Machine learning model for generating hemophilia-related predictions using sensor data
The use of machine learning models and sensor data to personalize hemophilia treatments addresses the variability in existing treatments, enhancing treatment efficacy by predicting factor levels and bleeding risks.
Patent Information
- Application Number
- JP2023521127
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2020-10-06
- Filing Date
- 2021-09-29
- Publication Date
- 2024-11-14
- Estimated Expiration
- 2041-09-29
AI Technical Summary
Existing hemophilia treatments lack personalization and effective scheduling, leading to variability in treatment effectiveness due to the disease's inherent variability and the development of inhibitors, necessitating improved methods for predicting and administering treatments based on individual patient data.
A computer-implemented method using machine learning models and sensor data to generate hemophilia-related predictions, including pharmacokinetic models and neural networks, to personalize treatment strategies and schedules, and provide real-time recommendations for treatment administration.
Enhances the personalization of hemophilia treatments by predicting factor levels, therapeutic agent levels, and bleeding probabilities, thereby improving treatment efficacy and reducing the risk of abnormal bleeding events.
Smart Images

Figure 0007583928000001 
Figure 0007583928000002 
Figure 0007583928000003
Abstract
Description
[Technical field]
[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims the benefit of and priority to European Patent Application No. 20200358.8, filed October 6, 2020, which is incorporated by reference in its entirety for all purposes.
[0002] Field Methods and systems disclosed herein generally relate to using machine learning models to facilitate intelligent selection of treatment strategies for subjects suffering from hemophilia. More specifically, workflows can be configured to use one or more machine learning models and sensor data representative of a subject's condition (e.g., movement, exercise, etc.) to generate results to inform or indicate treatment selection, treatment schedules, treatment administration, and / or activity recommendations for a particular subject. [Background technology]
[0003] background Blood circulation supports the flow of nutrients and oxygen to various parts of the body, as well as carbon dioxide and waste products from organs. The circulatory system generally supports the free flow of blood, but blood clotting may also be desirable when injury occurs and blood begins to flow out of the organism. A cascade of coagulation factors then often supports clotting (i.e., clot formation). More specifically, multiple initial reactions lead to the formation of thrombin, which converts fibrinogen to fibrin, and a fibrin clot is formed.
[0004] Blood contains proteins (clotting factors) that aid in the generation of more thrombin and therefore blood clotting quickly after injury. People with hemophilia experience longer bleeding and more blood loss after injury because they lack enough clotting factors. As a result, people with hemophilia can have excessive internal bleeding in their joints and brain. If undetected, this internal bleeding can cause seizures, permanent damage to joints, or reduced consciousness.
[0005] Hemophilia is most often inherited through the X chromosome, but it is possible for hemophilia to develop in other ways. For example, mutations may occur during development, or antibodies against clotting factors may develop later in life. In addition, cancer, autoimmune disorders, and pregnancy are associated with the development of hemophilia.
[0006] The severity of hemophilia varies greatly from subject to subject. Some subjects have severe hemophilia, which may correspond to spontaneous bleeding not caused by injury. These subjects are at risk for internal bleeding and frequently undergo prophylactic treatment to reduce the probability of spontaneous bleeding occurring. Some subjects have mild hemophilia, where spontaneous bleeding is unlikely, but injury (e.g., surgery or tooth extraction) may induce excessive bleeding. These subjects often do not undergo prophylactic treatment, although treatment may be administered before or after surgery (for example).
[0007] Traditionally, treatment of hemophilia involved receiving injections of the missing blood clotting factor. Recently, alternative treatments have become available. The majority of treatments require repeated administration to maintain therapeutic drug levels at sufficiently high levels. Determining treatment schedules is complicated by the fact that hemophilia itself accommodates high diversity of expression and subjects exhibit high variability with respect to pharmacokinetics (e.g., absorption rate, excretion rate, etc.). Furthermore, a significant portion of hemophilia subjects (e.g., approximately 1 in 5 with hemophilia A) will develop inhibitors that reduce the effectiveness of factor concentrate treatment.
[0008] US 2019 / 365317 discloses a system for recommending administration of a therapeutic agent to a patient by tracking physiological data. Sensor data is used to quantify the amount and intensity of the patient's physical activity. The patient's pharmacokinetic profile, current plasma concentration levels and the patient's activity level can be used to determine the rate at which the patient metabolizes and / or will metabolize the therapeutic agent. The frequency of activity can be used to determine whether to recommend on-demand treatment or an adjusted treatment schedule.
[0009] However, there remains a need to improve personalized selection of hemophilia treatments and personalized identification of treatment schedules to improve treatment efficacy for individual subjects suffering from hemophilia. Additionally, there is a need to establish interfaces, networks, and communication channels that facilitate rapid detection and characterization of hemophilia-related conditions that warrant treatment. Summary of the Invention
[0010] overview In some embodiments, a computer-implemented method is provided. The method uses one or more data processing workflows to generate a hemophilia-related prediction for a subject (e.g., a subject diagnosed with hemophilia, a subject suspected of having hemophilia, a subject with a family history of hemophilia, a subject with a genetic predisposition to hemophilia, etc.). The data processing workflows can use one or more machine learning models, optionally one or more pre-processing functions, and optionally one or more post-processing functions. For example, an input data set can be generated for the subject that includes a representation of demographic information, diagnostic information, laboratory information, treatment information, symptom manifestation information, etc. Some or all of the input data set can be used to (e.g.) configure and / or select (e.g., via a classifier machine learning model) one or more workflows and / or one or more machine learning models to be used for the subject. The same or different portions or all of the input data set can be processed by the workflows and / or machine learning models to generate a hemophilia-related prediction. The machine learning models used to generate the prediction can include (e.g.,) a pharmacokinetic model and / or a neural network model.
[0011] In some embodiments, a method is provided that includes receiving a subject-specific dataset corresponding to a subject, the subject-specific dataset including or identifying: a type of hemophilia; a treatment type; demographic data; and / or a photograph of a portion of the particular subject or information derived based on a photograph of a portion of the particular subject. At least a portion of the subject-specific dataset is processed using a classifier model to identify one or more population-level machine learning models from among a set of population-level machine learning models. Each of the set of population-level machine learning models includes a machine learning model trained using a training set corresponding to a set of other subjects with hemophilia. One or more instructions are received that identify one or more times at which a treatment-type treatment was administered to the subject. A hemophilia-related time course is predicted for the subject using the one or more times and a data processing workflow that uses one of the one or more population-level machine learning models. A representation of sensor data collected at a device associated with the subject is received. A transformed data processing workflow that generates a hemophilia-related prediction for the subject is determined (based on the representation of the sensor data). A hemophilia-related prediction is generated for the subject using the transformed data processing workflow. A result corresponding to the hemophilia-related prediction is output.
[0012] Generating the hemophilia-associated prediction can include identifying a post-processing algorithm based on the sensor data. Determining the transformed data processing workflow can include identifying an initial result using the data processing workflow, determining a post-processing function based on the sensor data, and transforming the initial result using the initial result and the post-processing function. Determining the transformed data processing workflow can include selecting an activation function.
[0013] The population-level machine learning model can include a pharmacokinetic model.
[0014] Processing the subject-specific dataset with a classifier model to identify one or more population-level machine learning models may include encoding the subject-specific dataset and using a distance-based algorithm to identify a distance between the encoding of the subject-specific dataset and the encoding of other subject-specific datasets included in the training set.
[0015] The method may also include adding at least a hemophilia-related prediction to the predicted hemophilia-related time course, where the one or more predicted factor levels include the predicted factor levels; receiving another representation of new sensor data collected at a device associated with the subject; determining to resume use of the data processing workflow based on the another representation; generating another hemophilia-related prediction for the subject using the data processing workflow and the added time course; and outputting another result corresponding to the another hemophilia-related prediction.
[0016] The method may include determining, based on the sensor data, that a workflow transition condition is satisfied, wherein the transformed data processing workflow is determined in response to determining that the workflow transition condition is satisfied.
[0017] Generating the hemophilia-associated prediction may include determining one or more state or interim variables calculated for a recent time step by a population-level machine learning model, and configuring a transformed data processing workflow to use the one or more state or interim variables.
[0018] The sensor data can represent the movement of the subject.
[0019] The sensor data may represent physiological attributes of the subject.
[0020] The predicted hemophilia-related time course can include predicted levels of a factor, therapeutic agent or active ingredient for each of a set of time points, and the hemophilia-related predictions include different predicted levels of the factor, therapeutic agent or active ingredient associated with different time points.
[0021] The predicted hemophilia-related time course can include, for each of a set of time points, a predicted clotting time or a probability that an abnormal bleeding event will occur, and the hemophilia-related predictions include different predicted clotting times or different probabilities that an abnormal bleeding event will occur associated with different time points.
[0022] The results may indicate a recommendation to reduce the subject's activity level.
[0023] The results can provide recommendations for treating the subject in light of increased probability of abnormal bleeding.
[0024] The method may also include treating the subject with a hemophilia treatment based on the results. The hemophilia treatment may be different from the treatment type administered at one or more times.
[0025] The results may correspond to a recommendation to seek medical attention for bleeding shown in the photograph.
[0026] The subject-specific data set can include data identified in an electronic medical record corresponding to the subject.
[0027] The results may be output on the subject's device via a chatbot.
[0028] In some embodiments, a system is provided that includes one or more data processors and a non-transitory computer-readable storage medium containing instructions that, when executed on the one or more data processors, cause the one or more data processors to perform some or all of one or more of the methods disclosed herein.
[0029] In some embodiments, a computer program product is provided that is tangibly embodied in a non-transitory machine-readable storage medium and includes instructions configured to cause one or more data processors to perform some or all of one or more of the methods disclosed herein.
[0030] Some embodiments of the present disclosure include a system including one or more processors. In some embodiments, the system includes a non-transitory computer-readable storage medium including instructions that, when executed on the one or more processors, cause the one or more processors to perform some or all of one or more methods and / or some or all of one or more processes disclosed herein. Some embodiments of the present disclosure include a computer program product tangibly embodied in a non-transitory machine-readable storage medium including instructions configured to cause one or more processors to perform some or all of one or more methods and / or some or all of one or more processes disclosed herein.
[0031] The terms and expressions used are used as terms of description and not of limitation, and in the use of such terms and expressions there is no intention to exclude any equivalents of the features shown and described or portions thereof, but it is recognized that various modifications are possible within the scope of the invention as claimed. Thus, although the invention as claimed has been specifically disclosed by embodiments and optional features, it will be understood that modifications and variations of the concepts disclosed herein may be resorted to by those skilled in the art, and such modifications and variations are considered to be within the scope of the invention as defined by the appended claims. [Brief description of the drawings]
[0032] The present disclosure is described in conjunction with the accompanying drawings, in which:
[0033] [Figure 1]1 illustrates a network environment in which cloud-based applications are hosted in accordance with some aspects of the present disclosure. [Diagram 2] 1 is a flowchart illustrating an example of a process performed by a cloud-based application to deliver an abbreviated subject record to a user device in connection with a consultation broadcast requesting assistance with a subject's treatment, in accordance with some aspects of the present disclosure. [Diagram 3] 1 is a flowchart illustrating an example process for monitoring user integration of a treatment plan definition (e.g., a decision tree or treatment workflow) and automatically updating the treatment plan definition based on the results of the monitoring, in accordance with some aspects of the present disclosure. [Figure 4] 1 is a flow chart illustrating an example of a process for recommending a treatment for a subject, according to some aspects of the present disclosure. [Diagram 5] 1 is a flowchart illustrating an example of a process for obfuscating query results to comply with data privacy rules in accordance with certain aspects of the present disclosure. [Figure 6] 1 is a flowchart illustrating an example of a process for communicating with a user using a bot script, such as a chatbot, in accordance with some aspects of the present disclosure. [Figure 7] 1 illustrates a network for using an artificial intelligence model to generate an output to facilitate diagnosis and / or treatment of a subject. [Figure 8] 1 illustrates a process for using machine learning models to facilitate the identification of hemophilia treatment strategies. [Figure 9] 1 illustrates a process for generating hemophilia-related outcomes using sensor data. [Figure 10] 1 illustrates a process for generating model predictions using machine learning models and sensor-based post-processing.
[0034] In the accompanying drawings, similar components and / or features may have the same reference label. Furthermore, various components of the same type may be distinguished by following the reference label with a dash and a second label that distinguishes the similar components. If only a first reference label is used in this specification, the description is applicable to any of the similar components having the same first reference label, regardless of the second reference label. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0035] Detailed Description I. Overview In some embodiments, a computer-implemented method is provided. The method uses one or more data processing workflows to generate a hemophilia-associated prediction for a subject (e.g., a subject diagnosed with hemophilia, a subject suspected of having hemophilia, a subject with a family history of hemophilia, a subject with a genetic predisposition to hemophilia, etc.). The data processing workflows can use one or more machine learning models, optionally one or more pre-processing functions, and optionally one or more post-processing functions. For example, an input dataset can be generated for the subject that includes representations of demographic information, diagnostic information, laboratory information, treatment information, symptom manifestation information, etc. Some or all of the input datasets may have been received from one or more electronic medical records (e.g., stored on a server and / or a care provider device). Some or all of the input datasets can be used to (e.g.) configure and / or select (e.g., via a classifier machine learning model) one or more workflows and / or one or more machine learning models to be used for the subject. The same or different portions or all of the input datasets can be processed by the workflows and / or machine learning models to generate a hemophilia-associated prediction. The machine learning models used to generate the predictions may include (for example) pharmacokinetic models and / or neural network models.
[0036] The prediction can include (for example) predicted factor levels, therapeutic levels, active ingredient levels, abnormal bleeding event probability, probability of spontaneous abnormal bleeding events, etc. In some cases, the prediction can include or be used to identify treatment specifications and / or treatment approaches. For example, the prediction can predict when one or more treatments will be administered, one or more doses of the treatments administered, and / or which treatments will be administered to facilitate a given objective (e.g., maintaining the level of a therapeutic agent, active ingredient, and / or factor above a predetermined threshold; and / or maintaining the risk of abnormal bleeding events and / or the risk of abnormal spontaneous bleeding events below a predetermined threshold). The treatment can be provided and / or recommended to the subject according to the prediction.
[0037] In some cases, a given situation may affect the effectiveness and / or bleeding event risk of various hemophilia treatment strategies. For example, a user may be more likely to suffer an injury (which may cause an unusual bleeding event) when engaging in high-intensity activity. As another example, a user who frequently engages in high-intensity activity may be at lower risk of joint bleeding due to increased muscle strength to support the joints, increased coordination, and reduced obesity (thereby reducing joint inflammation). As yet another example, high-intensity activity may alter the pharmacokinetics that affect how quickly a hemophilia treatment is administered and remains effective (e.g., via changes in rates of distribution, metabolism, absorption, and excretion).
[0038] Accordingly, some embodiments generate one or more hemophilia-related predictions based on sensor data associated with a subject. The sensor data may be indicative of (for example) the subject's movements and / or physiological attributes of the subject. For example, the sensor data may be collected via an accelerometer, a gyroscope, a GPS sensor, a pedometer, and / or a heart rate monitor. In some cases, inferences may be made regarding the type of activity the subject is engaging in, the intensity of the activity the subject is engaging in, and / or the activity state of the subject.
[0039] The sensor data and / or inferences made based on the sensor data can be used to (for example) select and / or configure a machine learning model, select and / or configure a pre-processing function, and / or select and / or configure a post-processing function. For example, a first set of rate constants can be defined for a pharmacokinetic model to be used when the user is inferred to be stationary. A second set of rate constants can be defined for a pharmacokinetic model to be used when the user is inferred to be engaged in low intensity activity. A third set of rate constants can be defined for a pharmacokinetic model to be used when the user is inferred to be engaged in high intensity activity. In some cases, the first, second, and third sets of rate constants may have been separately trained using different training sets. In some cases, one of the first, second, and third sets of rate constants can be trained using a training data set, and each rate constant can be adjusted by a corresponding absolute or relative amount to determine a rate constant that corresponds to another rate constant of the second or third sets of rate constants. A dynamic selection of the rate constant set can then be made, and a smooth transition is facilitated by utilizing and / or sharing state variables and / or other interim variables.
[0040] As another example, a function can be learned for each of one or more rate constants that relates the rate constant to the sensor data. At each of one or more time steps, the rate constants can be dynamically determined based on recent sensor data.
[0041] As yet another example, a pharmacokinetic model can be used to generate a hemophilia-associated prediction when a moving average of readings from an accelerometer falls below a threshold, and a neural network can be used to generate a hemophilia-associated prediction when a moving average of accelerometer readings exceeds a threshold.
[0042] As yet another example, thresholds, activation functions, and / or post-processing functions can be configured to be defined separately based on the sensor data. For example, thresholds for issuing warnings (e.g., for treatments that change the treatment and / or activity to be performed) and / or for recommending changes in treatment schedules can depend on the sensor data. Thresholds can be selected using logic that depends on the sensor data (and / or its processed version), one or more conditional statements, step functions, monotonic functions, etc.
[0043] Cloud-based applications storing data across large populations of hemophilia subjects (e.g., associated with diverse hemophilia types, hemophilia severity, demographics, treatments, locations, genetic mutations, activity levels, etc.) can facilitate learning the multidimensional dependencies of hemophilia-associated predictions on various factors (including sensor data).
[0044] The applications (e.g., operating locally on the device and / or using at least in part the results of computations performed on one or more remote and / or cloud servers) can be used (for example) by a subject with hemophilia and / or a caregiver caring for a subject with hemophilia. The present application can perform one or more of the operations disclosed herein. In some cases, the one or more applications can facilitate communication between a subject with hemophilia and a caregiver. Such communication can facilitate (for example) alerting a caregiver to an abnormal bleeding response or injury and / or can facilitate telemedicine (which may be particularly beneficial, for example, if the subject or part of the community has a communicable disease, if the subject has a movement disorder, and / or if the subject is physically remote from the care provider's facility).
[0045] II. Overview of Hemophilia Subtypes, Diagnostic Protocols, Related Medical Tests, Progression Assessment, and Available Treatments II.A. Types of Hemophilia The medical community has defined several types of hemophilia, which can affect treatment options and inform prognosis.
[0046] II.A.1. Hemophilia A, Hemophilia B, and Hemophilia C Hemophilia A, hemophilia B and hemophilia C are all congenital disorders caused by genetic mutations present at birth. Mutations in the FVIII (or F8) gene cause hemophilia A, mutations in the FIX or (F9) genes cause hemophilia B, and mutations in FXI (or F11) cause hemophilia C. Of people with hemophilia, approximately 80% have hemophilia A, approximately 15% have hemophilia B, and approximately 5% have hemophilia C.
[0047] The FVIII and FIX genes are located on the long arm of chromosome X (Xq28 and Xq27.1, respectively). The FXI gene is located on the long arm of chromosome 4 (4q35.2). Mutations in the FVIII gene result in reduced production of clotting factor VIII or the production of abnormal forms of this factor. Similarly, mutations in the FIX gene result in reduced production of clotting factor IX or the production of abnormal forms of this factor. Mutations in the FXI gene result in factor XI deficiency. Factors VIII, IX, and XI are all involved in the clotting reaction.
[0048] More specifically, factor VIII is typically in a stable form as a result of binding to von Willebrand factor. However, when injury occurs, thrombin is generated. Thrombin binds to and "activates" factor VIII, separating it from von Willebrand factor. Factor XI is activated by factor XIIa, thrombin, and activated FXI itself.
[0049] Activated factor VIII serves as an essential cofactor for factor IX (which itself is activated in response to injury by activated factor XI) in the intrinsic coagulation cascade, and activation of factor VIII by factor IX triggers the subsequent chain of reactions that result in blood clots. Low or absent levels of factor VIII, factor IX or factor XI prevent the formation of stable blood clots, in that blood may not clot initially or that a formed clot may be destroyed.
[0050] II.A.2. Mild, moderate, and severe hemophilia Hemophilia A and hemophilia B are further typically characterized as mild, moderate, or severe based on the activity level of clotting factors (e.g., factor VIII or factor IX) in the subject's blood. Hemophilia is characterized as mild when the factor activity level is between 6%-49% (e.g., 0.06-0.49 IU / mL in adults). Hemophilia is characterized as moderate when the factor activity level is between 1-5% (e.g., 0.01-0.05 IU / mL in adults). Hemophilia is characterized as severe when the factor activity level is less than 1% (e.g., less than 0.05 IU / mL in adults). The normal level of factor IX in subjects of very young age (e.g., at birth or less than 6 months) may be lower than the normal factor level in adult subjects. Thus, testing for factor IX levels may, in some cases, be delayed until the child reaches a certain age (e.g., 6 months) or, if low levels are detected at an earlier age, may be repeated when the child is a certain age.
[0051] Severe cases are the most common, with 60% of people with hemophilia having this form, while 15% of hemophilia cases are moderate and 25% are mild. Typically, the severity does not change throughout life, except that a form of factor IX deficiency called hemophilia B Leiden may become less severe after puberty.
[0052] Individuals with severe hemophilia experience frequent spontaneous and abnormal bleeding (e.g., bleeding for long periods of time) even from small injuries. Spontaneous bleeding occurs most frequently in joints, but internal bleeding (e.g., in the kidneys, brain, and gastrointestinal tract) is not uncommon. Recurrent joint bleeds can cause inflammation, damage cartilage, and permanently damage joints. Minor injuries can result in muscle hematomas or internal bleeding that may appear several days after the initial injury.
[0053] Individuals with moderate hemophilia also experience abnormal bleeding after minor injuries (e.g., potentially to a lesser extent), but rarely experience spontaneous bleeding. Individuals with mild hemophilia experience abnormal bleeding after major injuries, surgery, and / or tooth extractions, but not after minor injuries or spontaneously. Thus, severe hemophilia is typically diagnosed early in life (e.g., within the first year), moderate hemophilia is typically diagnosed later, however in early childhood, and severe hemophilia is often diagnosed later in life (e.g., after surgery or tooth extraction).
[0054] Hemophilia C is generally not characterized as mild, moderate or severe, even when factor XI activity falls into the "severe" range (0.15-0.20 U / mL), because bleeding abnormalities are fairly minor (e.g., the individual does not experience spontaneous bleeding, but bleeding after surgery may be abnormal).
[0055] II.B. Diagnosis of Hemophilia Diagnostic protocols for hemophilia can depend on the subject's age and whether the subject has a family history of hemophilia.
[0056] Diagnosis of hemophilia and hemophilia subtypes frequently involves prothrombin time (PT) assays and / or partial thromboplastin time (PTT) assays; platelet counts (eg, and / or complete blood counts); mixed assays; and / or factor assays.
[0057] II.B.1.Complete blood count / platelet count Platelet counts in hemophilia are generally normal. Platelet counts can be determined by estimating the count on a blood smear.
[0058] II.B.2. Prothombin Time Assay Prothrombin times in hemophiliacs are also generally normal. To determine prothrombin time, a sample of the subject's blood is first mixed with an anticoagulant (e.g., 3.2% buffered sodium citrate). The sample mixture is then mixed with a prothrombin time reagent (containing thromboplastin and calcium chloride), which overcomes the anticoagulant properties of the anticoagulant. The time to clotting is measured optically. A normal prothrombin time is approximately 15-19 seconds. Prolongation of PT is consistent with deficiencies of either factor II, factor V, factor VII or factor X; consistent with oral anticoagulants, liver disease or vitamin K deficiency. Notably, prolongation of PT is not consistent with deficiencies of factor VIII, factor IX or factor XI.
[0059] II.B.3. Partial Thromboplastin Time Assay Partial thromboplastin time (PTT) and / or activated partial thromboplastin time (aPTT) of hemophilia patients are generally prolonged. To measure PTT or aPTT, assays can be performed using partial thromboplastin time (PTT) or activated partial thromboplastin time (aPTT) methods, and may also use factor-deficient substrates. More specifically, plasma can be obtained from a subject and mixed with a specific amount of anticoagulant (e.g., 3.2% buffered sodium citrate). The sample can then be evaluated using aPTT-based and / or PTT-based clotting time assays. Clotting time assays can include one-stage clotting time assays.
[0060] The assay can be performed by combining the sample mixture with a substrate that is deficient in a particular factor (e.g., factor VIII, factor IX, or factor XI) but contains normal levels of other factors. To obtain the aPTT, an aPTT reagent is further introduced and then incubated for a specific time. Then, for either the PTT or aPTT assay, calcium can be added to induce clotting, and the clotting time can be measured optically. The clotting time of the sample can be compared to the clotting time of a control plasma (e.g., pooled from multiple subjects). A normal PTT is between 60-70 seconds, and a normal aPTT is between 30-40 seconds. A PTT of more than 100 seconds indicates a subject that has experienced spontaneous bleeding, and an aPTT of more than 70 seconds indicates a subject that has experienced spontaneous bleeding.
