical data from patients not participating in the clinical trial (patients whose data were used to construct the predictive model).

[0072] The output of the analysis of patient baseline records using predictive model 150 is a set of predictive data indicating the predicted progression of the condition of patients who participated in a previously conducted clinical trial. For example, for a predictive model configured to predict ALSFRS-R scores one year after enrollment in a previously conducted clinical trial to test the effectiveness of a drug to treat ALS, including 100 study patients, the predictive data includes the 100 predicted ALSFRS-R scores (the predicted ALSFRS-R scores for each patient in the previously conducted clinical trial). In some embodiments, the predictive data includes the probability of reaching a certain milestone. For example, the predictive data may include the likelihood that the patient will die in 10 years, or the log-likelihood value itself.

[0073] In step 110, clinical trial patient data is grouped into subsets based on the predictive data. Each subset corresponds to a subgroup of clinical trial patients. For each subgroup, the effectiveness of treatment is evaluated by analyzing the data of the corresponding subset. The clinical trial patient data subsets are associated with subgroups of clinical trial patients with similar predictive values ​​within a certain range. For example, a data subset may include data from patients with predictive values ​​in the range of 1 to 1.5. In some embodiments, the data subsets are selected based on the proportion of patients belonging to the relevant subgroup. For example, three subsets of clinical data may be defined, each associated with one-third of the clinical trial patient population. Within each data subset, each third of the subgroups consists of patients with consecutive ranges of predictive values. For example, the first subgroup may include patients with predictive values ​​in the range of 0.0 to 0.5, the second subgroup may include patients with predictive values ​​in the range of 0.5 to 1.0, and the third subgroup may include patients with predictive values ​​in the range of 1.0 to 1.5. Subsets may be defined for subgroups of any size. For example, a subset may refer to a subgroup having 50%, 25%, 10%, 5%, 2%, or any other percentage of a clinical trial population. Subsets need not be mutually exclusive. For example, a first subgroup may include patients with a predictive value in the range of 0.0 to 0.5, a second subgroup may include patients with a predictive value in the range of 0.0 to 1.0, and a third subgroup may include patients with a predictive value in the range of 0.0 to 1.5.

[0074] The subsets of clinical trial data include trial outcome data generated during clinical trials, i.e., data reflecting the effectiveness or lack thereof of treatments and placebos. Thus, the clinical trial data includes measurements of patient status relevant to the conduct of the clinical trial. Each subset includes data for patients who received a therapeutic agent during the clinical trial and may also include data for control patients, such as patients who received a placebo. The clinical trial outcome data may be extracted from the clinical trial database in step 112.

[0075] The effect of a clinical trial on a subgroup of patients is assessed by analyzing data from a relevant subset of clinical trial patient data. For example, clinical trial outcomes measured during the clinical trial are compared between treatment and control groups to generate a measure of treatment effectiveness. The measure of effectiveness for any subgroup may be less than, equal to, or greater than the measure of effectiveness for the full analysis set (the complete set of data on which the success of the previously conducted clinical trial was determined). The measure of effectiveness may be, for example, a probability value (p-value) or other appropriate value for measuring the effectiveness of a treatment in a clinical trial, or more generally, the outcome of the trial.

[0076] In step 114, the efficacy measures for the subsets of the clinical trial data are queried to identify one or more subsets of the clinical data (corresponding to subgroups of patients) that exhibit different (e.g., higher) efficacy measures compared to the full analysis set. This identification includes determining whether any of the efficacy measures are above or below a threshold. For example, the success of a clinical trial may be determined based on whether there is a statistically significant difference between the treatment and control groups based on a p-value of 0.05 or less, or the p-value for the subset of data may be searched for values ​​equal to or less than 0.05.

[0077] To the extent that any subgroup of patients produces improved or sufficiently improved results compared to the full data set, it can be concluded that the subgroup of patients shares a common characteristic that leads to a detectable result. Subgroups are generally more homogeneous in one or more characteristics than the full patient population, and the homogeneity of subgroup members associated with a sufficiently high measure of efficacy contributes to a detectable effect of the subgroup relative to the full patient population. A detectable result for an identified subgroup can be due to an improved treatment effect for patients within the group, e.g., a treatment effect greater for patients with a certain characteristic than for patients without that characteristic. The detection of an identified subgroup can also be due to the removal of noise in the data, i.e., a reduced RMSE. For example, patients whose disease progresses more rapidly than other patients in a clinical trial will have greater variance in clinical outcome data (e.g., a larger RMSE) and will fail to demonstrate a statistically significant effect from the treatment. Subgroups lacking these patients may demonstrate statistically significant results that the full analysis set does not. Thus, for example, the treatment effect for the full group and the subgroup may be the same, but the RMSE for the subgroup may be smaller than for the full group as long as the p-value for the subgroup effect size is less than 0.05 when the p-value for the full group effect size is greater than or equal to 0.05. The improved effect size associated with the subgroup may be due to a combination of increased treatment effect and reduced noise.

[0078] In step 116, one or more predictive values ​​associated with one or more subsets showing improved results compared to the full analysis set (e.g., those with p-values ​​less than 0.05) are identified and established as future clinical trial patient enrollment screening criteria. The one or more predictive values ​​can be used to screen candidates for future clinical trials for the treatment, as described further below. The one or more predictive values ​​are the predictive values ​​used to define the identified subsets. For example, a subset of clinical trial data identified as showing a detectable effect of the treatment can include data for patients with predicted progression values ​​ranging from 0.5 to 1.0. These upper and lower predicted state progression values ​​for the identified subset of clinical trial data are established as upper and lower screening thresholds for screening candidates for future clinical trials. In some embodiments, the method concludes by providing the screening criteria to a user for use in designing future clinical trials.

[0079] According to some embodiments, the method includes screening candidate patients for enrollment in future clinical trials. In step 118, clinical data for screening relevant patients for enrollment in future clinical trials for the same treatment is received. This clinical data can be received from the clinician who generated some or all of the clinical data or from a third party that collects the data from the clinician. For example, a physician specializing in ALS treatments can enter clinical data generated about one or more of his or her patients into a candidate screening portal to determine whether one or more of his or her patients should be enrolled in a clinical trial, or a drug development company planning a future clinical trial can collect candidate patient information to provide for the screening process.

[0080] In step 120, one or more condition progression predictions are generated for the candidate patient by inputting the candidate's clinical data into predictive model 150 (i.e., the same predictive model used to generate predictions for the previously conducted clinical trial). Thus, for example, if predictive model 150 was used to generate ALSFRS-R scores for patients in the previously conducted clinical trial, step 120 generates an ALSFRS-R score for the candidate patient.

[0081] In step 122, the candidate patient's predicted value(s) from step 120 are compared to the future clinical trial patient enrollment screening criteria established in step 116 to determine whether the candidate patient should be enrolled in the clinical trial. If the candidate patient's predicted value(s) meet the criteria, the candidate patient is identified for enrollment in the clinical trial. If the candidate patient's predicted value(s) do not meet the criteria, the candidate patient is identified for denial of enrollment in the clinical trial.

[0082] In some embodiments, the screening criteria may be upper and lower predictive value thresholds, such that candidate patients having predictive values ​​that fall within the range defined by the upper and lower predictive value thresholds are identified for enrollment in the clinical trial, and candidate patients having predictive values ​​that fall outside the range are identified for denial of enrollment in the clinical trial. In some embodiments, the screening criteria may include only an upper threshold or only a lower threshold, such that a candidate patient's predictive value must be greater than or less than the threshold, respectively, to be identified for enrollment in the clinical trial.

[0083] Steps 118-122 may be repeated for any number of candidate patients. In step 124, the clinical trial is enrolled with the patients selected for enrollment. In step 126, the clinical trial is conducted, and enrolled patients may receive treatment or a placebo according to well-known clinical trial methodologies. In some embodiments, enrolled patients receive different levels of treatment based on screening. For example, the clinical trial may be defined to include low-dose and high-dose groups, and screening may identify which patients should be included in which groups.

[0084] At the end of the clinical trial, the results of the trial are determined. The results may indicate a detectable therapeutic effect not detected in previous clinical trials, which may be due to screening candidate patients based on the results of previous clinical trials, as described above. As a result of method 100, a therapeutic developer may have evidence to pursue further therapeutic development or approval from a regulatory agency, such as the Food and Drug Administration. In such cases, the therapeutic would otherwise have been abandoned due to the failure of the initial clinical trial. This may prevent the loss of significant investment in therapeutic development and increase the number and quality of treatments available to patients, potentially leading to improved quality of life, relief of suffering, and prolonged survival.

[0085] 3A-3D illustrate one technique for identifying patient subgroups associated with improved treatment outcomes relative to the full set of 1,373 subjects analyzed, which can be used to establish screening criteria for screening candidates for future clinical trials. This technique can be used, for example, in step 114 of method 100. In this technique, outcome data from a previously conducted clinical trial are grouped into subsets based on patient subgroups, and the treatment effect, root mean square error (RMSE), and effect size are determined for each subset. The effect size is used as a measure of the effectiveness of the treatment for the patient subgroup corresponding to each subset, and is calculated by dividing the treatment effect by the RMSE. Subgroup members defining the data subsets are selected based on predicted condition progression data generated using a predictive model (e.g., the predicted data generated in step 108 of method 100). Subgroups associated with increased effect sizes may represent patient populations for which future clinical trials are likely to show detectable or improved treatment outcomes relative to previously conducted clinical trials. Therefore, the predicted condition progression data for such subgroups can be established as screening criteria for candidate patients for future clinical trials.

[0086] Figures 3A-3D are graphs showing the treatment effect, root mean square error, effect size, and p-value for a given subset, each as a function of the proportion of the total population in the subgroup corresponding to that subset from the previous clinical trial. Each data point on the graph represents a given subgroup's respective metric. The groups are defined according to a "low risk," "intermediate risk," or "high risk" prediction of progression. In the illustrated example, the predicted progression metric is the 50% predicted vital capacity risk for ALS patients, but the illustrated concepts are applicable to any metric.

[0087] On the left side of each graph, the percentage of the population belonging to each group is shown as one-third, which is the minimum group size used in the illustrated embodiment. However, any group size may be used. The low-risk group 302 includes the one-third of the population with the lowest predicted 50% vital capacity risk, the high-risk group 304 includes the one-third of the population with the highest predicted 50% vital capacity risk, and the medium-risk group 306 includes the remaining one-third of the population. The respective metrics for each group are plotted. Toward the right of each graph, the size of each group is increased to include more patients in the effect analysis until the entire population is included. The size of the low-risk group is increased first by including lower members of the medium-risk group, the size of the high-risk group is increased first by including higher members of the medium-risk group, and the size of the medium-risk group is increased by incorporating members from both the low-risk and high-risk groups. For each increased group size, the effect size for that group is determined and plotted on the respective graph. The indices for the three groups converge to an index value for the full analysis set ("FAS"), which represents the respective index value for the entire population.

[0088] Figure 3C shows the maximum extent of the low-risk group 302, reaching approximately 80-90% of the population. This size means that the low-risk group has the lowest predicted status progression value and includes 80-90% of the overall clinical trial population. While the effect size for this subgroup is smaller than the medium effect size of 33%, this subgroup represents just under 90% of the population, and using this subgroup's predictive data as screening criteria may mean that a greater proportion of future trial candidates will be selected for enrollment. Using this information, users can simulate several trials to determine the characteristics of the trial that best suit their needs: a trial with a high therapeutic effect that rejects many candidates, or a trial with a low therapeutic effect that includes more candidate patients.