[0061] In some cases, multiple different dilutions are performed using the sample mixture (or control mixture) and substrate. The dilutions may include (for example) 1 / 10 (assuming to have full activity of the factor in the sample), 1 / 20 (assuming to have 50% activity of the factor in the sample), 1 / 50 and 1 / 100. PTT or aPTT values can be determined for each dilution. For each of the sample and control mixtures, the clotting time and dilution are used to determine a log-lin fit. Factor clotting activity can be determined based on the point on the control mixture fit that corresponds to the same estimated clotting time as the sample mixture is fitted at a 1 / 10 dilution (for example).
[0062] II.B.4. Mixed Testing There are multiple potential reasons why a subject's PTT or aPTT can be prolonged. In the case of congenital hemophilia, the prolongation is due to a deficiency of a clotting factor. However, another potential explanation (e.g., applicable to other subjects) is that the plasma sample contains an inhibitor of the factor, which can occur in acquired hemophilia. Mixing studies can facilitate the distinction between these situations. In a mixing study, the subject's plasma is mixed with a control plasma containing normal factor levels. If this mixing negates or "corrects" the prolonged clotting of the subject's plasma, it may be inferred that the subject's plasma was deficient in the factor. On the other hand, failure to correct may indicate or suggest that the subject's plasma contains an inhibitor (e.g., a clotting inhibitor).
[0063] Some control plasmas may include "normal plasma" associated with subjects who are not hemophilic, do not have a factor deficiency, and are not using anticoagulants. A determination that prolonged clotting can be corrected by mixing with normal plasma is consistent with a deficiency of factors VIII, IX, and XI. Some control plasmas may include "adsorbed plasma" associated with subjects who are deficient in a factor (e.g., factor IX) and an anticoagulant (e.g., an oral anticoagulant). A determination that prolonged clotting can be corrected by mixing with adsorbed plasma is consistent with a deficiency of a single clotting factor.
[0064] If clotting is determined to persist when mixed, the blood may contain lupus anticoagulant or the subject may have acquired hemophilia. The presence of low levels of factors (e.g., factor VIII or factor XI) and factor inhibitors (e.g., factor VIII inhibitors or factor XI inhibitors) is consistent with acquired hemophilia, whereas the presence of lupus anticoagulant is consistent with lupus.
[0065] II.B.5. Chromogenic Assay Chromogenic assays (e.g., chromogenic factor VIII assay, chromogenic factor IX assay, or chromogenic factor XI assay) can distinguish the amount of a particular factor in a sample. The rate-limiting step in each of these assays is the amount of factor VIII, factor IX, or factor XI present in the sample.
[0066] For the chromogenic factor VIII assay, in the first step, sample plasma from a subject is combined with a reagent containing activated factors IX and X. Factor VIII in the subject's plasma assists in the activation of factor X (in response to thrombin activation of factors VIII and VIIIa, factors VIII and VIIIa react with Ca2 + , together with phospholipids and factor IXa, activates factor X). In a second step, the chromogenic substrate is cleaved by the activated factor X. Factor VIII activity in the subject sample can then be determined based on the optical measurement (e.g., identifying absorbance at a particular wavelength).
[0067] For the chromogenic factor IX assay, in the first step, the sample plasma is combined with a reagent containing activated factors XI and VIII, calcium and phospholipids, resulting in activation of factors IX and X. A substrate is then cleaved, which is measured by an optical readout (e.g., identifying absorbance at a particular wavelength).
[0068] For chromogenic factor VIII or IX assays, the factor concentration can be determined using a reference relationship (eg, a reference linear relationship) that relates absorbance at a given frequency to factor concentration.
[0069] II.B.6. Bethesda Assay The factor assay can include the Bethesda assay (or modifications of the Bethesda assay, e.g., the Nimgen modification of the Bethesda assay), which can facilitate the distinction between factor inhibitors and factor deficiencies due to other causes. The Bethesda assay can detect the titer of a particular factor inhibitor (e.g., factor IX inhibitor or factor XI inhibitor), such as IgG subclass 4 and / or IgG subclass 1. One Bethesda unit (Bu) corresponds to the amount of inhibitor in a plasma sample required to neutralize 50% of a single unit of a given factor after incubation at 37° C. for 2 hours. The assay is performed by mixing diluted plasma from a subject with control plasma from multiple other subjects (in a 1:1 ratio), incubating at 37° C. for 2 hours, and detecting residual factor activity. This residual factor activity is compared to a reference factor activity detected after incubation of a mixture of control plasma and buffer at 37° C. for 2 hours. The Bethesda titer is determined based on the comparison by using a defined relationship relating inhibitor concentration to residual factor activity.
[0070] The Nijmegen modification of the Bethesda assay derives the Bethesda titer in the same way, but the mixtures used to detect residual factor activity are different: the first mixture is a combination of diluted subject plasma with a buffer control sample, and the second mixture is a combination of a plasma sample lacking the factor of interest with a buffer control sample.
[0071] II.B.7. von Willebrand factor assay Von Willebrand factor is the carrier protein for factor VIII. Von Willebrand factor levels in hemophilia are generally normal. To determine the quantity and quality of Willebrand factor, one or more von Willebrand factor assays can be performed. Von Willebrand factor assays can include a multimer assay (VWF:multimer analysis), an antigen assay (VWF:Ag), a ristocetin cofactor assay (VWF:RCof) or a collagen binding activity assay (VWF:CBA). The antigen assay (VSF:Ag assay) is a quantitative assay that indicates the level of von Willebrand factor in the subject's plasma but does not detect many quality defects (e.g., those that may be present in subjects with type 2 von Willebrand disease). The VSF:CBA and VSF:multimer analyses provide an output that indicates the quality of von Willebrand factor but provide limited quantitative data. The VSF:RCof detects both the quantitative level of von Willebrand factor in effect and the quality of the factor. However, the VSF:RCof assay has challenges with consistency and reliability.
[0072] Both qualitative and quantitative von Willebrand factor deficiencies are important to detect when determining a hemophilia diagnosis. Qualitative defects, such as those associated with type 2 von Willebrand disease, may impair binding of factor VIII to von Willebrand factor and therefore may be associated with symptoms and various other assay metrics that mimic those of hemophilia A. Similarly, quantitative defects in which levels of von Willebrand factor are low may result in reduced factor VIII binding capacity.
[0073] II.B.8. Factor Activity and Diagnostic Criteria Thus, factor activity can be measured using a variety of techniques, including PTT assays, aPTT assays, chromogenic factor assays, or enzyme immunoassays. The measured activity can be compared to one or more thresholds established based on normal factor levels. If the measured activity is below the threshold and alternative explanations for the reduced activity are excluded by other assays (e.g., via indications of normal plasma counts, normal von Willebrand factor activity, absence of coagulation inhibitors, etc.), the data can be consistent with and / or suggestive of a diagnosis of the subject with hemophilia or a subtype of hemophilia.
[0074] The observed decrease in factor VIII activity (e.g., below a normal range threshold, such as below 50% or 40% of normal levels) is consistent with hemophilia A. However, as noted above, the decrease in factor VIII activity may also be due to one or more other mutations, bleeding disorders, or diseases. Factor VIII levels can depend on factors such as body mass index, glucose (e.g., diabetes mellitus), insulin, fibrinogen, triglycerides, age, contraceptive use, pregnancy, and various diseases (e.g., renal disease, hyperthyroidism, and liver disease). In some cases, when a potential confounding condition is known, the subject's clotting time may be compared to a reference clotting time generated based on control plasma from an individual with the same or similar condition, and / or a normalization of the subject's level (or reference threshold) may be performed. If an alternative explanation is not applicable (e.g., the subject is no longer pregnant), another test may be performed.
[0075] The observed decrease in factor IX activity (e.g., below a normal range threshold, such as below 50% or 40% of normal levels) is consistent with hemophilia B. However, as indicated above, decreased factor IX activity may also result from one or more other mutations, bleeding disorders, or diseases. Factor IX levels may also depend on other factors, such as pregnancy, liver disease, heparin use, warfarin use, or disseminated intravascular coagulation drug use. In some cases, when a potential confounding condition is known, the subject's clotting time may be compared to a reference clotting time generated based on control plasma from an individual with the same or similar condition, and / or normalization of the subject's level (or reference threshold) may be performed. If an alternative explanation does not apply (e.g., if the subject is no longer pregnant or has not received heparin or warfarin for at least a defined period of time), another test may be performed.
[0076] The observed decrease in factor XI activity (e.g., below a normal range threshold, such as below 70% or 50% of normal levels) is consistent with hemophilia C. However, the decreased factor XI activity may also be due to one or more other mutations, bleeding disorders or diseases (e.g., Noonan syndrome).
[0077] II.C. Genetics of Hemophilia Hemophilia is generally congenital (e.g., for about 2 / 3 of hemophilia A cases and about 4 / 5 of hemophilia B cases), but may alternatively be acquired. Congenital hemophilia A and B are X-linked recessive disorders (hemophilia A corresponds to one or more mutations in the FVIII gene, and hemophilia B corresponds to one or more mutations in the FIX gene). Males have a single X chromosome, whereas females have two. Thus, the majority of congenital hemophilia patients are male. On the other hand, females with FIX or FVIII mutations may be carriers of hemophilia.
[0078] Occasionally, women may present with hemophilia A or B. For example, women may have hemophilia A or B as a result of having an affected father or carrier mother, or by having a disorder that results in having only one X chromosome (e.g., Turner syndrome). As another example, a woman may be diagnosed with hemophilia despite having a FIX or FVIII mutation in only one of the two X chromosomes. Bleeding symptoms may occur due to freezing or partial inactivation of normal X chromosome function. Hemophilia diagnostic criteria are met when factor VIII or factor IX levels are below a certain threshold (e.g., 40%).
[0079] Hemophilia C corresponds to one or more recessive chromosome 4 mutations. Thus, there are approximately as many females with hemophilia C as there are males with hemophilia C. The highest prevalence of hemophilia C is observed among the general population of Ashkenazi or Afro-Israeli descent.
[0080] II.C.1. Genetic counseling and hemophilia Genetic counseling can be used to inform parents about the probability that their children will have hemophilia if one or both parents are hemophilia patients and / or carriers. For example, if the father has hemophilia A or B and the mother is not a carrier, each daughter will be a carrier and each child will be neither hemophilia patients nor carriers. On the other hand, if the mother is a carrier for hemophilia A or B and the father is not hemophilia patients, each child has a 50% chance of having hemophilia A or B (and each daughter has a 50% chance of being a carrier). If the father has hemophilia patients and the mother is a carrier, each offspring has a 50% chance of having hemophilia (whether the offspring are daughters or offspring). Beyond these congenital probabilities, hemophilia can present in response to spontaneous mutations (which account for 1 / 3 of hemophilia A cases and 1 / 5 of hemophilia B cases).
[0081] Genetic counseling can further be used to predict the potential severity of hemophilia in offspring. For example, genotypes typically associated with severe phenotypes include: FVIII intron 22 inversion; FVIII intron 1 inversion mutation, FVIII intron 22 inversion mutation, nonsense point mutations in FVIII or FIX (resulting in a new stop codon); splice junction mutations in the FIX gene. Missense mutations in factor IX are characteristic of mild or moderate phenotypes, and single base mutations in the 5' promoter region of factor VIII are associated with mild phenotypes.
[0082] II.C.2. Prenatal / infant testing for hemophilia Prenatal or infant testing can be performed if the mother is identified as a carrier of hemophilia, or if either parent has hemophilia, to test for established linkage within the family. Amniotic fluid (e.g., collected from the uterus by amniocentesis at approximately 16-20 weeks of gestation) or umbilical cord blood (e.g., collected at approximately 20 weeks of gestation or at birth) can be used to determine if a subject (e.g., fetus or newborn) has hemophilia, identify the subject's hemophilia subtype, and / or determine if the subject is a carrier of hemophilia. For chorionic villus sampling, a small sample is collected from the placenta at the uterine wall insertion site (e.g., at approximately 10-12 weeks of gestation). Amniotic fluid, umbilical cord blood, or placenta samples can be used for genetic sequencing to establish if a hemophilia mutation is present.
[0083] Cord blood testing may be sufficient to diagnose a subject with hemophilia A at birth. However, factor IX is vitamin K dependent and is generally present at reduced levels at birth. Thus, subjects suspected of having hemophilia B (e.g., due to family history) are often diagnosed later (e.g., at about 6 months of age), especially if the subject has a mild form of hemophilia B. Factor XI is also reduced at birth and reaches relatively normal levels around 6 months of age.
[0084] II.D. Hemophilia Treatment Hemophilia treatments can include prophylactic treatments, which are performed on a regular or continuous basis; episodic (or episodic) treatments, which are performed in response to monitoring for the reduction of clinically evident bleeding events or complications.
[0085] II.D.1. Preventive measures Prophylactic treatment can be administered using a schedule that is not tied to bleeding events. Prophylactic treatment can be administered to subjects with severe hemophilia and subjects with potentially moderate hemophilia. Prophylactic treatment is generally not provided to subjects with hemophilia C, subjects with mild hemophilia, and subjects with potentially moderate hemophilia.
[0086] Continuous prophylactic treatment can be delivered more frequently than 45 times per year. Continuous prophylactic treatment can be divided into primary prophylactic treatment (initiated without evidence of current or recent bone or joint bleeding and without the occurrence of two bleeding episodes in a large joint); secondary prophylaxis (initiated after the occurrence of two or more bleeding episodes in a large joint and without the onset of joint disease); or tertiary prophylaxis (initiated after, for example, detection of joint disease). Primary prophylactic treatment can include multiple injections per week. Intermittent prophylactic treatment can be delivered to prevent bleeding episodes, but can be delivered less than 46 weeks per year.
[0087] Prophylactic treatment can include administration of the factor (e.g., recombinant factor, plasma-derived factor) that the subject is deficient in. Recombinant products may be particularly beneficial for individuals with more severe mutations (e.g., intron 22 inversions), while purified non-recombinant products may be suitable for individuals with milder mutations. The administered factor can include a factor concentrate. The administration of prophylactic treatment can be determined based on disease severity. One problem with this approach is that subjects (even those corresponding to the same disease severity) exhibit different pharmacokinetics, so a given dose administered to a comparable subject may have different half-lives, clearances and in vivo recovery. In addition, the half-lives of factors VIII and IX are relatively short, so factor VIII concentrate is administered frequently (intravenously) multiple times (e.g., three times) per week.
[0088] Recently, individualized regimens have become more frequently used. Pharmacokinetic computer simulations can be trained using data from a population of subjects. Subject attributes and / or laboratory measurements corresponding to a particular subject are selected and then fed into a trained model that can identify a recommended dosage and / or treatment schedule.
[0089] The subject-specific measurements used to train the model and / or inputs for a particular subject may include a pharmacokinetic curve including a time series of factor levels versus time of bolus. The model may be used to iteratively identify dosing and / or treatment schedule recommendations based on (for example) recent factor levels from the subject (e.g., some or all of a recent pharmacokinetic curve) and / or target minimum (or trough) factor levels. The model may use Bayesian analysis to generate these recommendations. The output of the model may further depend on one or more other dynamic or static subject attributes, such as weight, blood type, and / or von Willebrand factor level.
[0090] The laboratory measurements used to train the model, processed by the model (e.g., to generate subject-specific outputs), and / or monitored for any treatment evaluation can include factor levels measured at specific time intervals relative to treatment administration. The levels and time intervals can be analyzed to determine whether changes in dosage or treatment schedule are recommended.
[0091] In some cases, the factor level is a trough level. Trough level testing can include identifying a minimum factor level. The minimum factor level is most likely to be present just before the prophylactic drug of the maintenance treatment is administered. Thus, blood samples can be collected just before treatment to identify this level. It can be advantageous to maintain the trough factor level above a certain threshold (e.g., 1%, 5%, 10% or 15%) of baseline factor activity. If the trough falls below the threshold, the administration of the treatment and / or the identity of the treatment can be adjusted.
[0092] Gene therapy has recently been explored as another approach to increase factor levels. This approach involves the use of modified viral vectors (e.g., adeno-associated viral vectors) to introduce non-mutated FVIII or FIX genes into the cells of a subject. This viral introduction is designed to induce the subject to produce the same genes so that sustained levels of clotting factors are increased without periodic bolus treatments. Current gene therapies being investigated for hemophilia B include AMT-060, AMT-061, FLT180a, SB-FIX, and SPK-9011. Current gene therapies being investigated for hemophilia A include SPK-8001, SPK-8016, BMN270, and SB-525.
[0093] Subjects with acquired hemophilia may be administered a therapy designed to reduce or eliminate active inhibitors. Treatment may include immunosuppressants, steroids, chemotherapeutic agents (e.g., cyclophosphamide), rituximab, calcineurin inhibitors, and / or mycophenolate mofetil. The dosage of treatment may be tapered slowly and / or repeatedly adjusted to support factor levels within the target open or closed range (e.g., above threshold).
[0094] Several research efforts have attempted to identify hemophilia treatments that are not based on factor replacement. Such approaches can reduce or avoid the risk that subjects will develop inhibitors (and thus become less responsive to treatment). For example, emicizumab is a bispecific antibody that binds to factors IXa and X, thereby mimicking factor VIII. As another example, desmopressin acetate induces the release of factor VIII from tissues. As yet another example, a multispecific antibody (e.g., Q499-z121-J327-z119 / L404-k) containing a first antigen-binding site that binds a clotting factor (e.g., factor IX and / or activated factor IX) and a second antigen-binding site that binds clotting factor X can be administered with factor IX to function as a functional substitute for factor VIII.
[0095] II.D.2. Expression Procedures Hemophilia symptoms can include cases where a subject experiences unexplained or excessive bleeding; prolonged, large, or deep wounds; and / or joint abnormalities (e.g., swelling, pain, or tightness). Symptoms can be treated to slow or stop bleeding by promoting blood clotting. Hemophilia symptoms are typically treated across all types of hemophilia. Treatment can include administering (e.g., intravenously) plasma-derived or recombinant factors (e.g., factor VIII or factor IX), which can include concentrated versions thereof. More recent treatment options include extended half-life factor concentrates (e.g., those that fuse polyethylene glycol, IgG1-Fc, or albumin to the factor), bispecific antibodies (e.g., emicizumab, which partially mimics factor VIIIa for binding to factors IX and X), and agents that activate anticoagulation pathways. Providing treatment immediately after an episode begins can reduce the likelihood that blood will enter the subject's joints, reduce the amount of blood introduced into the subject's joints, and / or reduce the likelihood that bleeding will disrupt essential bodily functions (e.g., breathing).
[0096] The goal, as proposed by the World Federation of Hemophilia, is a symptomatic treatment administered within 2 hours of the onset of a bleeding event. Otherwise, blood in the joint can lead to iron and subsequent macrophage recruitment in the synovium of the joint, resulting in soft tissue damage and subsequent damage to cartilage and bone. Joint damage can lead to chronic pain, reduced range of motion, and impaired activity. Symptoms that may indicate that a subject is experiencing bleeding in the joint can include joint pain, joint tingling, bubbling sensation in the joint, and resistance to movement in the joint with arthritis. Target joints are defined as joints with at least three spontaneous bleedings within a 6-month period. Each bleeding can result in new vulnerable blood vessels that are prone to spontaneous bleeding, resulting in a positive feedback situation that can lead to permanent joint damage.
[0097] Other bleedings that may represent medical emergencies requiring rapid treatment (e.g., via concentrated factors, factor concentrates and / or blood transfusions) include intracranial bleeding, central nervous system bleeding and / or sublingual bleeding (e.g., particularly in children due to their small airways).
[0098] Muscle and nosebleeds should also be addressed with prompt treatment. Muscle bleeding is a significant cause of disability in hemophilic subjects. More specifically, muscle hematomas can induce significant mobility restriction muscle spasms, superficial ecchymosis and inflammation.
[0099] Excessive bleeding after surgery, circumcision, or tooth extraction is common in patients with hemophilia. If the health care provider is aware of the subject's hemophilia, they may administer clotting factors and possibly antifibrinolytic agents before and after the procedure.
[0100] In general, dosage and / or schedule symptomatic treatment can be prescribed in an effort to maintain the level of one or more factors not below a given symptomatic lower threshold. The threshold can be determined based on the type of bleeding. For example, the lower threshold of factor VIII can be set at 30% of the baseline value of normal subjects for early arthritis, bleeding gums and nosebleeds, while the threshold can be set at 50% for later treatment of bleeding or muscle bleeding, and severe bleeding episodes can be associated with a threshold of 80%. Dosage can be calculated based on the lower threshold, current factor level and subject's weight.
[0101] In some cases, additional or alternative treatments are provided in response to symptom manifestations. For example, additional or alternative treatments can promote the cessation of bleeding. Additional or alternative treatments may include (for example) desmopressin acetate (DDAVP) or vasopressin can promote the cessation of bleeding. DDAVP is often administered in response to muscle or joint bleeding, mucosal bleeding, and for surgery. Other agents, such as aminocapric acid, can be used to inhibit the destruction of blood clots.
[0102] Administering treatment in response to bleeding exhibited by subjects with acquired hemophilia is complicated by the fact that the relationship between inhibitors and kinetics is nonlinear and multiphasic, with the first phase showing rapid inactivation and the second phase showing more consistency. Mild bleeding can be resolved by itself or can be treated with factors (e.g., plasma-derived or recombinant factors) or with DDAVP or vasopressin. Heavy bleeding can be treated with DDAVP, vasopressin, or bypassing agents (e.g., plasma-derived activated prothrombin complex concentrate or recombinant activated factor VII). Bypassing agents can trigger the generation of thrombin, resulting in hemostasis. To date, bypassing agents are not highly predictable in terms of efficacy.
[0103] II.E. Hemophilia Surveillance As mentioned above, one type of monitoring that is frequently performed is monitoring factor levels at various times during drug administration (to potentially adjust dosage and / or treatment schedule).
[0104] Factor activity levels may be less informative, especially when evaluating newer non-factor therapies (e.g., emicizumab or fitusiran). For example, the results of aPTT-based one-stage clotting assays and two-stage chromogenic activity assays typically scale similarly based on the amount of factor activity, but the results of these two assays do not show the same correspondence for non-factor therapies. Although aPTT-based one-stage clotting assays may not yield factor activity level results that are monotonically related to the concentration of the non-factor treatment, this type of relationship is observed using chromogenic assays. These results can be the result of emicizumab not inducing factor-VIII-typical actions such as phospholipid binding, thrombin activation or APC inactivation. One approach to monitor non-factor therapies is to monitor thrombin generation assay parameters (e.g., peak thrombin), such as parameters from reactions initiated using non-standard factor XIa. Another approach to monitor non-factor therapies is to use chromogenic factor assays (e.g., using all human reagents). The results can then be scaled using a conversion factor to estimate a metric that corresponds to traditional factor activity levels. Yet another approach is to measure the levels of the therapeutically active agent (eg, using an ELISA-based assay).
[0105] Another type of surveillance that is frequently performed is to determine whether a subject with hemophilia has developed factor inhibitors. Approximately 20% of subjects with hemophilia A and 3% of subjects with hemophilia B develop inhibitors to clotting factor treatment products.
[0106] Inhibitor testing can be performed to determine whether and / or to what extent a subject has developed inhibitors against a coagulation factor. Prior to inhibitor testing, the subject may undergo a washout period (e.g., or 1, 2 or 3 days) (depending on the study design) during which no factor treatment is received. Alternatively, heat treatment of the sample may be performed. To determine whether a subject has developed an inhibitor, two samples can be collected: one before the treatment (e.g., plasma-derived or recombinant factor) is performed and one after. A shorter half-life of the infused factor can indicate the presence of an inhibitor. Inhibitor screening can be performed periodically and / or based on scheduled or unscheduled invasive or intensive treatments.
[0107] Some non-factorial therapies (e.g., emicizumab) interfere with the measurement of the Bethesda assay. An alternative approach is to use chromogenic-based inhibitor assays (e.g., using bovine reagents) that can generate factor inhibitor levels.
[0108] If an inhibitor is detected, the subject may undergo immune tolerance induction. During immune tolerance induction, a high dose of the factor is concentrated. Immune tolerance induction is successful in clearing the inhibitor in approximately 70% of hemophilia A subjects and approximately 30% of hemophilia B subjects, although the timeline for clearance varies significantly between subjects (e.g., between several months to several years). Immune tolerance induction involves repeated administration of the factor (or factor concentrate). Treatment regimens can be fixed or adjusted from onset to the subject's prognostic state. Treatment regimens vary based on dosage (e.g., 25 U / kg, 50 U / kg, or 100 U / kg) and / or frequency (e.g., daily, twice a week, or three times a week).
[0109] In some cases, rather than undergoing immune tolerance induction (or upon failure of immune tolerance induction), expression and / or prophylactic treatment may instead be adjusted to use a bypassing agent instead of the factor concentrate. The bypassing agent may include (for example) FEIBA, NovoSeven RG, or Obizur. In some cases, the bypassing agent includes a recombinant factor.
[0110] Some treatments (e.g., non-factorial therapies) may result in the development of antibodies to the treatment. Detection of prolonged times from a conventional aPTT assay and / or low activity metrics from factorial clobase or chromogenic assays is consistent with the development of antibodies and may indicate the need for further assays to test for the presence of potential antibodies.
[0111] II. Network environment for hosting cloud-based applications configured with intelligent capabilities 1 illustrates a network environment 100 in which an embodiment of a cloud-based application is hosted. The network environment 100 may include a cloud network 130 that includes a cloud server 135 and a data registry 140. The cloud server 135 may execute source code underlying the cloud-based application. The data registry 140 may store data records captured from or identified using one or more user devices, such as a computer 105, a laptop 110, and a mobile device 115.
[0112] The data records stored in the data registry 140 may be structured according to a skeleton structure of fixed parts (e.g., data elements). The computer 105, the laptop 110, and the mobile device 115 may each be operated by a different user. For example, the computer 105 may be operated by a doctor, the laptop 110 may be operated by an administrator of an entity, and the mobile device 115 may be operated by a subject. The mobile device 115 may connect to the cloud network 130 using the gateway 120 and the network 125. In some cases, each of the computer 105, the laptop 110, and the mobile device 115 is associated with the same entity (e.g., the same hospital). In other examples, the computer 105, the laptop 110, and the mobile device are associated with different entities (e.g., different hospitals). The user devices of the computer 105, the laptop 110, and the mobile device 115 are examples for illustrative purposes, and thus the present disclosure is not limited thereto. The network environment 100 may include any number or configuration of user devices of any device type.
[0113] In some embodiments, the cloud server 135 can obtain data (e.g., subject records) for storage in the data registry 140 by interacting with either the computer 105, laptop 110, or mobile device 115. For example, the computer 105 interacts with the cloud server 135 by using an interface to select a subject record or other data record stored locally (e.g., stored on a network local to the computer 105) for inclusion in the data registry 140. As another example, the computer 105 interacts with the interface to provide the cloud server 135 with an address (e.g., a network location) of a database that stores the subject record or other data record. The cloud server 135 then retrieves the data record from the database and populates the data record into the data registry 140.
[0114] In some embodiments, the computer 105, the laptop 110, and the mobile device 115 are associated with different entities (e.g., a medical center). The data records that the cloud server 135 retrieves from the computer 105, the laptop 110, and the mobile device 115 may be stored in different data registries. The data records from each of the computer 105, the laptop 110, and the mobile device 115 may be stored within the cloud network 130, but the data records are not intermingled. For example, the computer 105 cannot access the data records retrieved from the laptop 110 due to constraints imposed by data privacy rules. However, the cloud server 135 may be configured to automatically obfuscate, obscure, or mask portions of the data records when the data records are queried by different entities. Thus, the data records captured from the entities can be exposed to the different entities in an obfuscated, obscured, or masked form to comply with the data privacy rules.
[0115] Once the data records are collected from the computers 105, laptops 110, and mobile devices 115, the data records can be used as training data for training machine learning or artificial intelligence models to provide the intelligent analytic functions described herein. If a user device associated with an entity queries the data registry 140 and the query results include data records originating from different entities, the data records may be available for querying by any entity, given that these data records may be provided or exposed to the user device in an obfuscated format that complies with data privacy rules.
[0116] The cloud server 135 can be configured in a special manner to execute code that, when executed, causes the cloud server 135 to execute an intelligent function using the transformed representation of the subject record (e.g., a vector that numerically represents information stored in the subject record). For example, the intelligent function can be executed by executing code using the cloud server 135. The executed code can represent a trained neural network model. The neural network model may be trained to perform an intelligent function, such as predicting a subject's responsiveness to a treatment regimen, identifying similar patients, generating treatment regimen recommendations for the patient, and other intelligent functions. The neural network model can be trained using a training dataset that includes subject records of subjects who have previously been treated for a condition and experienced an outcome (e.g., overcoming the condition, increasing the severity of the condition, decreasing the severity of the condition, etc.). Additionally, the executed code can be configured to cause the cloud server 135 to convert non-numeric values of existing subject records into a numerical representation (e.g., the transformed representation) that can be processed by the trained neural network model. For example, code executed by cloud server 135 can be configured to receive as input each subject record of a set of subject records, and for each subject record, the code, when executed, can cause cloud server 135 to perform the operations described herein to convert each data element of each subject record into a transformed representation, such as a vector representation. Performing the intelligent function can include inputting at least a portion of the data records stored in data registry 140 into a trained machine learning or artificial intelligence model to generate an output for further analysis. In some embodiments, the output can be used to extract patterns in the data records or to predict values or outcomes associated with data fields of the data records. Various embodiments of the intelligent functions performed by cloud server 135 are described below.
[0117] In some embodiments, the cloud server 135 is configured to enable a user device (e.g., operated by a physician) to access a cloud-based application to send a consultation broadcast to a set of destination devices. The consultation broadcast can be a request for support or assistance regarding the treatment of a subject associated with the subject record. The destination devices can be user devices operated by other users associated with other entities (e.g., physicians at other medical centers). If the destination device accepts the request for assistance associated with the consultation broadcast, the cloud-based application can generate a reduced representation of the subject record that omits or hides certain data fields of the subject record. The reduced representation can comply with data privacy rules, such that the reduced representation of the subject record cannot be used to uniquely identify the subject associated by the subject record. The cloud-based application can send the reduced representation of the subject record to the destination device that accepted the request for assistance. The user operating the destination device can evaluate the reduced representation and communicate with the user device using a communication channel to discuss options for treating the subject. For example, the communication channel can be configured as a secure chat room that allows a user device (e.g., operated by a physician requesting a consultation) to securely communicate with a destination device (e.g., operated by another physician providing a consultation).
[0118] In some embodiments, the cloud server 135 is configured to provide a treatment plan definition interface to the user device. The treatment plan definition interface allows the user device to define a treatment plan for the condition. For example, the treatment plan can be a workflow for treating a subject according to the condition. The workflow can include one or more criteria for defining a population of subjects as having the condition. The workflow can also include a particular type of treatment for the condition. The cloud server 135 receives and stores a treatment plan definition for a particular condition from each user device of the collection of user devices. A cloud-based application can deliver a treatment plan for a given condition to the set of user devices. Two or more user devices of the set of user devices may be associated with different entities. Each of the two or more user devices can provide an option to integrate any part or the entirety of the treatment plan into a customer rule set. The cloud server 135 can monitor whether the user device integrates the entire shared treatment plan or a portion of the treatment plan. The interaction between the user device and the shared treatment plan can be used to determine whether to update the treatment plan or rules created based on the treatment plan.
[0119] In some embodiments, the cloud server 135 allows a user operating a user device to access a cloud-based application to determine a proposed treatment for a subject having a condition. The user device loads an interface associated with the cloud-based application. The interface allows the user operating the user device to select a subject record associated with a subject being treated by the user. The cloud-based application can evaluate other subject records to identify previously treated subjects similar to the subject being treated by the user. The similarity between subjects can be determined, for example, using an array representation of the subject record. The array representation (e.g., a transformed representation such as a vector, an N-dimensional matrix, or any numerical representation of a non-numeric value) can be any numerical and / or categorical representation of the values of the data fields of the subject record. For example, the array representation of the subject record can be a vector representation of the subject record in a domain space, such as in Euclidean space. In some cases, the cloud server 135 can be configured to convert the entire subject record into a numerical representation, such as a vector. For a given subject record, the cloud server 135 can evaluate each data element to determine the type of data contained or included in that data element. The type of data can inform the cloud server 135 as to what process or technique to perform to convert the numeric or non-numeric values of that data element into a numeric representation. As an illustrative example, the cloud server 135 can convert the non-numeric values of a data element in a subject record (e.g., text in a doctor's note) into a numeric representation (e.g., a vector). The conversion can include using natural language processing techniques such as Word2Vec or other text vectorization techniques to generate numeric values representing each word of the text. The generated numeric values can serve as vectors that can be input into a trained neural network to perform intelligent analytics.As another illustrative example, for a data element that includes images (e.g., MRI data) or image frames (e.g., ultrasound video data) of a video, each image or image frame can be converted to a numerical representation (e.g., a vector) using a trained autoencoder neural network that is trained to generate a latent space representation of the input image. The reduced representation (e.g., latent space representation) of the input image can serve as a numerical representation of the input image. This numerical representation can be input to a neural network or other machine learning model to perform intelligent analysis of the associated subject record. As yet another example, for a data element that includes a time-varying information sequence (e.g., events occurring or measurements taken from a subject over a period of time), the time-varying information can be represented as a numerical representation using several illustrative transformations. In some cases, a count of events may be used as a vector representing the time-varying information. For example, if four measurements were taken on a subject over a year, the numerical representation can be "4." In other examples, the frequency or rate of events occurring (e.g., once a week, once a month, once a year, etc.) can be used as a vector representing the time-varying information. In yet other examples, an average or combination of measurements associated with each event in the time-varying information can be used as a vector representing the time-varying information. The disclosure is not limited to these examples and other numerical representations of the time-varying information can be used as a vector representing the numerical representation.
[0120] In some cases, multiple values in the array representation correspond to a single data element or data field. For example, the value of a data element may be represented by multiple binary values generated via one-hot encoding. As another example, each value of the multiple values in a single data element of a subject record may be individually converted to a numeric representation as described above. The numeric representations representing each value of the multiple values may be combined into a single numeric representation corresponding to the data element. The combination of the multiple numeric representations may be performed using any vector combination technique, such as averaging the magnitudes of the vectors, adding the vectors, or concatenating the multiple vectors into a single vector. In some cases, the cloud-based application may generate an array representation of each subject record of a group of subject records. The similarity between two subject records may be represented by comparing the two array representations to determine the distance between them. Instead of comparing the numeric representation of an entire subject record to other numeric representations of other subject records, subject records may also be compared along a dimension (e.g., data elements). For example, comparing two subject records along a dimension may include comparing the numeric representation of a data element of the subject record to other numeric representations of matching data elements of the other subject records. Additionally, the cloud-based application can be configured to identify subjects that are nearest neighbors to a subject record selected by a user device using the interface. The nearest neighbors may be determined by comparing the numerical representations of the various subject records to the numerical representation of a target subject record. The cloud-based application can identify treatments previously performed on the subjects that are nearest neighbors. The cloud-based application can utilize the treatments previously performed on the nearest neighbors on the interface.
[0121] In some embodiments, the cloud server 135 is configured to create a query that searches a database of previously treated subjects. The cloud server 135 may execute the query and retrieve subject records that satisfy the query constraints. However, in presenting the query results, the cloud-based application may present complete subject records only for subjects that have been or are being treated by the user creating the query. The cloud-based application masks or otherwise obfuscates portions of subject records for subjects not treated by the user creating the query. Masking or obfuscating portions of subject records included in the query results allows users to comply with data privacy rules. In some embodiments, the query results (whether or not the query results are obfuscated) may be automatically evaluated for patterns or common attributes in the subject records.
[0122] In some embodiments, the cloud server 135 embeds the chatbot in the cloud-based application. The chatbot is configured to automatically communicate with a user device. The chatbot can communicate with the user device in a communication session in which messages are exchanged between the user device and the chatbot. The chatbot may be configured to select an answer to a question received from the user device. The chatbot can select an answer from a knowledge base accessible to the cloud-based application. When a user device sends a question to the chatbot and the chatbot does not have an existing answer stored in the knowledge base, different phrases of the question for which there is an existing answer are stored in the knowledge base. A user communicating with the chatbot can be prompted as to whether the answer provided by the chatbot is accurate or useful.
[0123] It will be understood that any machine learning or artificial intelligence algorithm can be executed to generate any of the trained machine learning models described herein. A variety of different types and techniques of artificial intelligence-based models and machine learning models can be trained and then executed to generate one or more outputs that predict user outcomes for performing a protocol or function. Non-limiting examples of models include naive Bayes models, random forest or gradient boosting models, logistic regression models, deep learning neural networks, ensemble models, supervised learning models, unsupervised learning models, collaborative filtering models, and any other suitable machine learning or artificial intelligence models.
[0124] It will be understood that the cloud-based application can be configured to perform intelligent functions with respect to consulting external physicians, determining diagnoses, and suggesting treatments for any disease, condition, area of study, or disorder, including, but not limited to, COVID-19, oncology, including cancers of the lung, breast, colorectal, prostate, stomach, liver, cervical (cervix), esophagus, bladder, kidney, pancreas, endometrium, oral cavity, thyroid, brain, ovary, skin, and gallbladder, solid tumors such as sarcomas and carcinomas, cancers of the immune system including lymphomas (such as Hodgkin's lymphoma or non-Hodgkin's lymphoma), and cancers of the bone marrow, such as blood (hematological cancers) and leukemias (such as acute lymphocytic leukemia (ALL) and acute myeloid leukemia (AML)), lymphomas and myelomas. Further disorders include blood disorders such as anemia, bleeding disorders such as hemophilia, blood clots, ophthalmologic disorders including diabetic retinopathy, glaucoma, and macular degeneration, neurological disorders including multiple sclerosis, Parkinson's disease, spinal muscular atrophy, Huntington's disease, amyotrophic lateral sclerosis (ALS), and Alzheimer's disease, autoimmune disorders including multiple sclerosis, diabetes, systemic lupus erythematosus, myasthenia gravis, inflammatory bowel disease (IBD), psoriasis, Guillain-Barre syndrome, chronic inflammatory demyelinating polyneuropathy (CIDP), Graves' disease, Hashimoto's thyroiditis, eczema, vasculitis, allergies, and asthma.
[0125] Other diseases and disorders include, but are not limited to, kidney disease, liver disease, heart disease, stroke, gastrointestinal disorders such as celiac disease, Crohn's disease, diverticular disease, irritable bowel syndrome (IBS), gastroesophageal reflux disease (GERD) and peptic ulcers, arthritis, sexually transmitted diseases, high blood pressure, bacterial and viral infections, parasitic infections, connective tissue diseases, celiac disease, osteoporosis, diabetes, lupus, diseases of the central and peripheral nervous system such as attention deficit / hyperactivity disorder (ADHD), catalepsy, encephalitis, epilepsy and seizures, peripheral neuropathy, meningitis, migraine headaches, myelopathy, autism, bipolar disorder, and depression.
[0126] II.A. The cloud-based application enables user devices to broadcast consultation requests to other user devices and automatically condense subject records to comply with data privacy rules. 2 is a flow chart illustrating a process 200 performed by a cloud-based application to deliver an abbreviated subject record to a user device in association with a consultation broadcast requesting assistance regarding a subject's treatment. Process 200 can be performed by the cloud server 135 to enable user devices associated with different entities (e.g., hospitals) to collaborate or consult regarding a treatment for a subject while complying with data privacy rules.
[0127] Process 200 begins at block 210, where the cloud server 135 receives a set of attributes from a user device. Each attribute in the set of attributes can represent any characteristic of a subject (e.g., a patient). The set of attributes can be identified by a user using an interface provided by the cloud server 135. For example, the set of attributes identifies demographic information of the subject and recent symptoms experienced by the subject. Non-limiting examples of demographic information include age, sex, ethnicity, state or city of residence, income range, education level, or any other suitable information. Non-limiting examples of recent symptoms include subjects who currently or recently (e.g., at last clinic visit, intake, within 24 hours, within one week) experienced a particular symptom (e.g., difficulty in breathing, fever above a threshold temperature, blood pressure above a threshold blood pressure, etc.).
[0128] In block 220, the cloud server 135 generates a record for the subject. The record can be a data element that includes one or more data fields. The record indicates each of a set of attributes associated with the subject. The record can be stored in a central data store, such as the data registry 140 or any other cloud-based database. In block 230, the cloud server 135 receives a request submitted by a user using the interface. The request can be to initiate a consultation broadcast. For example, a user associated with an entity is a physician at a medical center that treats the subject. The user can operate a user device to access a cloud-based application and broadcast a request to assist in treating the subject. The broadcast may be sent to a set of other user devices associated with a different entity.
[0129] In block 240, the cloud server 135 queries the central data store using one or more recent symptoms included in the set of attributes associated with the subject. The query result includes a set of other records, each record of which is associated with another subject. In some cases, the cloud server 135 can query the central data store to identify other subject records similar to the subject record. Similarity can be determined by comparing the transformed representation of the entire subject record to the transformed representation of each of the other subject records. The comparison of the transformed representations can result in a distance (e.g., Euclidean distance) that represents the similarity between the two subject records. In other examples, similarity can be determined based on values included in the data elements. For example, the target subject record can include a subject data element that includes text that represents a symptom experienced by the subject. Each of the other subject records stored in the central data store can also include a data element that includes text that represents the associated subject's symptoms. The cloud server 135 can convert the text included in the subject data element into a numerical representation using techniques described above (e.g., trained convolutional neural networks, text vectorization techniques such as Word2Vec, etc.). The numeric representation of the text contained in the subject data element can be compared to the numeric representation of the text contained in the data element of the matching subject record. The result of the comparison (e.g., in a domain space such as Euclidean space) between the two numeric representations can indicate the degree to which the text contained in the subject data element is similar to the text contained in the data element of the other subject record. In block 250, the cloud server 135 identifies a set of destination addresses (e.g., other user devices associated with different entities). Each destination address in the set of destination addresses is associated with a care provider of another subject associated with one or more other records in the set of other records identified in block 240. In block 260, the cloud server 135 generates a reduced representation of the record for the subject.The condensed representation of a record omits, obscures, or obfuscates at least a portion of the record. The condensed representation of a record cannot be used to uniquely identify the subject associated with the record and therefore can be exchanged between external systems without violating data privacy rules. The cloud server 135 can perform any masking or obfuscation techniques to generate the condensed representation of the record.
[0130] At block 270, the cloud server 135 utilizes a reduced representation of the record having a connection input component (e.g., a selectable link such as a hyperlink that establishes a communication channel) for each destination address of the set of destination addresses. The connection input component may be a selectable element presented at each destination address. Non-limiting examples of a connection input component include buttons, links, input elements, and other suitable selectable elements. At block 280, the cloud server 135 receives a communication from a destination device associated with the destination address. The communication includes an indication that a user operating the destination device has selected a connection input component associated with the reduced representation of the record. At block 290, the cloud server 135 establishes a communication channel between the user device for which the connection input component was selected and the destination device. The communication channel allows a user operating the user device (e.g., a physician treating the subject) to exchange messages or other data (e.g., a video feed) with a destination device (e.g., a physician at another hospital who has agreed to assist in the treatment of the patient) associated with the destination address for which the connection input component was selected.
[0131] In some embodiments, the cloud server 135 is configured to automatically determine the location of the user device and the location of the destination device where the connection input component was selected. The cloud server 135 can also compare the locations to determine whether to generate a condensed representation of the record. For example, in block 260, the cloud server 135 can generate a condensed representation of the record because the cloud server 135 determines that each destination address in the set of destination addresses is not co-located with the user device that initiated the consultation broadcast. In this case, the cloud server 135 can automatically determine to generate a condensed representation of the record to comply with data privacy rules. As another example, if the set of destination addresses are associated with the same entity as the user device that initiated the consultation broadcast, the cloud server 135 can transmit the record in its entirety (e.g., without obfuscating any portion of the record) to the destination device associated with the destination address while complying with data privacy rules.
[0132] In some embodiments, the cloud server 135 generates a plurality of other condensed record representations. Each of the plurality of other condensed record representations is associated with another subject. The cloud server 135 transmits the plurality of other condensed record representations to a user device and receives a communication from the user device identifying a selection of a subset of the plurality of other condensed record representations. Each of the set of destination addresses is represented by one of the condensed record representations. For example, generating the condensed record representation includes determining a jurisdiction of the other subject associated with the condensed record representation, determining a data privacy rule governing exchange of subject records within the jurisdiction, and generating the condensed record representation to comply with the data privacy rule. A first other condensed record representation of the plurality of other condensed record representations can include a particular type of data. A second other condensed record representation of the plurality of other condensed record representations can omit or obscure the particular type of data. For example, the particular type of data can be contact information, identifying information such as a name, a social security number, and other suitable information that can be used to uniquely identify the other subject.
[0133] In some implementations, the communication can be received at a central data store. The communication can be sent by a user device operated by a user and can include an identifier of a target subject record for the target subject. When the communication is received at the central data store, it can cause the central data store to query a set of stored subject records to identify an incomplete subset of the set of subject records. Each subject record of the incomplete subset can be identified and included in the incomplete subset because the subject record is determined to be similar to the target subject record along at least one dimension. The similarity between two subject records along a dimension can represent similarity with respect to data elements of the subject records, such as similarity with respect to symptoms, diagnoses, treatments, or any other suitable data elements. The one or more dimensions along which similarity or dissimilarity is determined can be automatically defined or user defined. Determining the similarity or dissimilarity between the target subject record and each subject record of the set of subject records stored in the central data store may include at least the following operations: retrieving the target subject record based on the identifier included in the communication, generating a transformed representation of the target subject record (or retrieving an existing transformed representation of the target subject record), and performing a clustering operation using the transformed representation of the target subject record and the transformed representation of each subject record of the set of subject records. The clustering operation may be performed on one or more dimensions (e.g., one or more features of the subject records). For example, the clustering operation may cluster the set of subject records stored in the central data store based on a data element that includes a value representing a symptom of the subject. The transformed representation of the target subject record may include a vector representation of a data element that includes a value representing a symptom of the subject. The vector representation of this data element of the target subject record may be compared to a vector representation of a corresponding data element in each subject record of the set of subject records to define a cluster of subject records.Each cluster of subject records can define a group of one or more subject records that share a common characteristic associated with a data element selected as a dimension of similarity. In each cluster of subject records, a Euclidean distance between the transformed representation of the target subject record and other transformed representations of the set of subject records can be calculated. A subject record can be determined to be similar to a target subject record, for example, if the Euclidean distance between the transformed representation of the subject record and the transformed representation of the target subject record is within a threshold.
[0134] III.B. Update of Shareable Treatment Plan Definitions Based on Aggregated User Integration FIG. 3 is a flow chart illustrating a process 300 for monitoring the integration of a treatment plan definition (e.g., a decision tree or a processing workflow) by a user and automatically updating the treatment plan definition based on the monitoring results. The process 300 can be executed by the cloud server 135 to enable a user device to define a treatment plan for treating a population of subjects having a condition. The user device may distribute the treatment plan definition to user devices connected to an internal or external network. The user device receiving the treatment plan definition can decide whether to integrate the treatment plan definition into a custom rule base. The integration into the custom rule base can be monitored and used to automatically modify the treatment plan definition.
[0135] In block 310, the cloud server 135 stores interface data that causes a treatment plan definition interface to be displayed when a user device loads the interface data. The treatment plan definition interface is provided to each user device in the set of user devices when the user device accesses the cloud server 135 to navigate to the treatment plan definition interface. In some embodiments, the treatment plan definition interface allows a user to define a treatment plan for treating a population of subjects having a condition (e.g., hemophilia).
[0136] At block 320, the cloud server 135 receives a set of communications. Each communication in the set of communications is received from a user device in the set of user devices and generated in response to an interaction between the user device and the treatment plan definition interface. In some embodiments, the communications include, for example, one or more criteria for defining a population of subject records. Each criterion may be represented by a variable type. For example, the variable type may be a value or a variable used as a condition of the criterion. The variable type of the criterion of the rule may also be any value of the condition that constrains the population of subjects to an incomplete subgroup. For example, the variable type of a rule defining a population of pregnant women is "IF "subject is pregnant". The criteria may be filter conditions for filtering a pool of subject records. For example, a criterion defining a population of subject records for subjects likely to have a severe form of hemophilia A may include the filter conditions "mutation of factor VIII gene" and "amount of clotting factor VIII less than 0.01 IU / mL". The communications may also include a particular type of treatment for the condition. A particular type of treatment can be associated with a proposed action (e.g., undergoing preventative treatment) or non-action (e.g., minimizing non-essential surgery) to treat a condition associated with the subjects represented by the population of subject records.