[0089] Figure 3D shows the p-values ​​reached for the three risk groups depicted in Figures 3A-C, as determined using covariate-adjusted linear regression analysis. As can be seen, only the intermediate-risk group 306 achieved a p-value less than 0.05, and this is for a sample size that is approximately 65-75% of the total sample size (1,373 in this experiment). Therefore, trials using the intermediate-risk group threshold can be conducted with a sample size equivalent to 65% of 1,373, or fewer than 900 patients. This represents a significant savings in both time and cost for drug sponsors conducting trials.

[0090] The superimposed black dashed lines in Figure 3C show the leveling of effect sizes for the medium-risk group at approximately 70% of the study population (representing the 70% of the study population with the lowest risk prediction) and the leveling of effect sizes for the complete analysis set. According to some embodiments, the subgroup associated with this leveling of the medium-risk group can be selected to establish screening criteria because it is associated with the highest effect size of the analyzed subgroups and encompasses 70% of the study population. The chart in Figure 3C shows that the patient population of a clinical trial can be divided into subgroups whose size is systematically adjusted for analysis by progressively including the closest individuals.

[0091] Figures 4A and 4B show another technique for identifying patient subgroups associated with improved treatment effect compared to the full analysis set, which may be used to establish screening criteria for screening candidates for future clinical trials. Figures 4A and 4B show effect size "heat maps" representing measures of treatment effect for patient subgroups from previous clinical trials.

[0092] Each block in the heatmap represents a subgroup whose members are defined by upper and lower condition progression prediction thresholds. For each block, a statistical analysis is performed on the clinical trial data associated with the subgroup represented by the block to determine the subgroup's trial outcome. As described further below, the heatmap can be used to identify subgroups that exhibit statistically significant effect sizes (e.g., p-values ​​less than 0.05). Alternatively, heatmaps of treatment effect and, separately, RMSE can be generated. These can be used to identify zones with maximum treatment effect and minimum RMSE, which can be used to find the overlapping region of maximum treatment effect and minimum RMSE. These metrics can be used to simulate sample sizes that result in effect sizes with p-values ​​less than 0.05.

[0093] The y-axis of the heatmap in Figure 4A displays the upper predicted progression threshold, and the x-axis displays the lower and upper predicted progression thresholds. In the example shown, the predicted values ​​are log-likelihoods ranging from a low of -2.98 to a high of 0.64. Each axis also displays the percentage of the entire clinical trial patient population. To determine the size of a subgroup as a percentage of the patient population, use the corresponding x-axis and y-axis values ​​for the associated box. The top-left box represents the entire patient population because it corresponds to the upper and lower thresholds that capture 100% of the patient population. That is, 100% of patients had a predicted value of 0.64 or less (the y-axis value corresponding to the top-left box) and 0.64 or more (the x-axis value corresponding to the top-left box). The highlighted box, with upper and lower thresholds of -1.94 and -2.66, represents the subgroup of patients with predicted values ​​ranging from -2.66 to 1.94. The size of this subgroup is 40% of the total patient population and is derived from the chart by subtracting the proportion of the population corresponding to the upper predictive value threshold from the proportion of the population corresponding to the lower predictive value threshold ("54% - 14%").

[0094] The black diagonal line represents the boundary where the lower threshold falls below the upper threshold. The empty lower right half therefore represents the region of no possibility, where the lower threshold is greater than the upper threshold. The diagonal line does not quite touch the region where blocking begins, because the sample sizes in the squares near the diagonal are too small (e.g., subgroups with only one or two patients) to make meaningful comparisons.

[0095] In the illustrated embodiment, three colors are used to display the boxes, each corresponding to a p-value threshold calculated from the clinical trial outcome data for the patient subgroups corresponding to the box. In this example, the p-value is used as a measure of treatment effect. Boxes corresponding to data with a p-value of 0.05 or less are displayed in red, boxes corresponding to data between 0.05 and 0.10 are displayed in orange, and boxes corresponding to data greater than 0.10 are displayed in yellow. The yellow box in the upper left corner of the chart in Figure 4A (representing the complete analysis set) indicates a low probability that the clinical trial represented in this example will demonstrate a positive treatment effect because its p-value is greater than 0.10. Meanwhile, the red-highlighted box (with a p-value of 0.05 or less) indicates a subgroup with a high probability of demonstrating a positive treatment effect. Identifying this subgroup can be used to establish screening criteria for future clinical trials. The same is true for the subgroups represented by the boxes surrounding the highlighted box, which are also red.

[0096] FIG. 4B is a heatmap according to another exemplary embodiment. In this example, the highlighted blocks identify a group of patients described by specific predictive data (e.g., "patients who entered the study with predicted vital capacity 50 log-likelihoods between -1.87 and -0.84"), in which patients in the treatment group responded differently from patients in the control group. If the study had been successful, the top left block, representing the complete analysis set, would have been red. The region of the red blocks shows a robust "hot spot" centered around a lower threshold of -1.87 and an upper threshold of -0.84 for the log-likelihood of reaching 50% vital capacity (ranging from -1.99 to +1.50). This hot spot is robust in the sense that multiple blocks within the area show detectable effects.

[0097] A threshold value for the block at the center of the hotspot may be selected to establish screening criteria for future clinical trials. Thus, future clinical trials may be designed that include only patients whose baseline characteristics result in predictions that fall within the boundaries of the specified threshold. Blocks showing detectable effects close to orange or yellow blocks (e.g., red blocks) may not be good blocks to select for establishing screening criteria because they are close to containing predictive values ​​associated with patients that decrease the treatment effect value and / or increase the RMSE value. Therefore, according to some embodiments, a subgroup located at the center of a hotspot may be selected to establish screening criteria. On the other hand, a single red block may be rejected. In some embodiments, blocks may be selected based on the size of the corresponding subgroup. For example, for multiple adjacent blocks at the center of a hotspot, a block corresponding to a larger subgroup size may be selected, as this would result in broader screening criteria and more candidates being accepted for enrollment in future clinical trials.

[0098] In some embodiments, multiple hotspots may be generated, and multiple subgroups may be selected to establish multiple sets of screening criteria. For example, a block within a first hotspot corresponding to a relatively low predictive index (e.g., patients with slowly progressing disease) may be identified to establish screening criteria for a low-dose arm of a future clinical trial, and a block within a second hotspot corresponding to a relatively high predictive index (e.g., patients with rapidly progressing disease) may be identified to establish screening criteria for a high-dose arm (i.e., more aggressive treatment) of a future clinical trial.

[0099] 5 is a functional block diagram of an enrollment screening system 500 that can be used to determine which candidate patients should be enrolled in future clinical trials for treatments that have already undergone prior clinical trials, in accordance with the principles and methods described above. System 500 can be used to perform method 100 described above. System 500 can include one or more servers executing one or more computer programs stored on one or more non-transitory computer-readable media. System 500 can be communicatively coupled to one or more of a disease database 540, a clinical trial database 550, and a clinical trial client system 560.

[0100] System 500 includes functional units such as a model builder 502, a prediction generator 504, a screening criteria generator 506, and an enrollment screener 508. These functional units represent functional components of one or more executable software programs executed by one or more processors of system 500. In some embodiments, simulation generator 505 can be used to enable a user to find the correct sample size for zones with high treatment effects and / or low RMSE, with sample sizes that are too small to test the significance of the effect size. Simulation generator 505 is useful for finding hotspots in relatively small clinical trials.

[0101] The model builder 502 is configured to build one or more predictive models for predicting the progression of a condition in a patient. The model builder 502 may include one or more learning machines that can be trained to predict the progression of a patient's condition based on the patient's clinical records. The model builder 502 may build the predictive model of the condition by training the learning machine on training data that comprises historical clinical data collected from patients with the condition.

[0102] This data may be retrieved from a disease database 540 (such as database 140 described above with respect to method 100), which may be a third-party or local database that maintains clinical information for patients suffering from the disease. The disease database 540 may include patient data from clinics, hospitals, research institutes, government agencies, or any other suitable source of patient data. The disease database 540 may also include patient clinical data from previously conducted clinical trials or studies. Clinical trials and studies may include tests and studies related to any number of different treatments and research topics related to the disease. The patient clinical data may also be from patient registries, which are generally databases of observational patient data rather than from interventional clinical trials. Many patient organizations are creating these types of databases. Patient data may be collected from other data sources, including private non-EHR data aggregators, such as PATIENTSLIKEME (https: / / www.patientslikeme.com), APPLE'S IOS HEALTH KIT (https: / / developers.apple.com / healthkit / ), and GOOGLE FIT (https: / / developers.google.com / fit / ).

[0103] Model builder 502 builds a model that generates predictions for one or more clinical indicators that a clinician can use to measure the progression of a patient's condition, as described above with respect to step 102 of method 100. In some embodiments, model builder 502 builds a predictive model that generates predictions for a single indicator of disease progression, while in other embodiments, model builder 502 builds a predictive model that generates predictions for multiple indicators of disease progression.

[0104] In some embodiments, model builder 502 is configured to reformat, filter, screen, or otherwise manipulate historical clinical data before training a predictive model. Model builder 502 may be configured to facilitate such manipulation by a user, such as by allowing the user to select data for inclusion in a training dataset. In some embodiments, model builder 502 may be configured to manipulate historical data through one or more automated functions. For example, model builder 502 may automatically extract patient clinical data required to train a predictive model, discard unnecessary data, and / or reformat the data based on a predefined training data schema.

[0105] In some embodiments, rather than a model builder for building predictive models, the system 500 includes one or more previously built, saved predictive models. These predictive models may have been previously built by the system 500 or may be provided by a third party. The model builder 502 may also be configured to select a predictive model to use in screening candidates for future clinical trials based on past clinical trials. For example, a past clinical trial may have been conducted to test the effectiveness of a new drug in extending the survival of patients with a certain disease, and a predictive model for predicting the time to death of patients with the certain disease may be selected from the model builder 502.

[0106] The prediction generator 504 is configured to generate predictions of disease progression for patients who participated in a prior clinical trial. The prediction generator uses a predictive model constructed or selected by the model builder 502 and patient baseline data from the prior clinical trial. The prediction generator 504 may be configured to receive patient baseline data from a prior clinical trial database 550, which stores patient clinical data for the prior clinical trial, or may be configured to extract patient baseline data from patient records from the prior clinical trial received from the database 550. The database 550 may be a local or remote database and generally includes baseline data collected for each patient who participated in the prior clinical trial before the trial and study outcome data collected during and / or at the end of the clinical trial. This data may include data for treatment and control groups. The prediction generator 504 generates predictions using only the patient's baseline data.

[0107] The prediction generator 504 generates a set of prediction data that is a function of the prediction model used and the patient baseline data. This data may include a predicted disease progression indicator value for each patient in the patient baseline data. Thus, for example, if a previous clinical trial consisted of 100 patients, the prediction data generated for that clinical trial may be 100 disease progression prediction values. In some embodiments, the prediction generator 504 runs multiple prediction models, or runs a single prediction model that generates multiple indicators, resulting in a prediction data set that includes multiple predictive indicators for each patient.