[0137] In block 330, the cloud server 135 stores the set of rules in a central data store, such as the data registry 140 or any other centralized server in the cloud network 130. Each rule in the set of rules includes one or more criteria and a particular treatment type to be included in the communication from the user device. As an illustrative example, the rules represent a treatment workflow for treating the subject's hemophilia. The rules include the following criteria (e.g., a condition following an "IF" statement) and the following actions (e.g., a particular treatment type defined or selected by the user following a "THEN" statement): "IF "inhibitor titer" is greater than 5 BU and "spontaneous bleeding in the last month", then "increase treatment dosage", "treat with bypassing agent", or "modify treatment". Additionally, each rule in the set of rules is stored in association with an identifier corresponding to the user device from which the communication was received.
[0138] In block 340, the cloud server 135 identifies a subset of the set of rules that are available among the entities via the treatment plan definition interface. The subset of rules may include a subset of the set of rules that are associated with symptoms and distributed to external systems, such as other medical centers, for evaluation. For example, rules may be selected for inclusion in the subset of rules by evaluating characteristics of the rules or identifiers associated with the rules. The characteristics of the rules may include a code or flag that is stored or added to the stored rules. The code or flag may indicate that the rule is publicly available to external systems (e.g., utilized by the entities).
[0139] At block 350, for each rule of the subset of rules identified at block 340, the cloud server 135 monitors interactions with the rule. The interactions can include an external entity (e.g., outside the entity associated with the user that defined the treatment plan associated with the rule) integrating the rule into a custom rule base. For example, a user device associated with the external entity (e.g., another hospital) evaluates the rules available by the external entity. The evaluation includes determining whether the rule is suitable for integration into a rule set defined by the external entity. A rule can be suitable if a user device associated with the external entity indicates that a treatment workflow defined using the rule is suitable for treating a condition corresponding to the rule. Continuing with the illustrative example above, a rule for treating hemophilia can be made available by an external medical center. A user associated with the external medical center determines that the rule for treating hemophilia is suitable for integration into a rule set defined by the external medical center. Thus, after the rule is integrated into the custom rule base defined by the external medical center, other users associated with the external medical center can execute the integrated rule by selecting the integrated rule from the custom rule base. Additionally, the cloud server 135 monitors the integration of available rules by detecting a signal that is generated or caused to be generated when the treatment plan definition interface receives input from a user device associated with an external entity corresponding to the integration of a rule into the custom rule base.
[0140] As another illustrative example, a user device associated with an external entity uses a treatment plan definition to integrate a modified version of a rule specified by an interaction into a custom rule base. The modified version of a rule specified by an interaction is a portion of the rule selected for integration into the custom rule base. Selecting a portion of a rule for integration includes selecting fewer criteria than all criteria included in the rule for integration into the custom rule base. Continuing with the illustrative example above, a user device associated with an external entity selects the criterion "IF "inhibitor titer" is greater than 5BU" for integration into the custom rule base, but the user device does not select the criterion "spontaneous bleeding within the last month" for integration into the custom rule base. Thus, the interaction-specific modified version of the rule integrated into the custom rule base is "IF "inhibitor titer" is greater than 5BU, then "increase dosage of treatment", "treat with bypassing agent", or "modify treatment". The criterion "spontaneous bleeding within the last month" is removed from the rule to create a modified version of the interaction-specified rule and integrated into the custom rule base.
[0141] In block 360, the cloud server 135 may detect that a modified version of the rule specified by the interaction has been integrated into the custom rule base defined by the external entity. Upon detection, the cloud server 135 may update the rule stored in the central data store of the cloud network 130. The rule may be updated based on the monitored interaction. The term "base" in this example corresponds to "after evaluating" or "using the results of the evaluation" of the monitored interaction. For example, the cloud server 135 detects that a user device associated with the external entity has integrated a modified version of the rule specified by the interaction. In response to detecting the modified version of the rule specified by the interaction, the cloud server 135 may update the rule stored in the central data store from the existing rule to the modified version of the rule specified by the interaction.
[0142] In some embodiments, the cloud server 135 updates the rules by generating an updated version that is utilized across external entities. Another original version can remain unupdated and is available to a user associated with the user device as the source of one or more received communications that identified the criteria and the particular type of action. For example, the cloud server 135 updates a rule stored in the central data store, but the cloud server 135 does not update another rule of the set of rules stored in the central data store.
[0143] In some embodiments, the cloud server 135 can update the rule when an update condition is met. The update condition can be a threshold. For example, the threshold can be a number or percentage of external entities that have integrated a modified version of the rule into their custom rule base. As another example, the update condition can be determined using the output of a trained machine learning model. To illustrate, the cloud server 135 can input the detection signal received from the external entity into a multi-armed bandit model that automatically determines whether and / or when to utilize the rule and / or whether and when to utilize an updated version of the rule. To illustrate by way of non-limiting example only, the rule can be defined as executable code such that, when executed, the rule automatically queries a central data store to identify a subset of the set of subject records for further analysis. Additionally, the rule can include one or more treatment protocols for treating subjects associated with the identified subset of subject records. The rule can be defined as a workflow for defining a subset of the set of subject records and treating the subset associated with the subset of subject records. For example, a rule may include one or more criteria for filtering subject records from a set of subject records and performing a particular treatment protocol on subjects associated with the remaining subject records (e.g., the subject records remaining after filtering has been performed on the set of subject records). Although the rules are defined by a user of a first entity, the rules can be accepted (e.g., integrated into the rule base of the second entity), modified, or completely rejected by an external user (e.g., a doctor working at another hospital) of a second entity (e.g., the first entity and the second entity are two different medical facilities). In some cases, a feedback signal can be sent to the cloud server 135 each time the external user of the second entity accepts a rule and thus completely integrates the rule into its code base.In another example, a feedback signal may be sent to the cloud server 135 whenever a user of the second entity modifies a rule. In another example, a feedback signal may be sent to the cloud server 135 whenever a user of the second entity completely rejects a rule. In each of the above examples, the feedback signal may include the rule (e.g., a rule identifier) and data indicating whether the rule was accepted, modified, or rejected. The multi-armed bandit model (executable by the cloud server 135) may be configured to intelligently select one of the original rule, the modified rule, or a completely different rule for broadcasting to external users of other entities. The selection of the original rule, the modified rule, or the different rule may be based at least in part on the configuration of the multi-armed bandit. In some cases, the multi-armed bandit may be configured by an ε-greedy search technique. In the ε-greedy search technique, the multi-armed bandit model may select the original rule for broadcasting to external users of other entities with a probability of “1-ε”, where ε represents the probability of searching for a new or modified rule. Thus, the multi-armed bandit model can select a modified version of the original rule or a completely new rule with a defined probability of ε. The multi-armed bandit model can modify ε based on feedback signals received from other entities. For example, if the feedback signals indicate that a rule has been modified in a particular way by different external users a threshold number of times, the multi-armed bandit model can learn to select the rule modified in a particular way to broadcast to the external users instead of broadcasting the original rule.
[0144] In some embodiments, the cloud server 135 identifies multiple rules in the set of rules that include criteria corresponding to the same variable type and identify the same or similar types of treatments. The variable type may be a value or variable used as a condition of the criteria. The variable type of the criteria of the rule may also be any value of the condition that constrains the population of subjects into subgroups. For example, the variable type of a rule that defines a population of pregnant women is "IF "subject is pregnant". The cloud server 135 determines the new rule to be a contracted representation of the multiple rules when the new rule is sent to a server that is typically operated by other entities.
[0145] In some embodiments, the cloud server 135 provides another interface configured to receive the set of subject attributes. For example, a user operating a user device accesses the another interface and uses the other interface to select a subject record that includes the set of attributes. The selection of the subject record can cause the cloud server 135 to receive the set of subject attributes. The cloud server 135 identifies (e.g., determines) a particular rule whose criteria are met based on the set of subject attributes. For example, evaluates the set of subject record attributes against the criteria of the rule stored in the central data store. To illustrate, if the set of attributes includes a data field that includes a value of "is pregnant" and if the rule includes a single criterion of "IF "subject is pregnant", the cloud server 135 identifies the rule. The cloud server 135 updates the other interface to present the particular rule and each particular type of processing associated with the particular rule.
[0146] In some embodiments, the rule criteria is a variable type related to a particular demographic variable and / or a particular symptom-type variable. Non-limiting examples of demographic variables include any item of information that characterizes the demographics of a subject, such as age, sex, ethnicity, race, income level, education level, location, and other suitable items of demographic information. Non-limiting examples of symptom-type variables indicate whether a subject currently or recently (e.g., at last visit, at intake, within 24 hours, within 1 week) experiences a particular symptom (e.g., difficulty in breathing, loss of consciousness, fever above threshold temperature, blood pressure above threshold blood pressure, etc.).
[0147] In some embodiments, cloud server 135 monitors data in a registry of subject records, such as subject records stored in data registry 140. Cloud server 135 monitors data in the registry of subject records for each rule of the subset of rules (identified in block 340). Cloud server 135 identifies a set of subjects for which the criteria of the rule are met and a particular treatment has been previously prescribed for the subject. For each of the set of subjects, cloud server 135 identifies a reported status of the subject as indicated from or using an assessment or test. For example, a reported status is any information that characterizes the subject's status in such aspects as whether the subject has been discharged from the hospital, whether the subject is alive, a measurement of the subject's blood pressure, the number of times the subject has woken up during a sleep stage, and other suitable status. Cloud server 135 determines an estimated responsiveness metric of the set of subjects to a particular treatment based on the reported status. For example, if a particular treatment of the rule is to prescribe a medication, the estimated responsiveness metric is a representation of the extent to which the medication addressed the symptoms or symptoms experienced by the subject. As a non-limiting example, the estimated responsiveness metric of the set of subjects can be an average, weighted average, or any sum of the scores assigned to each subject of the set of subjects. The score can represent or measure the effectiveness of the subject's responsiveness to the treatment. In some cases, the cloud server 135 can generate a score representing the effectiveness of the subject's responsiveness to the treatment by using clustering techniques. To illustrate, as a non-limiting example, the set of subject records can represent subjects who have previously undergone a particular treatment protocol to treat a condition. Each subject record of the set of subject records can be labeled (e.g., by a user) as having one of a positive responsiveness to the particular treatment protocol, a neutral responsiveness to the particular treatment protocol, or a negative responsiveness to the particular treatment protocol. The set of subject records can then be divided into three subsets (e.g., clusters).The first subset of subject records can correspond to subjects who had a positive responsiveness to the particular treatment protocol, the second subset of subject records can correspond to subjects who had a neutral responsiveness to the particular treatment protocol, and the third subset of subject records can correspond to subjects who had a neutral responsiveness to the particular treatment protocol. The cloud server 135 can convert each subject record of the first subset of subject records to a transformed representation according to the implementations described above. The cloud server 135 can also convert each subject record of the second subset of subject records to a transformed representation using the techniques described above. Finally, the cloud server 135 can convert each subject record of the third subject of the subject records to a transformed representation using the techniques described above. In some implementations, determining the predicted responsiveness of the new subject to the particular treatment protocol can include converting the new subject record of the new subject to a new transformed representation. The new transformed representation can be compared to the transformed representation of each cluster or subset of subject records in a domain space (e.g., Euclidean space). If the new transformed expression is closest to the centroid of the transformed expression associated with the first subset, the new subject is predicted to have a positive responsiveness to the particular treatment. If the new transformed expression is closest to the centroid of the transformed expression of the second subset, the new subject is predicted to have a neutral responsiveness to the particular treatment. Finally, if the new transformed expression is closest to the centroid of the transformed expression of the third subset, the new subject is predicted to have a negative responsiveness to the particular treatment protocol. The centroid may be a multidimensional average of the transformed expression associated with the subset. The cloud server 135 may cause the estimated responsiveness metrics of the subset of the set of rules and the set of subjects to be displayed or presented in the treatment plan definition interface.
[0148] II.C. Presenting Treatment Recommendations with Correlated Efficacy Using Treatments Prescribed to Similar Subjects 4 is a flow chart illustrating a process 400 for recommending a treatment for a subject. The process 400 can be executed by the cloud server 135 to display the recommended treatments for the subject and the effectiveness of each recommended treatment on a user device associated with the medical entity. The recommended treatments can be identified using results of evaluating the effectiveness of treatments previously prescribed for similar subjects.
[0149] At block 410, the cloud server 135 receives input corresponding to a subject record that characterizes an aspect of the subject. The input is received from a user device associated with the entity. Additionally, the input is received in response to the user device selecting or identifying the subject record using an interface associated with an instance of a platform configured to manage a registry of subject records. The user device can access the interface by loading interface data stored on a web server (not shown) connected within the cloud network 130. The web server may be included in the cloud server 135 or may run on the cloud server.
[0150] At block 420, the cloud server 135 extracts a set of subject attributes from the subject record received at block 410. The subject attributes characterize an aspect of the subject. Non-limiting examples of subject attributes include any information found in an electronic health record, any demographic information, age, sex, ethnicity, recent or past symptoms, symptoms, symptom severity, and any other suitable information that characterizes the subject.
[0151] In block 430, the cloud server 135 generates an array representation of the subject record using the set of subject attributes. For example, the array representation is a vector representation of values contained in the subject record. The vector representation may be a vector in a domain space, such as Euclidean space. However, the array representation may be any numerical representation of values of data fields of the subject record. In some embodiments, the cloud server 135 may perform a feature decomposition technique, such as singular value decomposition (SVD), to generate values that represent the set of subject attributes of the array representation of the subject record.
[0152] In block 440, the cloud server 135 accesses a set of other array representations characterizing the plurality of other subjects. An array representation in the set of other array representations may be a vector representation of a subject record characterizing another subject (e.g., one of the plurality of other subjects).
[0153] At block 450, the cloud server 135 determines a similarity score representing the similarity between the array representation representing the subject and each of the array representations of the other subjects. For example, the similarity score is calculated using a function of the distance (in domain space) between the array representation representing the subject and the array representation representing the other subjects. As an illustrative and non-limiting example, the similarity score may be calculated using a range from "0" to "1", with "0" representing a distance above a defined threshold and "1" representing the array representations having no distance between them. By way of illustrative and non-limiting example only, the similarity score may be based on the Euclidean distance between the two array representations (e.g., vectors).
[0154] At block 460, the cloud server 135 identifies a first subset of the plurality of other subjects. A subject may be included in the first subset if the similarity score associated with the subject is within a predetermined absolute or relative range. Similarly, at block 470, the cloud server identifies a second subset of the plurality of other subjects. However, a subject may be included in the second subset if the similarity score of the subject is within another predetermined range.
[0155] In block 480, the cloud server 135 retrieves record data for each subject in the first and second subsets of the plurality of other subjects. The record data includes attributes included in the subject record that characterize the subject. For example, the subject record data identifies a treatment the subject received and the subject's responsiveness to the treatment. The responsiveness to the treatment can be represented by text (e.g., "the subject responded positively to the treatment") or a score indicating the degree to which the subject responded positively or negatively to the treatment (e.g., a score from "0" to "1" with "0" indicating negative responsiveness and "1" indicating positive responsiveness). In some cases, the treatment responsiveness can indicate the degree to which the subject responded positively to a treatment previously administered to the subject. For example, the treatment responsiveness may be numeric (e.g., a score from "0" to "10") or non-numeric (e.g., words assigned to represent responsiveness such as "positive," "neutral," or "negative"). In some cases, the treatment responsiveness for previously treated subjects may be user-defined. In other examples, the treatment responsiveness may be automatically determined based on the results of a test or measurement from a user. For example, treatment responsiveness may be determined automatically based on values contained in a blood test performed on the subject.
[0156] At block 490, the cloud server 135 generates an output that is presented to an interface on the user device. The output may, for example, indicate one or more treatment recommendations for the subject. The one or more treatment recommendations may be determined, for example, based on treatments received by other subjects in the first and second subsets, treatment responsiveness of the subjects in the first and second subsets, and differences between subject attributes of the subjects in the second subset and subject attributes of the subjects.
[0157] In some embodiments, the cloud server 135 determines that the subject and one of the subjects from the first or second subset are being or have been treated by the same medical entity. The cloud server 135 determines that the subject of the first or second subset and another subject are being or have been treated by different medical entities. The cloud server 135 can utilize different obfuscated versions of the subject's records via the interface. The cloud-based application can automatically provide the entity with different obfuscated versions of the records based on various constraints imposed on data sharing by data privacy rules of different jurisdictions. In some embodiments, the cloud server 135 identifies the first and second subsets of subject records by performing a clustering operation on a transformed representation of the set of subject records.
[0158] II.D. Automatic Obfuscation of Query Results from External Entities 5 is a flow chart illustrating a process 500 for obfuscating query results to comply with data privacy rules. The process 500 may be executed by the cloud server 135 as an execution rule to ensure that data sharing of subject records with external entities is compliant with data privacy rules. A cloud-based application may enable a user device to query the data registry 140 for subject records that satisfy the query constraints. However, the query results may include data records originating from the external entity. Thus, the process 500 enables the cloud server 135 to provide additional information about a procedure from the external entity to the user device while still complying with data privacy rules.
[0159] At block 510, the cloud server 135 receives a query from a user device associated with a first entity. For example, the first entity is a medical center associated with a first set of subject records. The query may include a set of symptoms associated with a medical condition, or any other information that constrains the query search of the data registry 140.
[0160] At block 520, the cloud server 135 queries the database using the query received from the user device. At block 530, the cloud server 135 generates a data set of query results corresponding to a set of symptoms and associated with a medical condition. For example, the user device submits a query for subject records of subjects diagnosed with hemophilia. The query results include at least one subject record from a first set of subject records (originating from or created at the first entity) and at least one subject record from a second set of subject records associated with a second entity (e.g., a different medical center than the first entity). Each of the subject records from the first set of subject records and the second set of subject records can include a set of subject attributes. The subject attributes can characterize any aspect of the subject.
[0161] At block 540, the cloud server 135 presents (e.g., utilizes or otherwise makes available) to the user device an entire set of subject attributes for the subject records included in the first subject record set as these records originate from the first entity. Presenting the subject records in their entirety includes making the set of attributes included in the subject records available by the user device for evaluation or interaction using an interface. At block 550, the cloud server 135 also or alternatively utilizes to the user device an incomplete subset of the set of subject attributes for each subject record included in the second subject record set. Providing the incomplete subset of the set of subject attributes provides anonymity to the subject because the incomplete subset of the subject attributes cannot be used to uniquely identify the subject. For example, providing the incomplete subset may include available 4 of the 10 subject attributes to anonymize the subject associated with the 10 subject attributes. In some embodiments, at block 550, the cloud server 135 utilizes an obfuscated set of subject attributes for each subject record included in the second subject. Obfuscating the set of attributes includes reducing the granularity of the information provided. For example, instead of utilizing a subject attribute of the subject's address, the obfuscated attribute may be a zip code or the state the subject lives in. Regardless of whether an incomplete subject or obfuscated subset is available, the cloud server 135 anonymizes the subjects associated with the subject record.
[0162] II.E. Chatbot Integration with Self-Learning Knowledge Base 6 is a flow chart illustrating a process 600 for communicating with a user using a bot script, such as a chatbot. The process 600 can be executed by the cloud server 135 to automatically link new questions provided by a user to existing questions in a knowledge base to provide responses to the new questions. The chatbot can be configured to provide answers to questions associated with symptoms.
[0163] In block 605, the cloud server 135 defines a knowledge base that includes a set of answers. The knowledge base may be a data structure stored in memory. The data structure stores text that represents a set of answers to defined questions. Each answer may be selectable by the chatbot in response to a question received from the user device during a communication session. The knowledge base may be automatically defined (e.g., by retrieving text from a data source and parsing the text using natural language processing techniques) or may be user-defined (e.g., by a researcher or physician).
[0164] In block 610, the cloud server 135 receives a communication from a particular user device. The communication corresponds to a request to initiate a communication session with a particular chatbot. For example, a doctor or a subject may operate the user device to communicate with the chatbot in a chat session. The cloud server 135 (or a module stored within the cloud server 135) may manage or establish the communication session between the user device and the chatbot. In block 615, the cloud server 135 receives a particular question from the particular user device during the communication session. The question may be a string of characters that is processed using natural language processing techniques.
[0165] In block 620, the cloud server 135 queries the knowledge base using at least some words extracted from the particular question. The words may be extracted from the string representing the particular question using natural language processing techniques. In block 625, the cloud server 135 determines that the knowledge base does not contain a representation of the particular question. In this case, the received question may be newly posted to the chatbot. In block 630, the cloud server 135 identifies other question representations from the knowledge base. The cloud server 135 may identify other question representations by comparing the question received from the user device with other question representations stored in the knowledge base. For example, if similarity is determined based on an analysis of the question representation using natural language processing techniques, the cloud server 135 identifies other question representations.
[0166] At block 635, the cloud server 135 retrieves the answer from a set of answers associated with other question representations in the knowledge base. At block 640, the answer retrieved at block 635 is sent to the particular user device as an answer to the received question, even if the knowledge base did not contain a representation of the received question. At block 645, the cloud server 135 receives an indication from the particular user device. For example, an indication may be received in response to the user device indicating that the answer provided by the chatbot responded to the particular question. In block 650, the cloud server 135 updates the knowledge base to include the representation of the particular question or a different representation of the particular question. For example, storing the representation of the question may include storing keywords included in the question in a data structure. The cloud server 135 may also associate the same or different representation of the particular question with more answers sent to the particular user device.
[0167] In some embodiments, the cloud server 135 accesses a subject record associated with a particular user device. The cloud server 135 determines multiple answers to a particular question. The cloud server 135 then selects an answer from the set of answers. However, the selection of the answer is based at least in part on one or more values included in the subject record associated with the particular user device. For example, the values included in the subject record can represent symptoms recently experienced by the subject. The chatbot may be configured to select an answer according to symptoms recently experienced by the subject. In some cases, the cloud server 135 can access a learning-to-rank machine learning model trained to predict the order of each answer in the set of answers. The learning-to-rank machine learning model can be trained using a training set of answers. Each answer in the training set of answers may be labeled with one or more symptoms and a relevance score for the symptoms. The relevance score can represent the relevance of the associated answer to a given symptom of the one or more symptoms. The relevance score may be user-defined or may be automatically determined based on certain factors, such as the frequency of words (e.g., words for symptoms) in the training answers. The set of training answers may differ from the set of answers used when the chatbot is operating in a production environment. The learn-to-rank machine learning model can learn how to order a set of answers (to be used in the production environment) with respect to relevance to the symptoms (detected from the subject profile) based on patterns (e.g., patterns between a training set of labeled answers for each symptom of one or more symptoms and associated relevance scores) learned by the learn-to-rank model. The chatbot can select an answer from the set of answers to be used in the production environment based on the predicted answer ordering. In some cases, each answer in the set of answers may be associated with a tag or code that indicates one or more symptoms associated with the answer. The cloud server 135 can compare a value representing a symptom recently experienced by the subject to the tag or code associated with each answer.
[0168] In some cases, the chatbot is configured to receive input regarding an injury or bleeding event experienced by a hemophilia subject. The input may include (for example) a photograph of the injury or bruise, an indication regarding the location of the injury, and / or characteristics of the bleeding (e.g., duration of incomplete or complete bleeding). The chatbot may be configured to return recommendations regarding whether the subject should self-perform a particular procedure, whether the subject should seek emergency medical attention, whether the subject should schedule an appointment with a caregiver, etc. In some cases, the chatbot selectively communicates a recommendation when the artificial intelligence engine (using the knowledge base) predicts that the subject should self-perform a particular procedure, seek emergency medical attention, and / or schedule an appointment, and the chatbot does not (recommendations that identify potential opposite actions).
[0169] The recommendation can be based on a knowledge base and / or a knowledge base. In some cases, the knowledge base is configured to generate an output based on an input. For example, the knowledge base can include a pharmacokinetic model defined for a given population (e.g., subjects with hemophilia, subjects with a particular type of hemophilia, and / or subjects with a particular severity of hemophilia) that can be updated and / or configured based on data related to the particular subject. The updating and / or configuration can be performed based on (for example) whether a prediction (e.g., that an abnormal bleeding event is likely not to occur) was accurate.