[0108] The screening criteria generator 506 is configured to group the clinical trial patient data into multiple subsets corresponding to patient subgroups. The subsets are defined based on the prediction data generated by the prediction generator 504. The screening criteria generator is configured to analyze the clinical trial data within each subset to determine the effectiveness of a treatment for the corresponding patient subgroup when only data associated with the patient is considered. The screening criteria generator 506 may generate any number of relevant subgroups of any relevant size. In some embodiments, the screening criteria generator receives input from a user to define a minimum subgroup size and / or subgroup size increments. For example, a user may define a minimum subgroup size of 5% of the entire population and may define 5% increments of subgroup size.

[0109] The screening criteria generator 506 is configured to analyze each subset of data to generate a measure of the effectiveness of the treatment for the patients for a relevant subgroup of patients. The screening criteria generator 506 may be configured to establish screening criteria for future clinical trials of the treatment by identifying subgroups associated with a measure of effectiveness that is higher than the measure of effectiveness of the treatment for the complete first group of patients. For example, the screening criteria generator may identify subgroups associated with a p-value of 0.05 or less when the p-value for the complete study population was greater than 0.05.

[0110] In some embodiments, the screening criteria generator 506 is configured to facilitate a user's identification of a subgroup or subgroups that establish the screening criteria. For example, the screening criteria generator may generate plots such as those shown in Figures 3A-D and 4A-B that allow a user to identify a promising subgroup or subgroups.

[0111] In some embodiments, the screening criteria generator 506 is configured to establish screening criteria by selecting a subgroup or subgroups, as described above, and using the disease progression predictor boundaries of the subgroups for the screening criteria. For example, for a selected subgroup of patients with disease progression predictors ranging from 0.0 to 0.5, the screening criteria can range from 0.0 to 0.5, such that candidates for a future clinical trial must have disease progression predictors ranging from 0.0 to 0.5 to be enrolled in the future clinical trial. In some embodiments, the screening criteria generator can be configured to establish screening criteria for determining subanalysis groupings for candidates. For example, screening criteria can be established for screening candidates into low-dose and high-dose groups for testing.

[0112] Enrollment screener 508 is configured to screen candidates for enrollment in future clinical trials. Enrollment screener 508 may be configured to receive candidate patient information and may generate one or more disease progression predictors for the patient using the predictive models used by prediction generator 504. In some embodiments, the candidate patient information is received from a prospective clinical trial client system 560 that can be used, for example, by a clinician to screen the clinician's patients for enrollment in future clinical trials.

[0113] The enrollment screener 508 may be configured to determine whether the predictive value(s) for the candidate meet the enrollment criteria established by the screening criteria generator. For example, the enrollment screener may determine whether the disease progression predictive value for the candidate falls within a range of predictive values ​​established as screening criteria. The enrollment screener may be configured to provide a notification to the future clinical trial client system 560 of whether the candidate should be enrolled in the future clinical trial. In some embodiments, the enrollment screener 508 is configured to determine a sub-analysis grouping for the candidate. For example, the enrollment screener 508 may determine that the candidate should be enrolled in a “low dose” grouping for the future clinical trial.

[0114] The future clinical trial client system 560 may be a system that is remote from the enrollment screening system 500 and communicatively coupled to the screening system, for example, via the Internet. The client system 560 may be any computer that allows a user to send data to and receive data from the screening system 500. For example, the client system 560 may be a computer in a clinician's office running a web browser that is in communication with the screening system's 500 web server.

[0115] FIG. 6 illustrates a user interface of an exemplary prospective clinical trial client system used by a clinician to provide candidate information to and receive screening results from the screening system 500, in accordance with some embodiments.

[0116] User interface 602 is an exemplary embodiment of a user interface for a user (e.g., a clinician) to enter candidate patient information for screening. A user, such as a clinician, may manually enter the candidate clinical information, upload the candidate's clinical records, or some combination of both. In some embodiments, the party uploading the candidate information is the party that generated the patient information, such as the candidate's attending physician or the attending physician's staff. In some embodiments, the party uploading the candidate information may be a third party that previously collected the candidate information from one or more clinicians who generated the data.

[0117] The user interface 604 provides the candidate with the results of the screening. These results may include whether to enroll the patient in a future clinical trial and / or the patient's grouping within the clinical trial (e.g., whether the patient should be enrolled in a low-dose group or a high-dose group). In some embodiments, the screening results may include an indication of a future clinical trial, such as if multiple future clinical trials are being designed. In some embodiments, the screening results are not provided to the user who originated the candidate information, but rather to a third party that is designing and / or will be conducting the future clinical trial. The party receiving the screening results may be the party that uses or otherwise conducts the screening system 500.

[0118] In some embodiments, the candidate patient's records are received and stored by the screening system 500 for future screening. In some embodiments, the screening system 500 provides notification of the candidate's enrollment at a later date from the time the candidate's information is provided to the registration system 500. For example, an on-site clinician, unaware of the planned future clinical trial, can routinely upload a patient's record and receive notification from the registration system 500 that the candidate has been selected for the future clinical trial. The clinician can then notify the candidate to expedite enrollment. In some embodiments, the party providing the candidate's patient information is the party that receives the screening results and will conduct the future clinical trial.

[0119] The principles, methods, and systems described above can be extended to many different applications, as will be readily recognized by those skilled in the art, examples of which are within the scope of this disclosure and are further described below.

[0120] Predictive Covariate Adjustment - In this use case, predictions are made at the start of the study using patient data measured before and up to the moment the patient begins treatment (the baseline visit). These predictions are additional baseline data ("covariates") that can be used to adjust for analyses using traditional statistical tools. The more predictive the baseline covariates are of the study's measured outcomes (the "endpoints" of the study being measured), the more the adjustment will increase the study's power (the chance of detecting a treatment effect, if there is one).