[0170] III.F. Module to facilitate selection and monitoring of hemophilia treatments 7 illustrates an interaction system 700 for facilitating hemophilia treatment selection and evaluation. The interaction system 700 includes a user device 705 that can be used, operated, and / or owned by a subject diagnosed with hemophilia. The user device 705 can include (for example) a smartphone, a smart wearable device, a mobile electronic device, etc.
[0171] The user device 705 may include one or more processors 710, which may include one or more single-core, multi-core, dual-core, quad-core, hexa-core, or octa-core processors. Each of the one or more processors may have a clock speed greater than (for example) 50 MHz, 100 MHz, 500 MHz, 1 GHz, 1.5 GHz, or 2.5 GHz. The one or more processors 710 may be included in a system-on-chip (SoC) of the user device 705. The one or more processors may include one or more central processing units and / or one or more graphical processing units.
[0172] The user device 705 may include one or more storages 715, which may include one or more volatile memories (e.g., RAM), an internal storage chip, and / or one or more non-volatile memories (e.g., ROM, such as flash memory). Data stored in the one or more storages 715 may include code that, when executed by at least one of the one or more processors 710, causes a set of operations to be performed. Alternatively, data stored in the one or more storages 715 may include non-executable information (e.g., that can be retrieved and / or used by executable code).
[0173] The one or more storages 715 can store an operating system 716, a hemophilia application 717, one or more models 718, and / or subject data 719. The operating system 716 can manage the use of computing resources (e.g., processing cycles and memory usage) and / or schedule tasks (e.g., generated by the execution of applications). The operating system 716 can be a single-tasking operating system or a multitasking operating system. The operating system 716 can include versions of (e.g.,) Windows, Linux, iOS, Android, macOS, etc.
[0174] The hemophilia app 717 may include code for facilitating collection of data related to the selection and / or evaluation of a subject's hemophilia treatment using the user device 705 and / or for facilitating presentation of hemophilia-related information to the subject via the user device 705. The collected data may identify (for example) the occurrence of a bleeding event, one or more symptoms of a bleeding event, identification of the time to receive treatment and / or administration, activity level, exercise engagement (or planned engagement), sleep time, weight, pulse, heart rate, and / or altitude.
[0175] The hemophilia app 717 can identify one or more models 718 to be used to predict a subject's current or future factor level, factor activity, and / or bleeding event risk. Each model 718 can include one or more machine learning models trained using training data corresponding to subjects associated with the user device 705 and / or other subjects. For example, the training data used to train a given model can include, for each of a set of other subjects (having a particular type or hemophilia), demographic information (e.g., weight and age), the measured activity level of a given factor (e.g., as measured by using an aPTT-based one-stage clotting assay or a two-stage chromogenic activity assay) at each of a set of time delays relative to the time a particular treatment (e.g., factor concentrate or recombinant factor) was administered, and the administration of the particular treatment administered. In some cases, the one or more models 718 include (for example) machine learning models and / or pharmacokinetic models that predict the concentration of an active agent at one or more time points, the level of factor activity at one or more times, the clotting time or clotting propensity at one or more times, the time when the clotting propensity falls below or reaches a threshold value, and / or the time when the active agent concentration falls below or reaches a threshold value.
[0176] Another example of training data that can be used to train a given model associated with a particular treatment may include, for each of a set of other subjects (having a particular type or hemophilia), demographic information (e.g., weight and age), details of the treatment received (e.g., dosing), an indication of whether any bleeding events occurred between successive treatments, the time relative to the most recent treatment at which any bleeding event occurred, treatment-related levels measured when treating any bleeding event (e.g., factor activity levels, clotting times, aPTT times, etc.), any triggering circumstances of any bleeding event, etc.
[0177] The model 718 can include (for example) an AI model and / or a machine learning model. The AI and / or machine learning model can include (for example) a pharmacokinetic model that uses rate constants (e.g., absorption, distribution, metabolism, and / or excretion) to model the kinetics of a therapeutic agent (e.g., a factor concentrate, recombinant factor, or non-factor therapy) to generate a predicted level of the therapeutic agent over a period of time or a specific time. The pharmacokinetics can include a compartmental model, a non-compartmental model, or a physiological model.
[0178] In a compartment model, kinetic variables can be defined for each compartment. Each compartment corresponds to a tissue and / or organ that is assumed to be kinetically homogeneous. Compartment models further generally assume that the transfer of a therapeutic agent between compartments can be represented using first-order kinetics. Once the kinetic variables (e.g., rate constants) have been identified, a pharmacokinetic output can be generated using (for example) nonlinear regression analysis. In some cases, the compartment model includes a central compartment representing plasma and one or more peripheral compartments. Rate constants can represent the flow of plasma between the central component and each peripheral component. The results of the model can be interpreted and / or the model can be configured to indicate the absorption of the active agent by the system, the volume of distribution (plasma volume required to hold the amount of active agent predicted to be present at a given time), and / or the clearance of the active agent. The half-life of the therapeutic agent can further be calculated.
[0179] Non-compartmental models do not use variables that correspond to estimated physiological or anatomical properties. Rather, non-compartmental models use fixed equations (e.g., algebraic equations). Thus, non-compartmental models generally produce faster and more reliable results compared to compartmental models. However, non-compartmental models are typically inferior to compartmental models in accurately predicting variability across subjects and situations. Non-compartmental model parameters are generally determined based on time-series drug concentration data.
[0180] As mentioned above, each compartment in the compartment model does not have to correspond to a physiological subject (e.g., an organ or tissue). A physiological model, on the other hand, incorporates anatomical meaning and models blood flow (and flow of therapeutic agent) through various parts of the body, absorption of therapeutic agent in various body parts, etc. Some tissues can be grouped together if they are rapidly equilibrating (e.g., lung, liver, brain and kidney), and the model can exclude certain human features (e.g., bone) if the absorbance of the therapeutic agent is assumed or known to be low. A physiological model is parameterized using empirical data corresponding to observed physiology. A physiological model can estimate the concentration of therapeutic agent in various organs or tissues using the predicted size, perfusion rate, and tissue-plasma partition coefficient of the individual organs or tissues.
[0181] For any of these types of pharmacokinetic models, parameters (e.g., corresponding to rate constants, kinetics, tissue volumes, etc.) can depend on subject characteristics (e.g., age, weight, sex, type of hemophilia, comorbidities, etc.). In some cases, relevant subject characteristics can be determined quite practically and accurately. In some cases, at least some of the subject characteristics are not available. The model can infer the missing subject information and / or use a framework (e.g., a Bayesian framework) to estimate parameter dependencies based on population-level data. In some cases, Markov chain Monte Carlo simulation can be used to estimate the posterior distributions of population and subject-specific parameters, and the covariate model can identify systematic variation that is explainable by measurable subject characteristics (e.g., age, height, type of disease).
[0182] In some cases, a modified generative adversarial network (GAN) can be used to predict relevant subject data. Specifically, a population distribution across one or more variables (e.g., rate constants, kinetics, etc.) can be identified. The generator network can identify a transformation of the distribution along one or more dimensions. The transformation can be defined based at least in part on subject-specific characteristics (e.g., weight, age) and / or data (e.g., one or more subject-related rate constants, kinetic variables, etc.). A sampling technique (e.g., Monte Carlo) can sample from the transformed distribution, and a discriminator network can predict whether the sample corresponds to a population or subject. The accuracy of the prediction can be fed back to the generator network until a threshold accuracy is obtained or a threshold number of iterations have been performed. The transformation can then be used to estimate subject-specific metrics (and / or their one-dimensional or multidimensional distributions) representing the pharmacokinetics corresponding to individual subjects. This approach can facilitate the use of limited and / or a small number of subject-specific variable values to generate subject-specific distributions that can more fully represent biological activity. Sampling techniques (eg, Monte Carlo methods) can sample from a distribution to generate data used to train another model (eg, a pharmacokinetic model or a neural network).
[0183] In some cases, a model 718 is used that includes a neural network. The neural network can include a feedforward neural network. The feedforward neural network can receive an input data set that includes (for example) demographic information, laboratory data, diagnostic data, physiological data, and / or other metrics. The feedforward neural network may be trained using population data to learn one or more physiological metrics (e.g., half-life, time constants, scaling factors).
[0184] In some cases, the feedforward network further receives inputs indicating when a treatment was administered to a subject (and, e.g., the dose administered), and the feedforward neural network can output a prediction of the time when the clotting tendency (e.g., corresponding to the clotting time in an assay) will fall below or reach a particular threshold, the predicted current clotting tendency, the predicted current level or activity of the active ingredient, and / or the predicted time when the level or activity of the active ingredient will fall below or reach a particular threshold. In such cases, the feedforward neural network may be trained to generate predictions for a particular type of treatment, or additional inputs may identify the type of treatment.
[0185] The feedforward network can include at least one, at least two, or at least three hidden layers, each of which can include (for example) at least 10, at least 15, at least 25, or at least 30 nodes.
[0186] Within the one or more storages 715, the stored model data can include model architecture, predefined unlearned hyperparameters, and / or learned parameters. The stored model data can include variables (e.g., one or more unlearned hyperparameters used by a machine learning model, or one or more unlearned parameters used by a pharmacokinetic model) that include or are determined based on some or all of the subject data 719 (data corresponding to and / or characterizing a particular subject, which may be a user of the user device 705). The hemophilia app 717 can configure the model architecture and data routing routines to define inputs to be provided to the model based on some or all of the subject data 719. In some cases, the hemophilia app 717 selects and / or trains a model using some or all of the subject data 719.
[0187] The subject data 719 may include demographic data (e.g., identifying age, sex, race, and / or place of residence), medical history data (e.g., identifying a diagnosis of type of hemophilia, a diagnosis of severity of hemophilia, a diagnosis of one or more other diseases, and / or a characterization of one or more previous bleeding symptoms), treatment history data (e.g., identifying the type of current hemophilia treatment, the date and / or time of treatment administration, the administration of treatment administration, the type and date of previous hemophilia treatment, and / or any type of adverse event experienced with current or previous hemophilia treatment), symptom data (e.g., identifying joint damage and / or movement disorders), and / or genetic data (e.g., identifying one or more mutations, such as one or more X-chromosome mutations and / or one or more chromosome-4 mutations). In some cases, the genetic data includes a severity classification of each hemophilia-associated mutation (e.g., classified based on International Society on Thrombosis and Haemostasis criteria) and / or some mutations associated with the subject characterized as having a particular severity level (e.g., severe). Subject data 719 may further include laboratory results (eg, results of blood tests, assays, mixing tests or other tests identified herein) or processed versions thereof.
[0188] The subject data 719 may further include dates and / or times of one or more scheduled future treatments. In some cases, the hemophilia app 717 may be configured to process such that the model generates a prediction corresponding to the planned treatment time, such as a predicted factor level, active ingredient concentration, clotting tendency, or clotting time (e.g., given a particular hypothetical situation, e.g., bleeding or assay related situation). In some cases, the hemophilia app 717 may be configured to predict a time at which the predicted factor activity level, active ingredient concentration, clotting tendency, or clotting time will fall below a threshold (e.g., given a particular hypothetical situation, e.g., bleeding or assay related situation) and determine whether the predicted threshold crossing time is before or after the planned treatment time. In some cases, the predicted time corresponds to a time at which the probability of threshold crossing (e.g., having a factor level below a lower factor level threshold) exceeds a probability threshold (e.g., 35%, 50%, 75%, etc.).
[0189] The user device 705 may include one or more user interface components 720 configured to receive user input and / or present output. In some cases, at least a portion of the subject data 719 includes data received as input via the user interface component 720 and / or data generated based on input received via the user interface component 720. For example, the user input may identify the time and type of injury (e.g., scraping, cutting, etc.), and whether the result was perceived by the subject to be normal or abnormal (e.g., with respect to clotting time). As another example, the user input may identify a hemophilia-related symptom (e.g., joint pain), the severity of the symptom, the approximate time of onset of the symptom, and / or whether the user recalls a potentially triggering event prior to the onset of the symptom. In some cases, at least a portion of the operations of the hemophilia app 717 (e.g., identifying the interface and / or data to be presented, transmitting data to other devices, defining one or more thresholds for alarm alerts, etc.) are performed upon receiving one or more particular types of input via the user interface components.
[0190] In some cases, the hemophilia app 717 triggers one or more outputs to be presented via an output component of the user interface component 720. For example, a predicted factor level may be presented visually, and a notification of satisfaction of an alert condition (e.g., indicating that the predicted value will fall below a threshold value prior to the next scheduled procedure, or that the rate of change in the predicted value exceeds a change threshold) may be presented via an audible or tactile stimulus.
[0191] The one or more input components of the user interface components 720 may include (for example) a touch screen, a keyboard, a track pad, a mouse, and / or a microphone. The one or more output components of the user interface components 720 may include (for example) a display screen, a speaker, and / or a haptic output.
[0192] The user device 705 may include one or more sensors. The sensors may include one or more motion sensors 725 and / or one or more physiological sensors 730. The motion sensors 725 may include (for example) an accelerometer and / or a gyroscope. The user device 705 may include one or more applications (e.g., the hemophilia app 717 and / or another app) configured to convert the accelerometer data and / or gyroscope data to predict the number of steps taken by the user over a recent period and / or the athletic activity in which the user is engaged (e.g., running, walking, standing / sitting). The physiological sensors 730 may include a heart rate or pulse monitor that may operate by outputting light (e.g., via an LED and / or having a known frequency profile) and monitoring data collected via an optical sensor (e.g., to detect intensity in one or more frequency bands). Absorption of certain frequencies (e.g., green light) can vary based on slight blood volume differences that occur throughout a heart cycle, such that heart rate or pulse can be calculated using the light sensor data (e.g., via an application on the user device 705 and / or by the hemophilia app 717). In some cases, the heart rate or pulse can further inform a prediction of recent user activity.
[0193] Other sensors may include (for example) a GPS receiver that receives GPS signals. An application on the user device 705 (e.g., the hemophilia app 717 or another app) may use the GPS signals to infer the user's location (e.g., GPS coordinates, address, city, state, country, etc.). Other sensors may also or alternatively include a barometer. Measurements collected via the barometer and / or signals received by the GPS receiver may be used to infer the user's altitude.
[0194] In some cases, the hemophilia app 717 may be configured to receive and process input data (e.g., via one or more usage interface components 720) and / or input data from one or more sensors 730 to communicate hemophilia-related information. The hemophilia-related information may include (for example) an identification and / or characterization of a bleeding event and / or a treatment response. The hemophilia-related information may include (for example) a photograph of a wound, bleeding, rash, or other response (e.g., collected using a camera of the user device 705). The hemophilia-related information may also or alternatively include textual information (e.g., received via a touch screen, cursor, keyboard, or speaker, receiving audio commands) characterizing the bleeding event or injury. The hemophilia app 717 may convert the hemophilia-related information into one or more communications.
[0195] The communication can be sent to a care provider system and / or a central server, for example, the communication can report and / or characterize the injury and / or abnormal bleeding event (e.g., can include identification of any triggers of the bleeding event, a photograph of the area of the body that experienced the injury or bleeding event, an identification as to whether a wound is present, and / or an identification as to the duration of the bleeding). The hemophilia application 717 executing on the subject device 705 may process the input (e.g., adjusting the resolution or size of the image to a predetermined resolution or predetermined size; and / or generating key-value pairs based on various text inputs) and send the processed data to the caregiver's device. The same or a different application executing on a provider system (e.g., provider system 745) can process the communication and determine whether to (e.g.) present an alert, initiate an update to the subject record, and / or facilitate the generation and display of a presentation (e.g., identifying at least a portion of the hemophilia-related information).
[0196] In some cases, the hemophilia app 717 may access and / or process at least some of the sensor data. The hemophilia app 717 may store the raw and / or processed sensor data as part of the subject data 719. For example, the hemophilia app 717 may use accelerometer and / or heart rate data to predict whether the user engaged in minimal activity (e.g., primarily sedentary), low intensity activity, or high intensity activity for various portions of an individual day, and store such predictions as the subject data 719. As another example, the hemophilia app 717 may use heart rate data to predict what times the user was asleep on a day-by-day basis, and store the sleep time predictions. As yet another example, the hemophilia app 717 may use GPS and / or barometer data to predict the user's location and / or altitude.
[0197] In some cases, the hemophilia app 717 processes the sensor data and other subject data to predict one or more relationships between the sensor data (or a processed version thereof) and hemophilia-related occurrences. The processing may include performing multidimensional analysis or may use machine learning models to predict how exercise intensity or user exercise will affect hemostasis, if any (e.g., as indicated by whether a bleeding event was normal or abnormal and / or whether spontaneous bleeding occurred). For example, it may be determined that motion and / or movement characteristic of high intensity exercise will temporarily increase the probability of a bleeding event (although the magnitude and / or duration of such increase may be subject specific). The evaluation may further account for time since last treatment and / or recent predicted active ingredient levels, clotting tendencies, etc. For example, the evaluation may predict how exercise intensity and / or user exercise will temporarily affect one or more time constants of a pharmacokinetic model. As another example, the evaluation may predict how the output of a pharmacokinetic model will be post-processed (e.g., temporally post-processed) for transient changes in hemostasis. Post-processing can include (for example) multiplying the results (e.g., risk of abnormal bleeding, predicted clotting time, recommended time interval to receive next treatment, etc.) by a value, adding or subtracting amounts to the results, and / or transforming the results using non-linear functions. In some cases, the effect of post-processing is limited to the prediction for which the post-processing is performed. In some cases, the model implements an iterative approach whereby the processing of successive time points depends on values from the previous time point. Thus, post-processing can have a lasting effect.
[0198] In some cases, the hemophilia app 717 processes the sensor data to select a model used to generate a hemophilia-related prediction for the subject. Different model selections can include selecting models trained using different data, trained using different loss and / or objective functions, having different fixed hyperparameters, and / or having different architectures. For example, a default model that the hemophilia app 717 selects for a subject can include a model that prioritizes accurately predicting levels of active treatment agents. Meanwhile, if the hemophilia app 717 infers that the user is engaged in high intensity activity, it may transition to a model that prioritizes accurately predicting the occurrence of abnormal bleeding events. Transitions between models can be facilitated by defining dynamic variable correspondences between models. For example, each of the multiple models can be configured to receive a predicted active ingredient level (which may have been generated, e.g., by processing of a previous time step). Other inputs can be objective and / or fixed (e.g., variables based on or including the subject's physical attributes and / or demographic attributes and / or sensor variables). Thus, when switching from one model to another, the input variables of the other model can be readily utilized. In some cases, post-processing is implemented to further smooth and / or filter the predictions produced by the two models.
[0199] The user device 705 further includes one or more communication components 735 that can facilitate receiving communications from and / or transmitting communications to one or more other devices. The communication component 735 may be configured to transmit signals via a wired or wireless connection. The communication component 735 can include (for example) a receiver, a transmitter, and / or a transceiver. The communication component 735 can include one or more antennas (e.g., for transmitting or receiving radio frequency signals).
[0200] It will be appreciated that in some cases, the user device 705 may be paired with another user device. The user device 705 and the paired user device may then collectively include the components shown in FIG. 7 to facilitate interaction with both devices to facilitate operation of the hemophilia app 717. For example, a first user device may include a wearable device including all of the illustrated components, and a second user device may include a smartphone including all of the illustrated components except for the physiological sensor 730. The two devices may communicate over a network (e.g., Bluetooth or WiFi) such that the second user device receives data corresponding to the physiological measurements. The hemophilia app 717 operating on the first device may coordinate the associated data collection, initial processing (e.g., to convert optical signals to heart rate) and transmission of the processed sensor data, and the hemophilia app 717 operating on the second device may coordinate the selection, configuration and execution of one or more models to process the pre-processed sensor data and other associated subject data 719.
[0201] The interaction system 700 may further include one or more laboratory systems 740, one or more care provider systems 745, and a central artificial intelligence system 750. Two or more of the user devices 705, the laboratory systems 740, the care provider systems 745, and / or the central artificial intelligence system 750 may communicate via one or more networks (e.g., the Internet).
[0202] The laboratory system 740 can be associated with one or more medical laboratories and / or testing facilities. The laboratory system 740 can include a system configured to generate, evaluate, access, and / or communicate (for example) one or more laboratory test results. For example, the laboratory system 740 can be configured to receive a sample (e.g., a blood sample, a urine sample, or a saliva sample) from a subject and process the sample to generate a result. Processing the sample can include (for example) performing techniques disclosed herein (e.g., the processes disclosed in Section II), performing assays, and / or running tests. Processing can be performed (for example) during a diagnostic phase, as part of disease management / monitoring, in response to a response to a bleeding event, prior to treatment administration, and / or at one or more defined times relative to treatment administration. The one or more results generated in and / or by the one or more laboratory systems can include results, metrics, or variables of the type disclosed herein (e.g., in Section II), such as clotting times, factor activity levels, treatment concentrations, admixture test results, platelet counts, etc. The laboratory system 740 may transmit one or more results, one or more identifications of the evaluations performed, and / or one or more subject identifications (e.g., name, social security number, etc.) to the user device 705 and / or one or more care provider systems 745.
[0203] The care provider system 745 can be associated with (for example) a doctor, clinic, nurse, urgent care, hospital, surgeon, or other health care provider. The provider system 745 can receive (e.g., via a user interface) input corresponding to an assessment made by the care provider, which can correspond to an identification of the subject's condition, hemophilia risk factors, diagnosis (e.g., specific type and / or severity of hemophilia), prognosis, prescribed preventive treatment, prescribed symptomatic treatment, etc. The identification of the treatment can identify the type of treatment, administration of the treatment, a schedule of the treatment (e.g., treatment intervals and / or schedule of loading doses and maintenance doses), and / or monitoring to be performed incidental to the treatment.
[0204] In various instances, a caregiver system 745 associated with the subject's caregiver and / or a user device 705 used by the subject provides input to the central artificial intelligence system 750 to initiate enrollment of the subject and generation of one or more data records for the subject. The data records may include (for example): ● Information about the subjects (e.g. demographic information, residential information, education level, marital status, occupation, etc.); ● Hemophilia diagnosis information (e.g., type and / or severity of hemophilia, whether diagnosis is suspected or confirmed, date of diagnosis); ● one or more comorbidities (e.g., diagnosis of asthma, autoimmune disease, cardiomyopathy, cardiovascular disease, cerebrovascular disease, congestive heart failure, COPD, coronary artery disease, deep vein thrombosis, diabetes, HIV, history of thrombotic microangiopathy, history of thrombotic events, history of intracranial hemorrhage, hepatitis, hypertension, myocardial infarction, pulmonary embolism); ● One or more previous surgeries (e.g., identifying the type of surgery, date of surgery, and whether abnormal bleeding occurred); ● Laboratory and / or medical test information (e.g., one or more factor levels, inhibitor titer level, DNA variants detected, basophil count, eosinophil count, hemoglobin count, INR, lymphocyte count, monocyte count, neutrophil count, NK cell count, platelet count, prothrombin time, partial thromboplastin time, red blood cell count, T cell count, white blood cell count, total protein count, albumin count, SGOT / AST count, SGPT / ALT count, gamma-GT count, bilumbin count, alkaline phosphatase count, calcium count, urea count, uric acid count, creatine count, amylase count, vitamin D level, anti-JC virus, anti-HCV, anti-HIV, anti-varicella, HIV antigen, urine JC virus DNA); ● Imaging data (e.g., including scans or summaries from one or more CT, MRI, PET, or X-rays); ● Treatment information (e.g., identifying each on-demand hemophilia procedure, each prophylactic hemophilia procedure, each surgery-related hemophilia procedure, each elective bleeding prophylaxis procedure, each immune tolerance induction procedure, each nonhemophilia procedure, the date of each procedure, the route by which each procedure was administered, the reason for each procedure, and any adverse events associated with each procedure); • secondary condition information (e.g., identifying the date associated with the occurrence of each adverse condition, the type of adverse condition, the action taken in response to the adverse condition, and the outcome corresponding to the action taken for the adverse condition); ● Joint condition information (e.g., pain and mobility in each of one or more joints); • Health score information (e.g., Hemophilia Community Health Score, quality of life score); ● Fault information (e.g., indicating whether an assisted mobility device is used, indicating which assisted mobility device is used, indicating when an assisted mobility device started to be used), and / or ● Bleeding event information (e.g., identifying the date of each bleed, the number of bleeds the subject experienced at one time or in total, the location of each bleed, the severity of each bleed, any cause of each bleed, whether each bleed was spontaneous or in response to injury, the results of any inhibitor tests performed at one time in relation to the bleed, whether the subject was hospitalized in response to the bleed and / or for how long, any treatments performed for each bleed, and any persistent disabilities or medical problems attributable to each bleed).
[0205] The data record may further indicate to which of one or more actions (e.g., limited data sharing and / or limited data processing) the subject consented. For example, the provider may use the provider system 745 to upload a consent form (e.g., signed by the subject) that identifies the particular type of data processing to which the subject consented. The data record may further store predictions generated for the subject using one or more types of data processing (e.g., using machine learning models). Different entities may be permitted to access at least a portion of the subject record, although there may be differences as to which portions are made accessible. For example, the subject's care provider and one or more operators of the remote computing system may have full access, while the laboratory and the subject may have access to select fields of information.
[0206] Each of the one or more laboratory systems and / or each of the one or more provider systems may include a computing system including (for example) one or more processors, one or more memories, one or more input / output components, and / or one or more communication components. The laboratory system 740 may further include a laboratory device that automatically determines one or more sample evaluation related metrics.
[0207] The central artificial intelligence system 750 may include a remote system (e.g., a cloud system) configured to perform processing to support the operation of the hemophilia app 717. In some cases, the entity that controls and / or manages the central artificial intelligence system 750 also controls, manages and / or owns the hemophilia app 717.
[0208] In some cases, the central artificial intelligence system 750 includes a cloud computing system. The central artificial intelligence system 750 can include one or more servers and / or one or more virtual servers.
[0209] The central artificial intelligence system 750 can include one or more communication components 755, which can include enabling wireless communication (e.g., via the Internet) with one or more other devices (e.g., user device 705, laboratory system 740, and / or provider system 745). The communication components 755 can include one or more transmitters, one or more receivers, and / or one or more transceivers. The communication components 755 may include one or more antennas.
[0210] The central artificial intelligence system 750 may include one or more processors 760, which may include one or more single-core, multi-core, dual-core, quad-core, hexa-core, or octa-core processors. Each of the one or more processors 760 may have a clock speed greater than (for example) 50 MHz, 100 MHz, 500 MHz, 1 GHz, 1.5 GHz, or 2.5 GHz. The one or more processors 760 may include one or more central processing units and / or one or more graphical processing units.
[0211] The central artificial intelligence system 750 may include one or more data storages 765, which may include one or more volatile memories (e.g., RAM), embedded storage chips, and / or one or more non-volatile memories (e.g., ROM, such as flash memory). The data stored in the one or more storages 765 may include code that, when executed by at least one of the one or more processors 760, causes a set of operations to be performed. Alternatively, the data stored in the one or more storages 765 may include non-executable information (e.g., that can be retrieved and / or used by executable code).
[0212] Storage 765 may store population-level training code 766, one or more artificial intelligence codes 767, one or more loss functions 768, subject-specific tuning code 769, one or more pre-processing functions 769, one or more post-processing functions 770, model usage data 772, and / or expression data 773. The central artificial intelligence system 750 may have access to more extensive hemophilia-related data (corresponding to multiple subjects) and / or a more powerful processor compared to the user device 705. Thus, it may be advantageous for the central artificial intelligence system 750 to have dominant or complete control over the training of the population-level models and / or partial or complete control over defining the subject-specific models.
[0213] Population-level training code 766 can control the training and / or definition of one or more population-level artificial intelligence (AI) models that generate hemophilia-associated predictions. Each population-level artificial intelligence model can be generated based on training data corresponding to multiple subjects and can be generated for potential use with multiple subjects. In some cases, the population-level artificial intelligence models can be trained using the set of records according to some or all of blocks 210-260 from process 200 shown in FIG. 2 and / or some or all of process 500 shown in FIG. 5.
[0214] Population-level AI models can include pharmacokinetic models, machine learning models, any type of model identified herein, or other models. For example, population-level artificial intelligence models can include compartmentalized models, non-compartmentalized models, physiological models, modified generative adversarial networks, neural networks, machine learning models using a Bayesian framework, models using Markov chain Monte Carlo simulation, etc.
[0215] A population-level AI model can be defined for a particular type of use, which can correspond to attributes of the data used to train the model. For example, a given population-level AI model can correspond to a particular type of hemophilia (e.g., hemophilia A, hemophilia B, or hemophilia C), a particular severity (e.g., mild, moderate, or severe), a particular current treatment (e.g., a particular factor concentrate, a particular recombinant factor, emicizumab, no prophylactic treatment, etc.), a particular subject's mobility (e.g., fully ambulatory without assistance, ambulatory without assistance up to a defined threshold distance, ambulatory without assistance on one side, ambulatory with bilateral assistance or walker, wheelchair-assisted), a particular type of genetic mutation, etc.
[0216] The population-level training codes 766 may identify the training data to use to train a particular AI model (of one or more AI models 767) by identifying data for which subjects have given permission for the data to be used for training purposes and / or data for which the constraints of the AI model are satisfied. For example, for a population-level AI model corresponding to hemophilia A subjects being treated with emicizumab, the population-level training codes 766 may identify subjects associated with hemophilia A diagnosis data, emicizumab treatment data, and an indication of permission to use the data for model training.
[0217] In some cases, the training data may be further selected to represent the predicted subject group for which the AI model is expected to be used. For example, the majority (e.g., about 90%) of people diagnosed with hemophilia A or B are male. Thus, the training data set may be defined to correspond to a similar male:female ratio that occurs for a given set of AI model constraints. On the other hand, hemophilia C is fairly evenly affected in males and females. Thus, the training data set may be defined to represent males and females equally. While this example is for a single attribute, the prevalence of various combinations of attributes may be further considered. In some cases, a classifier may be used to determine whether the subject attributes represented in the training set can be differentiated from the attributes represented in a representative set of potential user subjects.
[0218] In some cases, different AI models are defined for different subject groups. For example, multiple AI models can be generated that relate to a particular type and severity of hemophilia and a particular treatment. Each of the multiple AI models can be associated with subjects that correspond to one or more hemophilia-independent attributes (e.g., a particular geographic region, a particular age group, a particular gender) and / or one or more hemophilia-related subject attributes (e.g., gene mutations for which inhibitors have been developed and have hemophilia-related joint damage). The subject groups can be identified by performing (for example) component analysis (e.g., principal component analysis or independent component analysis), mutual information analysis, clustering analysis, etc. For example, component analysis can indicate the degree to which each of a set of subject attributes explains the variability of a hemophilia-related metric (e.g., whether a bleeding event occurs, time constants for factor levels or active ingredient concentrations, etc.). For each of the two most variable attributes, two attribute-specific subgroups can be defined (e.g., based on thresholds or binary criteria) such that a total of four groups are defined. As another example, mutual information analysis can be performed to identify subject attributes that most substantially reduce the entropy of the hemophilia-related metric, and for each attribute for which the mutual information exceeds a threshold, the attribute can be used to define subject groups. As yet another example, clustering analysis can be used to cluster representations of subjects independent of the hemophilia-related metric that corresponds to the output of the target AI model being trained, and a different AI model can be generated for each cluster. As yet another example, clustering analysis can be used to cluster one or more hemophilia-related metrics, and subject attributes separately associated with the clusters can subsequently be identified and used to define subject groups.
[0219] The training data may include static or deterministic data (e.g., subject age, race, age at hemophilia diagnosis, etc.), or data corresponding to a particular time point or particular time period (e.g., type of first hemophilia diagnosis, severity of hemophilia as of a given date, extent of joint damage as of a given date, administration of a given treatment beginning at a pre-identified time frame, etc.). Such static, deterministic, and / or snapshot data may correspond to input data variables. The training data may further include data corresponding to one or more subsequent time points and / or time periods (e.g., indicative of the number of bleeding events within a two-year period after initiating a given treatment, the extent of joint damage at three years after initiating a treatment, the number of adverse event thrombus occurring within a given period, in vivo active ingredient concentration at a set time relative to treatment administration, etc.). This additional data may be used as a true label and / or outcome during training.
[0220] In some cases, the training data is used to define various variables and / or parameters using fitting techniques. For example, the time constant of the decay of the active ingredient concentration can be calculated by fitting a given algorithm (e.g., exponential decay, regression, etc.). In some cases, the training data is used to define various variables and / or parameters using learning techniques that operate according to one or more loss functions 768. The loss functions can identify which aspects of the prediction should be optimized and how they should be optimized relative to the corresponding true labels. For example, for a given input dataset, the AI model may determine whether a given active ingredient concentration is at level c awhere the factor VIII activity level is a and the probability of a bleeding event occurring can be estimated as p. The loss function can indicate that for training, the important variable is p, and a large penalty is imposed if a bleeding event occurs when the predicted probability of a bleeding event was low. The loss function can be linear, non-linear, dependent on one predictor, dependent on multiple predictors, etc. The loss function can indicate that the AI model should be trained to optimize predictions corresponding to (for example) whether a bleeding event occurs within a given time period (e.g., occurring at a given time compared to treatment administration), the severity of the bleeding event (e.g., occurring at a given time compared to treatment administration), when the bleeding event occurs (e.g., compared to treatment administration), whether the subject continues on treatment, the occurrence of an adverse event, the occurrence of a blood coagulation adverse event, the concentration of the active ingredient at a given time point (e.g., compared to treatment administration), the activity level of the factor at a given time point (e.g., compared to treatment administration), the clotting time at a given time point (e.g., compared to treatment administration), the assay result at a given time point (e.g., compared to treatment administration), subject mobility, change in subject mobility (e.g., over a given time period), and / or combinations thereof.
[0221] The subject-specific tuning code 769 can be configured to perform AI model selection, training, and / or configuration based on data corresponding to a particular subject. In some cases, a set of AI models is selected in which each model-associated condition of use is satisfied for a particular subject. For example, the conditions of use can identify a particular type of hemophilia, a particular treatment, a particular inhibitor status (e.g., with or without inhibitors), etc. The set of AI models can include (for example) different types of AI model architectures and / or different parameters. The subject-specific tuning code 769 can select an AI model from among the set of AI models to be used for a particular subject.
[0222] The selection can be made (for example) by determining, for each AI model of the set of AI models, the degree to which a particular subject's attribute is represented in the training data used to train the AI model and / or the data used to generate the fit used by the AI model, and selecting the AI models for which such a metric of representation exceeds a threshold or is highest across the set of AI models. To determine the representation, an attribute similarity metric can be determined between the particular subject and each subject in the training data. For example, the attribute similarity metric may be calculated as a weighted sum of whether and / or the degree to which each of a plurality of particular attributes matches an attribute of the training data subject. For example, with respect to age, the similarity metric can depend on the inverse of the age difference between the particular subject and the subjects of the training data (e.g., potentially capped to avoid the possibility of infinite values). As another example, with respect to age, a plurality of age difference ranges may be defined and associated with corresponding point values that contribute to the similarity metric. The weights used to calculate the similarity metric can be determined based on (for example) the degree to which the attribute is associated with predictive accuracy in the test data (e.g., a higher weight is assigned if the difference between the attribute values of the correct predictions and the attribute values of the inaccurate predictions is greater and / or more significant), the degree to which the attribute is unique across a population of hemophilia subjects (e.g., a higher weight is assigned if the subject attribute is more unique), and / or the variability of the attribute in the training data (e.g., a higher weight is assigned if the variability of the attribute is low). In some cases, the similarity metric and / or weights are defined differently for each of the set of AI models (e.g., the scale of the similarity metric may remain the same to facilitate inter-model comparison). In some cases, the similarity metric and / or weights are defined in the same way for each of the set of AI models.
[0223] The selected AI model may correspond to (for example) the AI model associated with the highest or above threshold median, mean, mode, maximum or percentile similarity metric (across the metrics calculated for all pairs between a particular subject and each subject in the training data). Thus, in some cases, AI model selection may include selecting the AI model for which the "nearest neighbors" of a particular subject are represented in the training data.
[0224] In some cases, the test data can indicate whether and / or to what extent the AI model predictions match or correspond to the observed data. An AI model can be selected for a particular subject by determining, for each of a set of AI models, a similarity metric between the particular subject's attributes and the attributes of subjects represented in the test data for which the predictions were accurate (or more accurate), potentially compared to a similarity metric between the particular subject's attributes and the attributes of other subjects represented in the test data for which the predictions were inaccurate (or less accurate). The selected AI model may be the one associated with the highest or above threshold response to the subject associated with an accurate prediction (e.g., potentially associated with a response to the subject associated with an inaccurate prediction).
[0225] In some cases, the subject-specific tuning code 769 can perform subject-specific training. Such training can include using (for example) transfer learning, where an AI model trained using data associated with a set of subjects (which may be selected from among a set of AI models using techniques disclosed herein) is accessed and the parameters are used as part of the subsequent training. The subsequent training can be performed using data associated with the particular subject and / or other similar subjects. For example, for each of a plurality of other subjects, a similarity metric may be determined (for example, using techniques disclosed herein) based on attributes of the other subjects and attributes of the particular subject. Similar subjects can be identified as a subset of the plurality of other subjects associated with a similarity metric above a threshold and / or a particular number of other subjects associated with the highest similarity metric.
[0226] Transfer learning can include using (for example) one or more learned parameters (e.g., weights) as initial values or frozen values to be incorporated into the subject-specific AI model. The parameters can include (for example) one or more time constants, one or more weights associated with connections between nodes in a neural network, one or more sub-networks, and / or one or more layers. In some cases, the subject-specific AI model includes one or more additional filters, time constants, neural network layers, etc. that are not included in the population-level AI model and are trained based on data associated with the particular subject and / or other similar subjects. In some cases, the subject-specific AI model includes one or more different parameters (e.g., weights, time constants, etc.), functions, and / or neural network layers compared to those in the population-level AI model. The subject-specific adjustment code 769 can implement constraints and / or limits on one or more parameters, where the constraints are determined based on the corresponding parameter values in the population-level AI model.
[0227] In some cases, the subject-specific adjustment code 769 can generate a subject-specific workflow using one or more AI models. The workflow can include one or more pre-processing functions 770, one or more AI models 767, and / or one or more post-processing functions 771. The pre-processing functions can normalize, standardize, encode, categorize, filter, and / or otherwise process the data (for example) before feeding the data to the AI model. For example, the raw data can include a level of a clotting factor or a concentration of an active ingredient collected immediately prior to the previous processing. The pre-processing can include implementing additive and / or multiplicative gain controls and / or transforming the raw data into a ratio of the raw level or concentration to the level or concentration predicted by the AI model. The pre-processing can include transforming the raw data points using a linear or non-linear relationship, which can correspond to a transformation between the subject-specific data and the population data used to train and / or test the model. In some cases, the transformation can include additive and / or multiplicative gain controls. As another example, subject-specific code can compare results from assays performed on a particular subject while experiencing an abnormal bleeding event with predicted results generated by an AI model corresponding to the same or similar instance in time, and corresponding additive or multiplicative transformations can be used to pre-process the subject's data or to post-process the output generated by the population-level AI model.
[0228] The pre-processing transformation can be determined by (for example) determining a relationship between one or more attributes of a particular subject to the subjects used to train and / or test the AI model, and / or a relationship between the labels and / or outcomes associated with the particular subject and the corresponding data used to train and / or test the AI model. For example, the transformation can be determined based on the frequency of abnormal bleeding during a particular condition for a particular subject, corresponding to the subject represented in the training data. As another example, a multi-dimensional surface can be generated that relates the attributes of the subjects in the training data to the statistics of the observations. A subject-related location in the multi-dimensional plane corresponding to a particular subject (e.g., based on attribute data including demographic data, diagnostic data, laboratory data, etc.) can be identified. A surface representative location having a multi-dimensional surface (e.g., a mean centroid or median centroid) can be identified, and a transformation can be identified based on the subject-related location and the surface representative location.
[0229] A post-processing transformation can be determined by relating (for example) one or more output predictions generated for a particular subject to one or more observations or predictions (e.g., observations corresponding to the particular subject, observations corresponding to other subjects used to train the model, observations corresponding to other subjects deemed similar to the particular subject and used to train or test the model, predictions generated for other subjects used to train or test the model, predictions for other subjects deemed similar to the particular and used to train or test the model, etc.). Post-processing can include implementing (for example) additive or multiplicative gain control and / or another transformation (e.g., activation functions).
[0230] In some cases, the post-processing function 771 can be defined based on (for example) distribution, trend, and / or point analysis. For example, it may be determined that a point sample associated with a particular subject falls below a predetermined probability threshold corresponding to the training data distribution, which can trigger the implementation of a transformation of the point sample (e.g., based on centroids and / or distribution statistics). As another example, a time constant can be calculated for the decay of the concentration / level of the active ingredient / factor level and compared to the time constant corresponding to the training and / or test data, and the post-processing can be defined based on the time constant.
[0231] In some cases, the post-processing function 771 may be determined based on an assay metric (e.g., an aPTT metric) determined for a particular subject for a particular treatment situation (e.g., two days after receiving a particular dose of a given treatment) relative to a statistical assay metric (e.g., an aPTT metric or a mean or median of a distribution of aPTT metrics) corresponding to a particular treatment situation represented in the training data. For example, a clotting time assay can be performed for a particular subject using blood collected after administering a particular treatment dose after washout, and the results can be compared to an equivalent clotting time associated with a subject represented in the training data determined based on a similar situation. Post-processing techniques can be implemented to exponentially scale (for example) the output from the AI model based on the comparison.
[0232] In some cases, the post-processing function associates the numerical model prediction with an action or recommended action (e.g., sending a communication to a care provider or subject, a recommendation that the subject seek medical care, a recommendation that the subject reschedule a procedure delivery, etc.). One or more rules and / or functions relating predictions to recommendations can be defined and / or adjusted based on subject-specific data (e.g., past bleeding event frequency, mutation-based severity, mobility status, etc.). For example, a consultation recommendation can be more liberally recommended (e.g., associated with a broader range of predictive outputs) if the subject is associated with increased bleeding prevalence, increased bleeding susceptibility and / or reduced mobility.
[0233] In some cases, the central artificial intelligence system 750 sends population-level AI models, subject-specific AI models, subject-specific workflows, pre-processing functions, post-processing functions, and / or workflows to the user device 705. In some cases, the user device 705 sends data to the central artificial intelligence system 750, which executes one or more models, functions, and / or workflows to generate results, which are then returned to the user device 705.
[0234] In some cases, the population-level AI model, subject-specific AI code, pre-processing functions, post-processing functions, and / or workflow results depend on sensor data collected at the user device 705. Sensor data dependencies can be learned based on user data and / or population data. For example, the population-level AI model can learn a relationship between the probability and / or type of bleeding event and preceding movement data (e.g., collected using an accelerometer and / or gyroscope), and the subject-level AI model can learn one or more post-processing modifications to convert the population-based prediction to a subject-based prediction. As another example, the population-level AI model can predict the probability of a bleeding event (or a given type of bleeding event, such as a spontaneous bleeding event) occurring based on how different recent sensor data associated with the subject is compared to previous sensor data associated with the subject.
[0235] The pre-processing function 770 may be configured to estimate an activity type based on the sensor data. For example, acceleration and / or gyroscope data may be used to infer whether the user is stationary, walking, or running. In particular, such inference may take into account the possibility that the user is stationary but in a moving vehicle. The AI model (e.g., population-level and / or subject-specific AI model) may then generate a prediction regarding a bleeding event (e.g., probability of occurrence of a possible bleeding event of a given type) based on whether a given type of activity (e.g., running) occurred within a recent time interval, the cumulative duration of the given type of activity (e.g., running or walking) over the recent time interval, estimated physiological data (e.g., maximum heart rate, median heart rate, or average heart rate) associated with the recent time interval based on the inferred recent activity, etc.
[0236] It will be appreciated that the population-level AI model may further or additionally be implemented in a workflow that includes one or more pre-processing functions 770 and / or one or more post-processing functions 771. The pre-processing functions 770 and / or post-processing functions 771 can be defined based on one or more variations of the input data, subject attributes, subject demographic data, subject assay data, subject genetic data, bleeding event timing, bleeding event frequency, bleeding event severity, model predictions, etc. In some cases, the subject-specific AI model is defined to be and / or include a population-level AI model (e.g., selected from among multiple population-level AI models).
[0237] The model usage data 772 may indicate which models are utilized and / or being used by devices associated with various subjects. For example, for a given subject, the model usage data 772 may indicate that 10 population-level models are potentially applied to the subject's current situation, that one subject-specific model has been generated for the subject, that the results of the subject-specific model have been presented via a hemophilia app operating on a user device associated with the subject, and that the subject-specific model recently generated a particular set of predictions that were presented via the user device. The model usage data 772 may further indicate interactions with the model result data, which may include (for example) requesting that the model result data be sent to the provider system 745, that processing be rescheduled, and / or that recalculation be performed. In some cases, the model usage data 772 may indicate the accuracy of the model predictions, for example, the model usage data 772 may include a user-inputted indication that a model prediction of bleeding events occurring by a particular date was inaccurate. The model usage data 772 may be used to further train population-level AI models and / or subject-specific AI models. In some cases, model usage data 772 indicates which subjects (and / or their attributes) are using the model, and population-level training code 766 and / or subject-specific adjustment code 769 can further train the population-level AI model and / or subject-level AI model to improve accuracy for similar subjects and / or other subjects not currently using the model.
[0238] Expression data 773 is an example of empirical outcome data that may be received, utilized, and / or stored by central artificial intelligence system 750. Expression data 773 (or other empirical outcome data) may be used to evaluate the accuracy of one or more population-level models, subject-specific models, population-level workflows, and / or subject-specific workflows. Expression data 773 (or other empirical outcome data) may be monitored to determine whether to initiate retraining of an AI model, whether to select a different AI model, whether to adjust pre-processing and / or post-processing functions used for a given subject, etc. Expression data 773 (or other empirical outcome data) may further indicate the accuracy of various model predictions, which may affect whether such models are subsequently used and / or retrained.
[0239] The display and description of FIG. 7 correspond to an example implementation of what types of operations can be performed in the user device and the central artificial intelligence system 750. It will be understood that other operational capabilities are possible. For example, one or more operations described as occurring in the user device 705 may instead be performed in the central artificial intelligence system 750 and / or vice versa. Furthermore, one or more types of data (e.g., code, functions, models, etc.) described and / or displayed as being stored in the user device 705 may alternatively or additionally be stored in the central artificial intelligence system 750, and / or one or more types of data described and / or displayed as being stored in the central artificial intelligence system 750 may additionally or alternatively be stored in the user device 705.
[0240] It will be appreciated that the computing network can include one or more components, one or more characteristics, and / or one or more functions identified in connection with FIG. 1 in addition to one or more components, one or more characteristics, and / or functions associated with FIG. 7. For example, the provider system 745 can include (and / or have one or more characteristics of) one or more of the computer 105, the laptop 110, and / or the mobile device 115, and / or any of the computer 105, the laptop 110, and / or the mobile device 115 can include (and / or have one or more characteristics of) the provider system 745. As another example, the central artificial intelligence system 750 can include (and / or have one or more characteristics of) the cloud server 135, and / or the cloud server 135 can include (and / or have one or more characteristics of) the artificial intelligence system 750.
[0241] IV. A Process for Using Machine Learning Models to Inform Hemophilia Treatment FIG. 8 illustrates a process 800 for using machine learning models to facilitate identification of hemophilia treatment strategies. Process 800 begins at block 805 where big data is collected. The big data can include data corresponding to a set of subjects including a plurality of subjects. Each of the plurality of subjects can include a person diagnosed with hemophilia. The data can include, for some or all of the plurality of subjects, demographic data (e.g., age, race, sex), physical data (e.g., weight, height), genetic data (e.g., identifying single nucleotide polymorphism mutations and / or copy number variants), treatment data (e.g., identifying a current treatment, one or more previous treatments, and / or one or more time periods during which the current treatment and / or previous treatments were used), health data (e.g., the type of hemophilia the subject was diagnosed with, comorbidities, any other diseases the subject suffers from, physiological metrics), laboratory data (e.g., identifying factor levels administered with or without recent treatment, factor levels administered with or without recent treatment, treatment concentration levels administered with or without recent treatment), and / or symptom data (e.g., frequency of bleeding events, time since last bleeding event). It will be appreciated that the treatment data can indicate whether and / or when the subject received a given treatment (e.g., a treatment identified in Section II.D.) It will also be appreciated that the laboratory data can include the results of an assay or test (e.g., an assay or test identified in Section II.B.).