[0121] Prediction-Based Randomization - In this use case, predictions are used to define strata (e.g., three groups of patients predicted to be "fast," "average," and "slow" progressors). Each of these three groups has a randomization schedule that assigns patients to various study arms (e.g., placebo, high-dose, medium-dose, low-dose). Randomization in this manner should result in a similar proportion of "fast," "average," and "slow" patients across all study arms. Furthermore, because the prediction algorithm utilizes dozens of input variables, the likelihood of confounding variables (i.e., statistically significant differences in that variable between two study arms) is reduced across all of those input variables.

[0122] Prediction-based clinical trial enrichment - In this use case, predictions are used to define strata (e.g., patients predicted to progress by 8 to 16 points in the next 12 months) that are then used as inclusion / exclusion criteria for enrolling patients in a clinical trial. To do this, predictive data must be readily available to investigators when reviewing patients for eligibility. This type of trial design is useful for either predictive enrichment (enrolling patients predicted to respond to the treatment being tested) or prognostic enrichment (enrolling homogeneous patients (predicted to progress in a similar manner) to highlight the drug's effect when they are randomized to different trial arms).

[0123] Prediction-based virtual control - In this use case, for each patient treated in the trial, a prediction is made using only data collected before the moment the therapeutic drug is administered. The patient's disease progression is then observed and compared to the predicted disease course.

[0124] One or more of the above use cases can be combined. For example, a virtual control use case can augment a concurrent placebo control. The virtual control predictions are combined with observed data from patients in the placebo group. This creates a larger placebo group for comparison with the observed treatment data. Another combination is the combination of the randomization use case with the covariate adjustment use case. According to some embodiments, the clinical trial enrichment use case is combined with the covariate adjustment use case, and in other embodiments, the clinical trial enrichment, randomization, and covariate adjustment use cases are combined.

[0125] API Tools for Populating EDC Systems with Predictive Data - Clinical trials often use electronic data capture (EDC) software systems to collect data. In this use case, the system can host a software API that can integrate with the EDC system using scripts hosted by the system. The script passes the data to the API, which calculates predictive values ​​and returns them to the EDC system where they are stored. Additional use cases (e.g., any of the use cases listed above or a combination thereof) can then leverage this data. One advantage of this use case is that by collecting predictive data at the time the data is flowing into the EDC, it can be incontrovertibly proven that all predictions were generated prospectively, without knowledge of the patient's true progression after the drug intervention was administered.

[0126] API tools for feeding predictive data into a test analytics system - This use case is similar to the previous one, except the predictions are not prospective. Instead, they are made using historical data and projected forward to another point in time (e.g., the present, or another past date).

[0127] Simulation Generator - A software platform that provides users (e.g., biostatisticians at a CRO or in-house at a pharmaceutical company) the ability to simulate clinical trial designs using patient forecasting applications to optimize clinical trial design parameters. Clinical trial simulations can use a database of historical patient data to select and simulate patients. A subset of patients is selected (randomly or otherwise) from the database and modified with simulated treatment effects. In this use case, the above use case can be packaged and provided to end users, allowing them to run these simulations themselves. Individual cell treatment effects and RMSEs can be used as inputs to the simulation generator, providing a method for determining sample sizes for future trials to achieve desired power.

[0128] Screening of healthcare patient outcomes – The software system can be integrated into the healthcare provider's EHR / EMR system.

[0129] Healthcare Patient Screening - A hospital (or clinic, health system, or other healthcare provider) has a UI built into their website. Prospective patients can enter their information into a form. This form sends the data to a prediction engine, which returns a preliminary analysis based on limited data. The patient then schedules an appointment with the healthcare provider to meet with a doctor for further testing and analysis. This provides the healthcare provider with a way to identify new patients who may benefit from their medical services.

[0130] Population Health - Based on predictions, healthcare organizations can understand the health status of patients under their care, which can inform them if patients are likely to experience a particular health event in the near future, effectively forecast resource needs, and analyze their performance compared to other clinics.

[0131] Health Insurance - By predicting the likely progression of a patient and the timing of disease-related events, health insurance ("payers") can better forecast and plan for cash outflows associated with individual patients.

[0132] Life Insurance - Unlike health insurance companies, life insurance companies have access to an applicant's entire medical records during underwriting. Life expectancy underwriting helps insurance companies price each insured's mortality risk.

[0133] Life Settlement (LS) - Similar to life insurance, life settlement (or "biotics") providers are investment companies that pay cash to individuals today, and in return, the LS provider is named as the beneficiary of the insured's life insurance. The LS provider is also responsible for paying any remaining premiums. Upon the insured's death, the LS provider receives the death benefit. These policies are traded among LS investment companies that hold portfolios of these policies. The value of the policy is determined by the amount of the death benefit and the likely timing of the payment, which occurs upon the insured's death. By more accurately estimating patients' life expectancies, LS providers can more easily assess the market value of these LS policies.

[0134] Figure 7 illustrates an example computer according to one embodiment. Computer 700 may be a component of a system for displaying patient information according to the systems and methods described above, such as system 500 of Figure 5. In some embodiments, computer 700 is configured to perform a method for displaying patient information, such as method 100 of Figures 1A and 1B.

[0135] The computer 700 may be a host computer connected to a network. The computer 700 may be a client computer or a server. As shown in FIG. 7 , the computer 700 may be any suitable type of microprocessor-based device, such as a personal computer, a workstation, a server, or a handheld computing device such as a phone or tablet. The computer may include, for example, one or more of a processor 710, input devices 720, output devices 730, storage 740, and communication devices 760. The input devices 720 and output devices 730 generally correspond to those described above and may be connectable to or integrated with the computer.

[0136] Input device(s) 720 may be any suitable device that provides input, such as a touchscreen or monitor, a keyboard, a mouse, or a voice recognition device. Output device(s) 730 may be any suitable device that provides output, such as a touchscreen, a monitor, a printer, a disk drive, or a speaker.

[0137] Storage 740 may be any suitable device providing storage, such as electrical, magnetic, or optical memory, including RAM, cache, hard drive, CD-ROM drive, tape drive, or removable storage disk. Communication device 760 may include any suitable device capable of sending and receiving signals over a network, such as a network interface chip or card. Computer components may be connected in any suitable manner, such as via a physical bus or wirelessly. Storage 740 may be a non-transitory computer-readable storage medium containing one or more programs that, when executed by one or more processors, such as processor 710, cause the one or more processors to perform methods described herein, such as method 500 of FIG. 5.

[0138] Software 750, which may be stored in storage 740 and executed by processor 710, may include, for example, programming that embodies functionality of the present disclosure (e.g., as embodied in systems, computers, servers, and / or devices such as those described above). In some embodiments, software 750 may include a combination of servers, such as an application server and a database server.

[0139] The software 750 may also be stored in and / or transmitted to any computer-readable storage medium for use by or in connection with an instruction execution system, apparatus, or device, such as those described above, that can fetch and execute instructions associated with the software from the instruction execution system, apparatus, or device. In the context of the present disclosure, a computer-readable storage medium may be any medium, such as storage 740, that contains or can store programming for use by or in connection with an instruction execution system, apparatus, or device.

[0140] The software 750 may also be propagated in any transmission medium for use by or in connection with an instruction execution system, apparatus, or device, such as those described above, that can fetch and execute instructions associated with the software from the instruction execution system, apparatus, or device. In the context of this disclosure, a transmission medium is any medium that can communicate, propagate, or transmit programming for use by or in connection with an instruction execution system, apparatus, or device. Transmission-readable media may include, but are not limited to, electronic, magnetic, optical, electromagnetic, or infrared wired or wireless propagation media.

[0141] The computer 700 may be connected to a network, which may be any suitable type of interconnected communications system. The network may implement any suitable communications protocol and may be secured by any suitable security protocol. The network may include any suitable arrangement of network links capable of implementing the transmission and reception of network signals, such as wireless network connections, T1 or T3 lines, cable networks, DSL, or telephone lines.

[0142] Computer 700 may implement any operating system suitable for operating on a network. Software 750 may be written in any suitable programming language, such as C, C++, Java, or Python. In various embodiments, application software embodying functionality of the present disclosure may be deployed in different configurations, such as, for example, in a client / server arrangement or via a web browser as a web-based application or web service.

[0143] The foregoing description has been set forth with reference to specific embodiments for purposes of explanation. However, the illustrative discussion above is not intended to be exhaustive or to limit the invention to the precise form disclosed. Many modifications and variations are possible in light of the above teachings. The embodiments were chosen and described in order to best explain the principles of the technology and its practical application. Those skilled in the art may thereby best utilize the technology and various embodiments with various modifications suited to the particular use contemplated.

[0144] Although the present disclosure and examples have been fully described with reference to the accompanying drawings, it should be noted that various changes and modifications will become apparent to those skilled in the art. Such changes and modifications are to be understood as falling within the scope of the disclosure and examples as defined by the claims. Finally, the entire disclosures of the patents and publications referenced in this application are incorporated herein by reference.

Claims

1. 1. A computer-implemented method for determining a treatment for a condition, comprising: calculating a prediction for at least one condition progression indicator that predicts progression of the condition for a first group of patients participating in a clinical trial for a treatment for the condition; grouping clinical trial data associated with the first group of patients into a plurality of subsets of clinical trial data based on the prediction of the at least one condition progression indicator; Including, the prediction is calculated using a machine learning predictive model and clinical data associated with the first group of patients; the machine learning predictive model is trained using clinical data relating to a second group of patients having the condition; each subset of clinical trial data is associated with a corresponding subgroup of said first group of patients; The method comprises: For each subset of clinical trial data, calculating a measure of the effectiveness of the treatment for the subgroup of patients corresponding to the respective subset; identifying at least one subset of the plurality of subsets having a measure of efficacy of treatment that is higher than a measure of efficacy of treatment for the first group of patients as a whole; determining a treatment for the condition based on a range of values ​​of the prediction for the at least one condition progression indicator associated with the identified at least one subset of the plurality of subsets having a measure of effectiveness of treatment that is higher than the measure of effectiveness of treatment for the first group of patients as a whole; The method further comprises:

2. 10. The method of claim 1, comprising selecting the machine learning predictive model from a plurality of different machine learning predictive models based on a goal of treatment.

3. training the machine learning predictive model; and standardizing clinical data associated with the second group of patients prior to training the machine learning predictive model. The method of claim 2.

4. The method of any of claims 1 to 3, wherein the clinical data relating to the second group of patients includes historical data of the at least one condition progression indicator.

5. The method according to any one of claims 1 to 4, wherein the clinical data relating to the first group of patients is clinical data generated prior to the clinical trial.

6. The method of any one of claims 1 to 5, wherein the number of patients associated with one of the plurality of subsets is equal to the number of patients associated with at least one other of the plurality of subsets.

7. 6. The method of claim 1, wherein the number of patients associated with one of the plurality of subsets is different from the number of patients associated with at least one other of the plurality of subsets.

8. The method according to any one of claims 1 to 7, wherein the measure of efficacy of said treatment is based on said at least one condition progression indicator.

9. 9. The method of any of claims 1 to 8, wherein the measure of effectiveness of treatment for the first group of patients as a whole indicates that there is no statistically significant difference in treatment outcome between the control group and the treatment group.

10. The method of any of claims 1 to 9, wherein the measure of efficacy is generated by comparing treatment outcomes of control patients with treatment outcomes of treated patients.

11. The method of any preceding claim, wherein identifying a subset of the plurality of subsets comprises identifying a subset having a measure of efficacy that satisfies a threshold test.