[0242] In some cases, each of the plurality of subjects is receiving the same hemophilia treatment. In some cases, each of the plurality of subjects is not and / or is not receiving a hemophilia treatment. In some cases, each of the plurality of subjects has the same type of hemophilia (e.g., hemophilia A, B, or C). In some cases, each of the plurality of subjects has the same or similar severity of hemophilia. In some cases, at least some of the plurality of subjects differ with respect to current hemophilia treatment, past hemophilia treatment, hemophilia type and / or hemophilia severity, and / or demographics.
[0243] At block 810, a classifier machine learning model can be used to divide the subject set into subgroups. In some cases, the subject set is divided based on static subject information (e.g., demographic information, the type of hemophilia the subject was initially diagnosed with, the severity of hemophilia the subject was initially diagnosed with, etc.) and / or recent subject information (e.g., current treatment, recent frequency of bleeding events, recent exercise status, etc.). Additionally or alternatively, the subject set can be divided based on results from one or more assays (e.g., as disclosed in Section II.B) and / or one or more hemophilia treatments (e.g., as disclosed in Section II.D) that the subject is currently undergoing or has undergone within a recent time frame. It will be appreciated that the degree to which any variable influences class assignment can depend on the learned parameters of the classifier and / or the distribution of values in the multidimensional space across the subject set.
[0244] The classifier machine learning model may include (for example) a perceptron, a decision tree, a regression, a nearest neighbor, a support vector machine, or a component (e.g., principal component) model. The classifier machine learning model may be trained using at least some of the fields collected in block 805 as input variables. In some cases, the subgroups are identified based on data corresponding to one or more initial time points (e.g., corresponding to a diagnosis date and / or a treatment selection date), based on static information (e.g., the type and / or severity of hemophilia with which the subject was initially diagnosed), and / or not based on a label identifying a subsequent disease characteristic. For example, the subgroups may be identified based on data reflecting the subject's initial demographic, diagnostic, and / or laboratory data. The classifier may be trained using unsupervised learning such that the subgroups are defined based on input data and not based on a corresponding label. In some cases, the subgroups are identified based on data corresponding to one or more initial time points and based on one or more subsequent time points. For example, the input data can reflect demographic, diagnostic and / or laboratory data associated with one or more initial time periods, and the label data can reflect the occurrence of bleeding events, factor concentrations, clotting times, subject mobility and / or survival at one or more subsequent time periods. The classifier can be trained using supervised learning such that subgroups are defined based on the input data (e.g., associated with one or more initial times) and the label data (e.g., associated with one or more subsequent times).
[0245] For each subgroup, a population-level machine learning model is trained using the subgroup's data at block 815. In some cases, some or all of the population-level machine learning models trained for different subgroups are the same type of machine learning model (e.g., the same type of pharmacokinetic model, pharmacokinetic model, machine learning model, or the same type of machine learning model, etc.).
[0246] Training can include supervised learning using training data associated with a set of subjects. Each of the population-level machine learning models can be trained to predict (for example) the probability of a bleeding event (e.g., any type of bleeding event or spontaneous bleeding event) occurring within a given time period, the number of bleeding events (e.g., within a given time period), the decay rate of a factor level, the decay rate of a therapeutic drug level, the rate of change of a factor level at a given time relative to the administration time of a treatment, etc. In some cases, the labels used to train the population-level machine learning models are binary. In some cases, the labels used are non-binary (e.g., numbers on a continuous scale, categories, etc.).
[0247] At block 820, data is received for particular subjects. The particular subjects may have been diagnosed with hemophilia. The particular subjects may include those not included in the set of hemophilia subjects.
[0248] The data may include (for example) demographic data, genetic data, treatment health data, laboratory data, and / or symptom data for the particular subject. For example, the data received in block 820 may identify (for example) the type of hemophilia with which the particular subject has been diagnosed, the severity of the hemophilia with which the particular subject has been diagnosed, the type of mutation detected through processing of the particular subject's sample, one or more test results (e.g., identification of one or more clotting times), a hemophilia treatment currently being received by the particular subject, one or more hemophilia treatments previously received by the particular subject, past bleeding event information (e.g., identification of the frequency of spontaneous and / or non-spontaneous bleeding events), and / or a level of mobility.
[0249] At block 825, a subject-specific machine learning workflow is defined for the particular subject using at least some or all of the data received at block 820. Defining the workflow may include (for example) selecting and / or training a particular population-level machine learning model for the particular subject. Defining the workflow may alternatively or additionally include selecting and / or configuring one or more pre-processing and / or post-processing techniques for use with the particular subject.
[0250] In some cases, further models are used to select a population-level machine learning model to use for a particular subject. Other models can include (for example) decision trees, classifiers, rule-based models, etc. Selection can be made by using data associated with the subject to (for example) assign the data to a cluster that is used to split the subject set into subgroups (and then to select a population-level machine learning model associated with the cluster), to identify nearest neighbors from a training dataset associated with the hemophilia subject set (and then to select a population-level machine learning model trained using data from the nearest neighbors). In some cases, the selected population-level machine learning model can be further trained using the subject-specific data.
[0251] As another example, block 825 may include identifying post-processing and / or pre-processing to be applied to a particular subject. For example, one or more linear or non-linear transformations (or one or more models) may be applied to transform data for a particular subject before feeding the data to a machine learning model. As another example, one or more transformations (or one or more models) may be applied to transform results generated by a machine learning model into subject-specific results. Pre-processing and / or post-processing techniques may be determined based on identifying distances and / or vectors between the multidimensional representations of a particular subject (e.g., based on all or a portion of an input dataset associated with the machine learning model) compared to representations of other subjects used to train the machine learning model, determining transformations based on portions of the subject dataset associated with similar distances and / or vectors (e.g., and the same subject-level machine learning model), and / or identifying scaling factors (e.g., using lookup data and / or defined relationships).
[0252] As an example, the level of a therapeutic agent can be measured in a sample of a particular subject at a particular time after administration of a particular dose of the therapeutic agent. The level (and possibly the particular time and / or the particular dose) can be used to calculate a clearance metric for the particular subject, which can then be used to configure a post-processing algorithm (e.g., a non-linear post-processing algorithm) to modify the output of a population-level machine learning model (e.g., based on a comparison between the clearance metric for the particular subject and the clearance metric associated with the subjects used to train the model).
[0253] At block 830, recent dynamic data associated with the particular subject is received. The recent dynamic data may characterize one or more occurrences (e.g., treatment prescriptions, treatment administrations, occurrences of bleeding events, identification of test results, etc.) that occur after one or more times that the data received at block 820 was identified and / or received (at block 820). In some cases, the recent data may characterize one or more occurrences that occur after the workflow was defined at block 825.
[0254] The dynamic data can indicate (for example) treatment schedules, when a particular administration of a therapeutic agent occurred (e.g., and / or which therapeutic agent was administered and / or its dosage), that a recent abnormal bleeding event occurred (potentially indicating whether a bleeding event was triggered, whether the bleeding event was internal, any therapeutic response received to the bleeding event, and / or whether the bleeding event resulted in any lasting damage), and / or new test results (e.g., identifying clotting times in the absence of any treatment, therapeutic concentrations and / or factor levels, and / or specific times compared to the administration of a particular treatment).
[0255] In block 835, the dynamic data can be used to define a new or modified workflow for the subject. For example, a new population-level machine learning model may be selected for the particular subject. The new selection can be made in response to determining that (for example) the recent dynamic data is distinguishably different from corresponding data in a training dataset associated with a subgroup used to train the population-level machine learning model, pre-processing techniques or post-processing techniques used in the particular subject's workflow. Alternatively or additionally, the new selection can be made in response to determining that (for example) the recent dynamic data differs by at least a threshold amount (e.g., 10%, 20%, 30%) from the corresponding prediction generated or would have been generated by the particular subject's workflow. Alternatively or additionally, the new selection can be made in response to determining that (for example) the probability that the recent dynamic data will occur as predicted via the workflow defined in block 825 falls below a predetermined threshold (e.g., 80%, 60%, 50%, 33%, etc.).
[0256] At block 840, a hemophilia-related prediction is generated for the subject using the new or modified workflow. The prediction may include (for example) predicted rate constants (e.g., for absorption, distribution, metabolism, and / or excretion of the therapeutic agent), predicted levels (e.g., concentrations) of the therapeutic agent or factor (e.g., at a particular time point, a particular time point relative to treatment administration, etc.), predicted dosages and / or dosing intervals of the therapeutic agent sufficient to maintain a predicted clotting time within a predetermined range, a predicted probability of a bleeding event occurring within a time interval, etc.
[0257] In block 845, a treatment specification is identified based on the prediction. Block 845 may include (for example) selecting a therapeutic agent, selecting a dose of the treatment (or active ingredient) to be performed, selecting a frequency of administration of the treatment, identifying a treatment administration schedule, etc. Block 845 may alternatively or additionally include determining whether to recommend short-term prophylactic treatment. For example, the treatment specification may include and / or match the current treatment specification if the prediction is satisfactory (e.g., if the predicted concentration of the therapeutic agent is above a predetermined threshold, the predicted probability of a bleeding event is below a predetermined threshold, and the predicted clotting time is within a predetermined range). As another alternative or additional example, a new or modified workflow may be used to identify an adjusted treatment administration and / or treatment schedule.
[0258] The treatment specification may be output. For example, the treatment specification may be output (e.g., as a recommendation) via a display of information about the particular subject's device and / or the particular subject's caregiver's device. As another example, the treatment specification may be sent to the particular subject's caregiver's device.
[0259] It will be understood that part or all of process 800 may be performed on a device of a particular subject, part or all of process 800 may be performed on a computing system associated with a healthcare provider associated with a particular subject, and / or part or all of process 800 may be performed on a remote computing system.
[0260] 9 illustrates a process 900 for generating hemophilia-related results using sensor data. Process 900 begins at block 905 where a subject-specific dataset corresponding to a particular subject diagnosed with hemophilia is received. The subject-specific dataset may include (for example) demographic data (e.g., age, race, sex), physical data (e.g., weight, height), genetic data (e.g., identification of single nucleotide polymorphism mutations and / or copy number variants), treatment data (e.g., identifying the current treatment, one or more previous treatments, and / or one or more time periods during which the current treatment and / or previous treatments were used), health data (e.g., the type of hemophilia the subject was diagnosed with, comorbidities, any other diseases the subject suffers from, physiological metrics), laboratory data (e.g., identifying factor levels administered with or without recent treatment, factor levels administered with or without recent treatment, therapeutic concentration levels administered with or without recent treatment), and / or symptom data (e.g., frequency of bleeding events, time since last bleeding event). In some cases, at least a portion of the subject-specific data set was initially identified via a communication from the subject's device (e.g., received at a care provider device and / or a remote server). In some cases, at least some of the subject-specific data set included or was based on one or more photographs (e.g., collected at the subject's device). The photographs may display (for example) previous bleeding, injury, bruising, or site. In some cases, at least a portion of the subject-specific data was initially identified via and / or received from an electronic medical record corresponding to the subject. The subject-specific data set may include and / or consist of recent, current, and / or static information. It will be appreciated that the treatment data may identify a treatment that the subject is undergoing or has recently undergone (e.g., a treatment identified in Section II.D). It will also be appreciated that the laboratory data may include results of an assay or test (e.g., an assay or test identified in Section II.B).
[0261] At block 910, at least a portion of the subject-specific dataset is processed using a classifier to identify a machine learning model used to generate a prediction for a particular subject. At least a portion of the subject-specific dataset may include a type of hemophilia, demographic data, and / or treatment data. At least a portion of the subject-specific dataset may alternatively or additionally include a photograph. The classifier may include (for example) a perceptron, a decision tree, a regression, a nearest neighbor, a support vector machine, or a component (e.g., principal component) model. The classifier may be trained using training data including at least some values of the fields included in the subject-specific dataset received at block 905. The values or representations thereof may be provided as inputs to the model, and outputs generated by the model during training may be compared to labels (e.g., numerical or categorical labels) in the training dataset via a loss function. The classifier may include (for example) a neural network (e.g., a convolutional neural network) configured to process the photographs. The neural network may generate an output that predicts (for example) whether a photograph depicts a bleeding event and / or the type of bleeding displayed (e.g., whether bleeding or bruising is displayed, location of the bleeding / bruising, size of the bleeding / bruising, etc.).
[0262] In some cases, initial subject data (e.g., corresponding to a particular subject or subjects in a training set) can be encoded, and the encoded representations can be provided as input to the model. The encoding can include categorical encoding techniques such as one-hot encoding, ordinal encoding, contrast encoding, binary encoding, etc. For example, one-hot encoding can represent categorical values as a set of binary numbers representing each potential category. In some cases, numeric values (e.g., age) can be converted to categorical values (e.g., age bins) and then encoded via one-hot encoding. In some cases, the numeric data can be converted to categorical data using a set of ranges, and the categorical data can be encoded using a categorical encoding technique.
[0263] The encoded representations of the values may then be aggregated and fed to a model. In some cases, each value of the encoded representation is fed to a separate node in an input layer of the neural network. In some cases, the encoded representations are concatenated and converted into a feature vector (e.g., using a classifier machine learning model). The feature vector may be used to identify a machine learning model. For example, each of a set of models may correspond to a location in a multidimensional space, and a distance may be calculated between the location of each model's representation and the location associated with the feature vector associated with the subject. The model associated with the smallest distance may be selected.
[0264] In some cases, the machine learning model is identified using subject-specific data corresponding to one or more initial time points (e.g., corresponding to the date of diagnosis and / or treatment selection date), based on static information (e.g., the type and / or severity of hemophilia the subject was initially diagnosed with), and / or not based on a label identifying a subsequent disease characteristic. For example, a machine-level model can be identified based on data reflecting a particular subject's initial demographic, diagnostic, and / or laboratory data. The identified machine learning model can include (for example) a model having one or more characteristics disclosed herein (e.g., Section III.F), a pharmacokinetic model, and / or a neural network.
[0265] The machine learning model may include a population-level machine learning model trained using training data corresponding to a population of subjects, or may be a subject-specific machine learning model trained at least in part using training data related to a particular subject. For example, the subject-specific machine learning model may be initialized with parameters from a population-level machine learning model and further trained using data from a particular subject. The subject-specific machine learning model and / or the subject-specific model may be generated using techniques disclosed herein (e.g., as disclosed in connection with network 700 and / or process 800) and / or may have characteristics disclosed herein (e.g., as disclosed in connection with network 700 and / or process 800).
[0266] It will be appreciated that a rule-based approach can be used instead of a classifier. For example, each of the set of machine learning models can be trained using training data corresponding to a set of subjects having one or more of the same characteristics (e.g., hemophilia A, hemophilia B, hemophilia C, a particular severity, a particular current treatment, the presence of inhibitors, etc.). The model selected in block 910 to be used for a particular subject can include a model trained using training data associated with other subjects who share one or more characteristics (e.g., hemophilia type, hemophilia severity, treatment, demographic attributes, etc.) with the particular subject. For example, the set of population-level machine learning models can include four models: a first model trained using training data associated with subjects with mild or moderate hemophilia A; a second model trained using training data associated with subjects with severe hemophilia A; a third model trained using training data associated with subjects with mild or moderate hemophilia B; and a fourth model trained using training data associated with subjects with severe hemophilia B. If a given subject has moderate hemophilia B, a third model may be selected.
[0267] In some cases, only a single machine learning model may be stored and / or trained, and a single machine learning model may be identified for use in block 910. In some cases, block 910 alternatively or in addition includes identifying one or more pre-processing and / or post-processing functions (e.g., using a classifier). In some cases, block 910 alternatively or in addition includes identifying a workflow (e.g., using a classifier).
[0268] At block 915, instructions are received identifying one or more times that hemophilia treatments have been received by a particular subject. The indication may identify (or information may be available indicating) what type of treatment was received and / or the administration of treatment received. Block 915 may include receiving information regarding the time that one or more prophylactic treatments were received and / or the time that one or more symptomatic treatments were received. In some cases, block 915 indicates that a particular subject has not received any hemophilia treatments within a given period of time.
[0269] In block 920, a hemophilia-related time course is predicted using the treatment information (e.g., treatment administration time, treatment duration and / or treatment dosing) and the selected machine learning model. The time course may predict how a hemophilia-related variable will change over time. The hemophilia-related variable may include (for example) a clotting time (e.g., any clotting time disclosed herein), a factor level, a therapeutic agent concentration (e.g., of or associated with a treatment), and / or a probability of occurrence of a bleeding event. Block 920 may include (for example) predicting one or more rate constants, predicting a value of the hemophilia-related variable for each of one or more times, and / or predicting a function that identifies how the hemophilia-related variable will change over time.
[0270] It will be appreciated that the workflow used to generate a prediction for a particular subject may be configured to receive an input indicating when the treatment was received, and may also receive other types of information. Other types of information may include some or all of the subject-specific data set received in block 905 (for example). For example, the workflow may receive the weight, height, and age of the particular subject, and / or historical laboratory data related to the particular subject, in addition to the treatment administration time. The workflow may include the population-level machine learning model identified in block 910 (or a subject-specific machine learning model built based on the population-level machine learning model identified in block 910). In some cases, the workflow may further include one or more pre-processing functions and / or one or more post-processing functions.
[0271] It will be appreciated that the treatment information may be encoded (e.g., using the techniques disclosed in connection with block 910) and / or represented as a feature vector. The encoded and / or feature vector representation may then be provided to a workflow and / or a machine learning model. In some cases, the input provided to the workflow and / or machine learning model further represents at least a portion of the subject-specific dataset received in block 905. For example, the input may represent the subject's height, weight, age and / or sex, and / or the subject's hemophilia severity level.
[0272] The prediction generated in block 920 can be used to determine (for example) whether and / or when a predicted value exceeds a threshold value (e.g., representing an increased risk of an abnormal bleeding event) so that an alert can be presented. The prediction may alternatively or additionally be used to generate a recommended treatment schedule (e.g., to maintain a predicted probability of an abnormal bleeding event, a predicted clotting time, a predictor level and / or a predicted treatment level within an acceptable predetermined closed or open range).
[0273] At block 925, a representation of sensor data collected at a device associated with the subject can be received. The device can include (for example) a smartphone, a wearable smart device, a smart watch, a smart headset, smart headphones, etc. The sensor data can include (for example) accelerometer data, gyroscope data, and / or GPS data. The sensor data can include physiological data, including (for example) a heart rate. The representation of the data can include the sensor data itself or a processed version thereof. For example, the processed version can include a normalized or regularized metric (e.g., relative to other data collected by the same sensor, other data collected on the same device, and / or other data collected on other devices) and / or one or more statistics (e.g., sum, average, maximum, or median of the raw or processed data over a period of time). As another example, the processed version can include an estimated number of steps taken by the subject during a given period of time, an estimated distance traveled by the subject during a given period of time, an estimated speed statistic (e.g., average, median, or maximum speed) over a period of time, or an estimated portion of the period during which the subject was stationary. The sensor data may be received (for example) relatively continuously, periodically, in response to a pull request, or in response to receiving a request for an updated prediction.
[0274] At block 930, a determination is made as to whether to transition to a new or modified workflow (e.g., from a workflow that includes the machine learning model identified at block 910). In some cases, block 930 includes determining whether a workflow transition condition is met, such that the workflow used to predict the hemophilia-related time course at block 910 continues to be used unless the workflow transition condition is met.
[0275] The determination of block 930 may be based on the representation of the sensor data, the processed version of the sensor data, the photograph, and / or the processed version of the photograph. For example, it may be determined that the workflow transition occurs if a statistical analysis (e.g., a Monte Carlo analysis) indicates that the probability of detecting the sensor data (or the representation and / or processed version thereof) is below a predetermined threshold or that the probability that the subject experienced a given type of bleeding event or bleeding events is below a predetermined threshold. The predetermined threshold may be determined based on past sensor data collected from the device, past sensor data collected in association with a particular subject, past sensor data collected in association with a set of other subjects, the frequency of bleeding events previously detected in association with a particular subject, and / or the frequency of bleeding events previously detected in association with a set of other subjects. By way of illustration, an analytical and / or probability-based approach may identify a threshold associated with the sensor data that represents a 5% (or other percentage) probability that such sensor data (or bleeding event) will be observed in a given data set. If threshold crossing or other condition satisfaction indicates that a particular sensor data (or its processed version and / or representation) is sufficiently unlikely to be observed within a given underlying data distribution, then at block 930, a decision may be made to discontinue use of the workflow (e.g., permanently or temporarily suspend) and / or transition to a new or modified workflow for hemophilia-related prediction.
[0276] If it is determined to continue using the workflow, the process 900 can return to block 920, and the workflow can be used to generate a new prediction. If it is determined not to transition the workflow, the process 900 can proceed to block 935, and a transformed data processing workflow can be determined based on the sensor data representation.
[0277] Abnormal, different, or outlier sensor data may indicate (for example) that the subject is becoming increasingly limited and / or disabled by hemophilia, and / or that the pharmacokinetic rate constants have changed, are changing, or will change. For example, increased disability may indicate that previous treatments were not aggressive enough. As another example, increased movement may correspond to more rapid distribution, absorption, metabolism, and / or excretion compared to less or no movement. Abnormal, different, or outlier sensor data may alternatively or additionally indicate that the subject is at higher risk for abnormal bleeding events (e.g., higher risk of spontaneous or induced abnormal bleeding events). Additionally, unexpected bleeding events (e.g., visible in a photograph) may suggest that the pharmacokinetic rate constants previously used for the subject do not adequately represent the subject's physiology.
[0278] Converting the data processing workflow can include selecting a new machine learning model, which can have the same or a different architecture as the model used in block 920. For example, both models can include the same type of pharmacokinetic model with different learning parameters (e.g., trained using different training data). As another example, the machine learning model used in block 920 can include a first type of pharmacokinetic model, and the machine learning model used in the converted data processing workflow can include a different second type of pharmacokinetic model. In one example, one or more activation functions, weights, and / or thresholds used in the machine learning model in block 920 are different from one or more activation functions, weights, and / or thresholds used in the converted data processing workflow.
[0279] The change in activation function or threshold can represent the probability that a bleeding event caused by (e.g.) the same level of therapeutic drug, factor level or injury is abnormal, but can correspond to different risks in different activity scenarios. The risk of injury may be higher during high intensity activity, and thus the machine learning model can learn to shift the activation function and / or lower the threshold. Alternatively or additionally, the indication of an unexpected bleeding event can trigger a shift in the activation function and / or a lowering of the threshold. This adjustment can facilitate more accurately and / or cautiously predicting whether an abnormal bleeding event will occur (e.g., by considering differences in injury risk) and / or better advising the subject whether to modify activity (to reduce the risk of abnormal bleeding) and / or undergo treatment.
[0280] It will be appreciated that where activation functions or thresholds differ, one or more other parameters may (but need not) differ between the models. For example, differences in coefficients and / or weights may represent differences in pharmacokinetic values, and differences in activation functions or thresholds may represent differences in predicted injury probabilities.
[0281] Converting the data processing workflow can include introducing or modifying (for example) pre-processing or post-processing. For example, in response to detecting sensor data showing high motion or in response to a photograph displaying a bleeding event, thresholds used in post-processing to determine whether to issue an alarm can be adjusted so that an alarm identifying the risk of an abnormal bleeding event is more likely to be generated. As another example, parameters used in the machine learning model can be adjusted to reflect different rate constants (corresponding to faster rates) of absorption, excretion, metabolism and / or distribution. As yet another example, a different machine learning model (e.g., a pharmacokinetic model or a neural network) can be selected to use in place of the machine learning model identified in block 910. The different machine learning model can have the same or a different architecture relative to the machine learning model selected in block 910. The different machine learning model may have been trained using training data corresponding to hemophilia subjects who were more (or less) active relative to the hemophilia subjects associated with the training dataset used to train the machine learning model identified in block 910.
[0282] In some cases, a first training data set is used to train the machine learning model (e.g., and / or one or more pre-processing functions and / or one or more post-processing functions) used in block 925, and a second training data set is used to train the machine learning model (e.g., and / or one or more pre-processing functions and / or one or more post-processing functions) in the transformed data processing workflow. The subjects corresponding to the first training data set may be the same as the subjects corresponding to the second training data set, may be completely different, or may be partially different. In some cases, the first training data set or another training data set used to further train the machine learning model (e.g., and / or one or more pre-processing functions and / or one or more post-processing functions) used in block 925 may (but need not) include data corresponding to the subjects for which predictions are generated in process 900. The second training data set or another training data set used to further train the machine learning model (e.g., and / or one or more pre-processing functions and / or one or more post-processing functions) used in block 935 may (but need not) include data corresponding to the subjects for which predictions are generated in process 900.