12. 12. The method of claim 11, wherein the threshold test is a p-value not exceeding 0.

05.

13. 13. The method of claim 1, wherein identifying at least one subset of the plurality of subsets comprises plotting a representation of a measure of effectiveness as the prediction for subgroup size and the at least one condition progression indicator.

14. Plotting a representation of the effectiveness measure includes generating a heat map; A change in appearance in the heat map indicates a change in the efficacy measure. The method of claim 13.

15. 15. The method of claim 14, wherein the heat map comprises a representation of a measure of effectiveness of a treatment for an entire group of patients.

16. The method of claim 14 , wherein identifying at least one subset of the plurality of subsets comprises identifying locations on the heat map that are surrounded by locations that have the same appearance.

17. The method of any preceding claim, wherein the at least one condition progression indicator comprises predicted patient regression or predicted time to an event.

18. The method of any preceding claim, wherein the patient is a human and the condition is a disease.

19. 19. The method of claim 18, wherein the disease is a neurodegenerative disease or cancer.

20. 20. The method of any of claims 1-19, wherein the clinical trial is configured to determine at least one of the effectiveness of a drug, the effectiveness of different dosages of a drug, or the effectiveness of different combinations of drugs.

21. 21. The method of any of claims 1 to 20, comprising treating the patient with the treatment based on a prediction for a condition progression indicator of the patient that is within a range of predicted values ​​based on at least one progression indicator associated with at least one identified subset of a plurality of subsets for which the effectiveness of the treatment is greater than the effectiveness of the treatment for the first group of patients as a whole.

22. The method of any of claims 1 to 21, wherein determining a treatment comprises determining a dose level.

23. 1. A system for enrolling candidates in a clinical trial for a treatment, the system including one or more processors, a memory, and one or more programs stored in the memory and executable by the one or more processors; The one or more programs: calculating, for a first group of patients participating in a clinical trial for a treatment for a condition, a prediction for at least one condition progression indicator that predicts progression of said condition; grouping clinical trial data associated with the first group of patients into a plurality of subsets of clinical trial data based on the prediction of the at least one condition progression indicator; Includes instructions for the prediction is calculated using a machine learning predictive model and clinical data associated with the first group of patients; the machine learning predictive model is trained using clinical data relating to a second group of patients having the condition; each subset of clinical trial data is associated with a corresponding subgroup of said first group of patients; The one or more programs: For each subset of clinical trial data, calculating a measure of the effectiveness of the treatment for the subgroup of patients corresponding to the respective subset; identifying at least one subset of the plurality of subsets having a measure of efficacy of treatment that is higher than a measure of efficacy of treatment for the first group of patients as a whole; determining a treatment for the condition based on a range of values ​​of the prediction for the at least one condition progression indicator associated with the identified at least one subset of the plurality of subsets having a measure of effectiveness of treatment that is higher than the measure of effectiveness of treatment for the first group of patients as a whole; Including instructions for system.

24. 24. The system of claim 23, wherein the one or more programs include instructions for receiving a selection of the machine learning predictive model from a plurality of different machine learning predictive models based on a goal of treatment.

25. training the machine learning predictive model; standardizing clinical data associated with the second group of patients prior to training the machine learning predictive model; 25. The system of claim 24, including instructions for:

26. The system of any of claims 23 to 25, wherein the clinical data relating to the second group of patients includes historical data of the at least one condition progression indicator.

27. The system of any of claims 23 to 26, wherein the clinical data relating to the first group of patients is clinical data generated prior to the clinical trial.

28. The system of any one of claims 23 to 27, wherein the number of patients associated with one of the plurality of subsets is equal to the number of patients associated with at least one other of the plurality of subsets.

29. 28. The system of claim 23, wherein the number of patients associated with one of the plurality of subsets is different from the number of patients associated with at least one other of the plurality of subsets.

30. 30. The system of any of claims 23 to 29, wherein the measure of effectiveness of the treatment is based on the at least one condition progression indicator.

31. 31. The system of any of claims 23-30, wherein the measure of effectiveness of treatment for the first group of patients as a whole indicates that there is no statistically significant difference in treatment outcome between a control group and a treatment group.

32. 32. The system of any of claims 23 to 31, wherein the measure of efficacy is generated by comparing treatment outcomes of control patients with treatment outcomes of treated patients.

33. The system of any of claims 23 to 32, wherein identifying a subset of the plurality of subsets comprises identifying a subset having a measure of effectiveness that satisfies a threshold test.

34. 34. The system of claim 33, wherein the threshold test is a p-value not exceeding 0.

05.

35. 35. The system of claim 23, wherein identifying at least one subset of the plurality of subsets comprises plotting a representation of a measure of effectiveness as the prediction for subgroup size and the at least one condition progression indicator.

36. Plotting a representation of the effectiveness measure includes generating a heat map; A change in appearance in the heat map indicates a change in the efficacy measure.

36. The system of claim 35.

37. 37. The system of claim 36, wherein the heat map comprises a representation of a measure of effectiveness of a treatment for an entire group of patients.

38. 37. The system of claim 36, wherein identifying at least one subset of the plurality of subsets comprises identifying locations on the heat map surrounded by locations that have the same appearance.

39. 39. The system of any of claims 23 to 38, wherein the at least one condition progression indicator comprises predicted patient regression or predicted time to event.

40. The system of any one of claims 23 to 39, wherein the patient is a human and the condition is a disease.

41. 41. The system of claim 40, wherein the disease is a neurodegenerative disease or cancer.

42. 42. The system of any of claims 23-41, wherein the clinical trial is configured to determine at least one of the effectiveness of a drug, the effectiveness of different dosages of a drug, or the effectiveness of different combinations of drugs.

43. The system of any of claims 23 to 42, wherein determining a treatment comprises determining a dose level.

Images