[0283] In some cases, the first training data set is used to train the machine learning model (e.g., and / or one or more pre-processing functions and / or one or more post-processing functions) used in block 925, and the third training data set is used to learn how to modify the machine learning model (e.g., and / or one or more pre-processing functions and / or one or more post-processing functions) in response to detecting different types of sensor data, inferring different types of activity, inferring exercise or intensity levels for different subjects, etc. The third training data set can (but need not) include data corresponding to subjects for which predictions are generated in process 900.
[0284] The first, second and / or third training data sets may include (for example) assay results and / or indicators corresponding to whether abnormal bleeding was observed.
[0285] At block 940, a new hemophilia-related prediction is generated for the particular subject using the transformed data processing workflow. In some cases, treatment information (e.g., identification of the time, dosage and / or type of one or more treatment administrations), some or all of the subject-specific data set received at block 905, and / or activity data based on and / or including the sensor data are input into the transformed workflow. In some cases, one or more state variables and / or intermediate variables are identified from the initially identified machine learning model and / or a workflow using the initially identified machine learning model. The transformed data processing workflow may be configured and / or fed to the state variables and / or intermediate variables. For example, block 920 may include iteratively predicting current levels of therapeutic agents, factors and / or inhibitors, and the transformed data processing workflow may be initialized to include the predicted current levels.
[0286] The hemophilia-related predictions may predict levels (e.g., concentrations) of factors and / or treatments, clotting times, risk of an abnormal bleeding event occurring (e.g., spontaneously or after injury). The hemophilia-related predictions may correspond to the same types of predictions generated in block 920. A post-processing function in the transformed data processing workflow may determine whether a given condition is met based on the output from the machine learning model used in the transformed data processing workflow. For example, the post-processing function may determine whether the predicted value exceeds a threshold (e.g., below a factor level threshold, below a therapeutic level threshold, above a clotting time threshold, or above an abnormal bleeding risk threshold). The thresholds used in the transformed data processing workflow may be different from the thresholds used in the initial workflow.
[0287] In some cases, the results generated based on the transformed data processing workflow are output in block 945. Outputting the results can include displaying or transmitting the results (e.g., to a device or a healthcare provider associated with a particular subject). The results can include output from the transformed data processing workflow or a processed version thereof. In some cases, the results correspond to a warning (e.g., identifying a low predictor level) or instructions (e.g., for the subject to contact a healthcare provider or vice versa, or to facilitate treatment of the subject). For example, the recommendation can be that the subject receive administration of a factor concentrate, recombinant factor concentrate, and / or other fast-acting treatment. The recommended and / or provided treatment may be different or the same as any, one, or all treatments that the subject receives prophylactically.
[0288] Whether the result is output can depend on whether a predetermined condition is satisfied. For example, the result may be selectively output when the prediction from the converted data processing workflow is within a predetermined open or closed range. As another example, the result can be selectively output when the prediction from the converted data processing workflow is equal to 1, indicating satisfaction of an alert condition.
[0289] Process 900 then proceeds from block 940 back to block 925 where a workflow is dynamically selected and used to process the new sensor data.
[0290] It will be understood that part or all of process 900 may be performed on a device of a particular subject, part or all of process 900 may be performed on a computing system associated with a healthcare provider associated with a particular subject, and / or part or all of process 900 may be performed on a remote computing system.
[0291] FIG. 10 illustrates a process 1000 for generating model predictions using machine learning models and sensor-based post-processing. Process 1005 begins at block 1005, where subject data is received. The subject dataset can include (for example) demographic data (e.g., age, race, sex), physical data (e.g., weight, height), genetic data (e.g., identification of single nucleotide polymorphism mutations and / or copy number variants), treatment data (e.g., identifying the current treatment, one or more previous treatments, and / or one or more time periods during which the current treatment and / or previous treatments were used), health data (e.g., type of hemophilia the subject was diagnosed with, comorbidities, any other disease the subject suffers from, physiological metrics), laboratory data (e.g., identifying factor levels administered with or without recent treatment, factor levels administered with or without recent treatment, therapeutic concentration levels administered with or without recent treatment), and / or symptom data (e.g., frequency of bleeding events, time since last bleeding event). The subject dataset can include and / or consist of recent, current, and / or static information. It will be appreciated that the treatment data can identify a treatment that the subject is receiving or has recently received (e.g., a treatment identified in Section II.D.) It will also be appreciated that the laboratory data can include the results of an assay or test (e.g., an assay or test identified in Section II.B.).
[0292] At block 1010, a schedule of treatment administration is received. The schedule may include one or more times at which one or more treatments are administered. The times may be, but need not be, evenly spaced. Block 1010 may include receiving information regarding the times at which one or more prophylactic treatments were received and / or the times at which one or more symptomatic treatments were received. In some cases, block 1010 indicates that a particular subject has not received any hemophilia treatments within a given period of time.
[0293] At block 1015, a representation of sensor data is received. The sensor data may have been collected at a user device associated with the subject. The device may include (for example) a smartphone, a wearable smart device, a smart watch, a smart headset, smart headphones, etc. The sensor data may include (for example) accelerometer data, gyroscope data, and / or GPS data. The sensor data may include physiological data (e.g., instantaneous heart rate or time-averaged heart rate). It will be appreciated that multiple types of sensor data (e.g., accelerometer and heart rate data) may be collected, in which case the multiple types of sensor data may be collected and / or received at the same or different frequencies.
[0294] The representation of the data can include the sensor data itself or a processed version thereof. For example, the processed version can include a normalized or regularized metric (e.g., relative to other data collected by the same sensor, other data collected on the same device, and / or other data collected on other devices) and / or one or more statistics (e.g., sum, average, maximum, or median of the raw or processed data over a period of time). As another example, the processed version can include an estimated number of steps taken by a subject during a given period of time, an estimated distance traveled by a subject during a given period of time, an estimated speed statistic (e.g., average, median, or maximum speed) over a period of time, or an estimated portion of a period during which the subject was stationary.
[0295] The sensor data can be received (for example) relatively continuously, periodically, in response to a pull request, or in response to receiving a request for an updated prediction. For example, receiving the representation of the sensor data can include an application executing at the subject device and locally accessing the sensor data collected at the subject device. As another example, receiving the representation of the sensor data can include a remote computing system receiving a communication from the subject device including a processed version of the sensor data collected at the subject device.
[0296] At block 1020, the subject's physical activity over a period of time is inferred. The period of time may correspond (for example) to a period during which sensor data was received or a period that ends when sensor data was received. For example, sensor data may be collected and / or received at the subject device every minute, and process 1000 may include inferring the physical activity for each 5 minute period based on a representation of the 5 sensor data points collected and / or received during that period. As another example, sensor data may be collected and / or received at the subject device every 15 minutes, and process 1000 may include generating, for each 15 minutes, a physical activity engaged in by the subject during the previous period based on recent sensor data associated with the time at which the period ends.
[0297] Inferring physical activity can include using the representation of the sensor data to infer which particular type of physical activity the subject engaged in. For example, a set of activity classes can be predefined, and the sensor data representation can be fed to a classifier to predict the activity. The predefined classes can include particular activities (e.g., sleeping, sitting, walking, running, biking) and / or activity characteristic categories (e.g., inactivity, low intensity activity, and high intensity activity).
[0298] The classifier can include a machine learning model or a rule-based model. For example, the machine learning model can use a clustering-based approach (e.g., K-means clustering, density-based spatial clustering for noisy applications, mean-shift clustering, expectation-maximization clustering, decision tree clustering, or component-based clustering) to identify distinct clusters of sensor data and infer the physical activity of each cluster based on an initial rule or signature (e.g., indicating that running is more likely to correspond to larger accelerometer data than walking, etc.). As another example, a rule-based approach can identify which of a set of predefined ranges (e.g., one-dimensional or multi-dimensional ranges) contains the sensor data representation and then infer the activity corresponding to that range.
[0299] In some cases, block 1020 includes estimating a number of steps taken by the subject over the time interval. Data from a pedometer sensor can directly identify the number of steps. The number of steps can be estimated using data from one or more accelerometers, one or more gyroscopes, and / or one or more GPS sensors.
[0300] At block 1025, a physiological state of the subject over a period of time is inferred. The time period may, but need not, correspond to the same time period for which physical activity was inferred at block 1020. In some cases, a physiological state is inferred based on the inferred physical activity. For example, high intensity activity (e.g., running, speed walking, high intensity biking) may be inferred to increase muscle blood flow (which may speed up distribution of the therapeutic agent, rapid absorption at active sites, and slow absorption at non-active sites), decrease plasma volume (which may slow down distribution of the therapeutic agent), decrease renal blood flow (which may slow down excretion of the therapeutic agent), and / or increase sweating (which may facilitate excretion of the therapeutic agent).
[0301] In some cases, a physiological state is inferred based on one or more representations of the sensor data. For example, if a representation of sensor data from an accelerometer or heart rate monitor (e.g., optical signals can be used to detect oxygenated blood flow under the skin and thus detect heart rate) is high, it can be inferred that high intensity activity (e.g., running, speed walking, high intensity biking) increases muscle blood flow (which can speed up the distribution of the therapeutic agent, its absorption at sites of exercise and its absorption at sites of non-exercise), decreases plasma volume (which can slow down the distribution of the therapeutic agent), decreases renal blood flow (which can slow down the excretion of the therapeutic agent), and / or increases sweating (which can facilitate the excretion of the therapeutic agent).
[0302] In some cases, process 1000 includes blocks 1020 and 1025. In some cases, process 1000 includes blocks 1020 and does not include 1025. In some cases, process 1000 includes blocks 1025 and does not include 1020. In some cases, block 1000 does not include blocks 1020 or 1025.
[0303] At block 1030, a hemophilia-related prediction is generated using the machine learning model and the sensor data representations (e.g., and / or physical activity and / or physiological state inferred based on the sensor data representations). The hemophilia-related prediction may predict factor and / or treatment levels (e.g., concentrations), clotting times, risk of developing an abnormal bleeding event (e.g., spontaneously and / or after injury).
[0304] The machine learning model may include one or more models disclosed herein (e.g., Section III.F). The machine learning model may include (for example) a pharmacokinetic model and / or a neural network. The hemophilia-related predictions may be generated using a workflow that includes the machine learning model, optionally includes one or more pre-processing functions, and optionally includes one or more post-processing functions. The machine learning model and / or workflow may be received and / or otherwise configured based on treatment information indicating when the subject received a hemophilia treatment, the administration of the hemophilia treatment, and / or which hemophilia treatment was received. The machine learning model and / or workflow may be configured to predict how the availability, concentration, and / or efficacy of the treatment changes over time (e.g., as a result of distribution, metabolism, absorption, and / or excretion) and / or to predict the risk of abnormal bleeding events. It will be appreciated that physical activity may affect the hemophilia-related predictions due to (for example) changes in pharmacokinetics and / or changes in risk of injury (which may alter the risk of involuntary abnormal bleeding), and the machine learning model may capture such relationships.
[0305] In some cases, the machine learning model and / or workflow has been trained to learn how the sensor data affects a variable of interest. For example, assay results (e.g., indicative of clotting time and / or drug concentration) and / or symptom expression data (e.g., whether and / or when an abnormal bleeding event occurred) may have been collected from a hemophilia subject undergoing a given treatment over a period of time relative to administration of the treatment. Sensor data may be further collected during this period, and the model and / or workflow may learn one or more parameters (e.g., one or more weights, rate constants, scaling coefficients) that result in the sensor data affecting an output (e.g., corresponding to an assay result or expression prediction). Some or all of the hemophilia subjects may have been and / or may have been asked to engage in one or more activities (e.g., walking, running, etc.) during a portion of the time they are being monitored. In some cases, the trained machine learning model and / or trained workflow may then receive a representation of the sensor data as an input. In some cases, separate models are trained, each model associated with a different type of activity, activity intensity, and / or physiological state. The workflow can then dynamically select which model to use based on the sensor data, and state variables and / or intermediate values can be passed from the previous model to the current model to facilitate smooth transitions and accurate results.
[0306] In some cases, one or more rules or techniques are used to determine how rate constants (e.g., used in a pharmacokinetic model) and / or weights (e.g., used in a neural network, classifier, pre-processing function, or post-processing function) change in response to various sensor data representations, inferred activity types, inferred activity intensity levels, and / or inferred physiological states. For example, a rule may indicate that the excretion rate constant increases by 5% (or increases the rate constant by another relative or absolute amount, or decreases the rate constant by an absolute or relative amount) during high intensity activity. The increase may be temporary and / or may affect subsequent predictions (e.g., in response to a model that uses previous predictions to generate new predictions). As another example, a rule may be indicated that post-processing should be applied that increases the probability of abnormal bleeding events by 5% (or increases the output by another relative or absolute amount, or decreases the output by an absolute or relative amount) during high intensity activity.
[0307] In some cases, the conditions evaluated in the post-processing function are modified depending on various sensor data representations, inferred activity types, inferred activity intensity levels, and / or inferred physiological states. For example, thresholds or ranges used to determine whether to issue an alert may be defined based on the sensor data representations, inferred activity types, inferred activity intensity levels, and / or inferred physiological states. The thresholds (or range boundaries) may be defined based on (for example) linear relationships, non-linear relationships, step functions, and / or logic-based analyses. By way of example, the post-processing function (and / or a collection of post-processing functions that may be selectively used based on various sensor data representations, inferred activity types, inferred activity intensity levels, and / or inferred physiological states) may instruct to issue an alert if the predicted probability of abnormal bleeding occurring is 10% or greater if the subject is inferred to be stationary, 8% or greater if the subject is inferred to be engaged in low intensity activity, and 5% or greater if the subject is inferred to be engaged in high intensity activity. As another example, the condition may indicate that a factor threshold should be determined using a sigmoid function that relates a threshold to motion data (e.g., estimated intensity data, estimated steps per time interval) or that relates a threshold to heart rate data. The condition may indicate that an alert should be issued if a predicted factor level (e.g., as predicted using a machine learning model) falls below a threshold.
[0308] In some cases, at block 1035, a result including or based on the hemophilia-associated prediction can be generated and output. Outputting the result can include displaying or transmitting the result (e.g., to a device or a health care provider associated with the particular subject). The result can include (for example) the hemophilia-associated prediction and / or a content selected based on the hemophilia-associated prediction. For example, the result can include a warning identifying a low predicted treatment level and / or a high predicted risk of an abnormal bleeding event. As another example, the output result can include a recommendation or instruction to facilitate communication between the subject and a health care provider (e.g., for the subject to contact the health care provider or vice versa), to modify a current activity level or activity selection generally (e.g., for the subject to stop taking in a particular physical activity or high intensity activity), or to administer a treatment. For example, the recommendation can be for the subject to receive administration of a factor concentrate, recombinant factor concentrate, and / or other fast-acting treatment. The recommended and / or provided treatment can be different or the same as any, one, or all treatments the subject receives prophylactically.
[0309] Whether or not a result is output can depend on whether a predetermined condition (e.g., which may be evaluated as part of the workflow generating the hemophilia-related prediction or which may be evaluated separately) exists. For example, a result may be selectively output when a prediction from the transformed data processing workflow is within a predetermined open or closed range. As another example, a result can be selectively output when a prediction from the transformed data processing workflow is equal to 1, indicating satisfaction of an alert condition.
[0310] Process 1000 may further return from block 1030 to block 1015 such that hemophilia-associated predictions are dynamically selected, dynamically configured, and / or iteratively determined using one or more workflows that process the sensor data representations.
[0311] It will be understood that part or all of process 1000 may be performed on a device of a particular subject, part or all of process 1000 may be performed on a computing system associated with a healthcare provider associated with a particular subject, and / or part or all of process 1000 may be performed on a remote computing system.
[0312] V. Further Considerations Some embodiments of the present disclosure include a system including one or more data processors. In some embodiments, the system includes a non-transitory computer-readable storage medium including instructions that, when executed on the one or more data processors, cause the one or more data processors to perform some or all of one or more methods and / or some or all of one or more processes disclosed herein. Some embodiments of the present disclosure include a computer program product tangibly embodied in a non-transitory machine-readable storage medium including instructions configured to cause one or more data processors to perform some or all of one or more methods and / or some or all of one or more processes disclosed herein.
[0313] The terms and expressions used are used as terms of description and not of limitation, and in the use of such terms and expressions there is no intention to exclude any equivalents of the features shown and described or portions thereof, but it is recognized that various modifications are possible within the scope of the invention as claimed. Thus, although the invention as claimed has been specifically disclosed by embodiments and optional features, it will be understood that modifications and variations of the concepts disclosed herein may be resorted to by those skilled in the art, and such modifications and variations are considered to be within the scope of the invention as defined by the appended claims.
[0314] This description provides only preferred exemplary embodiments and is not intended to limit the scope, applicability, or configuration of the present disclosure. Rather, this description of preferred exemplary embodiments provides those skilled in the art with an enabling description for implementing various embodiments. It will be understood that various changes can be made in the function and arrangement of elements without departing from the spirit and scope of the appended claims.
[0315] Specific details are given in the following description to provide a thorough understanding of the embodiments. However, it will be understood that the embodiments can be practiced without these specific details. For example, circuits, systems, networks, processes, and other components may be shown as components in block diagram form so as not to obscure the embodiments in unnecessary detail. In other examples, well-known circuits, processes, algorithms, structures, and techniques may be shown without unnecessary detail so as to avoid obscuring the embodiments.
[0316] VI. Further Examples a first example includes a method including receiving a subject-specific dataset corresponding to a subject, the subject-specific dataset indicating a hemophilia type, a treatment type, and demographic data corresponding to the particular subject; processing the subject-specific dataset using a classifier model to identify one or more population-level machine learning models from a set of population-level machine learning models, each of the set of population-level machine learning models including a machine learning model trained using a training set corresponding to a set of other subjects with hemophilia; receiving one or more instructions identifying one or more times at which a treatment-type treatment was administered to the subject; predicting a hemophilia-related time course for the subject using the one or more times and predicting a data processing workflow using one of the one or more population-level machine learning models; receiving a representation of sensor data collected at a device associated with the subject; determining a transformed data processing workflow based on the representation of the sensor data to generate a hemophilia-related prediction for the subject; generating a hemophilia-related prediction for the subject using the transformed data processing workflow; and outputting a result corresponding to the hemophilia-related prediction.
[0317] A second example includes an example where generating a hemophilia-related prediction includes identifying a post-processing algorithm based on the sensor data, and determining a transformed data processing workflow includes identifying an initial result using the data processing workflow, determining a post-processing function based on the sensor data, and transforming the initial result using the initial result and the post-processing function.
[0318] A third example includes the first or second example, where determining the transformed data processing workflow includes selecting an activation function.
[0319] A fourth example includes any of the first through third examples, where the population-level machine learning model includes a pharmacokinetic model.
[0320] A fifth example includes any of the first through fourth examples, where processing the subject-specific dataset with the classifier model to identify one or more population-level machine learning models includes encoding the subject-specific dataset and using a distance-based algorithm to identify distances between the encoding of the subject-specific dataset and encodings of other subject-specific datasets included in the training set.
[0321] A sixth example includes any of the first to fifth examples, further including adding at least the predicted hemophilia-related prediction to the predicted hemophilia-related time course, where the one or more predicted factor levels include the predicted factor levels; receiving another representation of new sensor data collected at a device associated with the subject; determining to resume use of the data processing workflow based on the another representation; generating another hemophilia-related prediction for the subject using the data processing workflow and the added time course; and outputting another result corresponding to the another hemophilia-related prediction.
[0322] A seventh example includes any of the first through sixth examples, further including determining that a workflow transition condition has been met based on the sensor data, wherein a converted data processing workflow is determined in response to determining that the workflow transition condition has been met.
[0323] An eighth example includes any of the first through seventh examples, wherein generating the hemophilia-associated prediction includes determining one or more state or interim variables calculated for recent time steps by a population-level machine learning model, and configuring a data processing workflow transformed to use the one or more state or interim variables.
[0324] A ninth example includes any of the first through eighth examples, wherein the sensor data represents movement of the subject.
[0325] A tenth example includes any of the first through eighth examples, where the sensor data represents a physiological attribute of the subject.
[0326] An eleventh example includes any of the first through tenth examples, wherein the predicted hemophilia-associated time course includes a predicted level of the factor, therapeutic agent or active ingredient for each of a set of time points, and the hemophilia-associated prediction includes another predicted level of the factor, therapeutic agent or active ingredient associated with another time point.
[0327] A twelfth example includes any of the first through tenth examples, wherein the predicted hemophilia-related time course includes, for each of a set of time points, a predicted clotting time or a probability that an abnormal bleeding event will occur, and the hemophilia-related prediction includes a different predicted clotting time or a different probability that an abnormal bleeding event will occur associated with a different time point.
[0328] A thirteenth example includes any of the first through twelfth examples, where the results indicate a recommendation to reduce the subject's activity level.
[0329] A fourteenth example includes any of the first through twelfth examples, where the results indicate a recommendation to treat the subject in view of an increased probability of abnormal bleeding.
[0330] A fifteenth example includes any of the first through fourteenth examples, further including treating the subject with a hemophilia treatment based on the result.
[0331] A sixteenth example includes any of the fifteenth examples, wherein the hemophilia treatment is different from the treatment type treatment administered at one or more times.
[0332] A seventeenth example includes a system including one or more data processors and a non-transitory computer-readable storage medium including instructions that, when executed on the one or more data processors, cause the one or more data processors to perform some or all of one or more methods disclosed herein (e.g., any of the first through sixteenth examples).
[0333] An eighteenth example includes a computer program product tangibly embodied in a non-transitory machine-readable storage medium that includes instructions configured to cause one or more data processors to perform some or all of one or more methods disclosed herein (e.g., any of the first through sixteenth examples).
Claims
1. A method implemented by a computer system, comprising: receiving a subject-specific data set corresponding to a particular subject, the subject-specific data set comprising: Type of hemophilia; Treatment type; Demographic data; and A photograph of a portion of a particular subject, or information derived based on a photograph of a portion of a particular subject receiving a subject-specific dataset corresponding to the subject, the subject-specific dataset including or identifying: processing at least a portion of the subject-specific dataset with a classifier model to identify one or more population-level machine learning models from a set of population-level machine learning models, each of the set of population-level machine learning models including a machine learning model trained using a training set corresponding to a set of other subjects with hemophilia to predict a time course associated with hemophilia for each of the set of other subjects with hemophilia, the training set including static or deterministic data, or data corresponding to a time point or time period, of the set of other subjects; receiving one or more indications identifying one or more times that a treatment of the treatment type was administered to the subject; predicting a hemophilia-associated time course for the subject by inputting the one or more times into a population-level machine learning model of the one or more population-level machine learning models included in a data processing workflow; receiving a representation of sensor data collected at a device associated with the subject; determining a transformed data processing workflow in response to determining that a workflow transition condition is satisfied using the representation of the sensor data to generate a hemophilia-related prediction for the subject; generating a hemophilia-associated prediction for the subject by inputting a representation of the sensor data into the transformed data processing workflow; and outputting a result corresponding to said hemophilia associated prediction.
2. generating said hemophilia associated prediction, determining the transformed data processing workflow, comprising identifying a post-processing algorithm based on the sensor data; identifying an initial outcome using said data processing workflow; determining a post-processing function based on the sensor data; transforming the initial result using the initial result and a post-processing function; The method of claim 1 , comprising:
3. The method of claim 1 or 2, wherein determining the transformed data processing workflow comprises selecting an activation function.
4. 4. The method of claim 1, wherein the population-level machine learning model comprises a pharmacokinetic model.
5. processing the subject-specific dataset with the classifier model to identify the one or more population-level machine learning models; encoding the subject-specific data set; using a distance-based algorithm to identify distances between the encoding of the subject-specific data set and encodings of other subject-specific data sets included in the training set; The method of any one of claims 1 to 4, comprising:
6. adding at least the hemophilia-associated predictions to the predicted hemophilia-associated time course, wherein one or more predicted factor levels comprise predicted factor levels; receiving another representation of new sensor data collected at the device associated with the subject; and determining to resume use of the data processing workflow based on the alternative representation; generating another hemophilia-related prediction for the subject using the data processing workflow and the appended time course; outputting another result corresponding to said another hemophilia-associated prediction; The method of claim 1 , further comprising:
7. generating said hemophilia associated prediction, determining one or more states or interim variables calculated for a recent time step by the population-level machine learning model; and configuring the transformed data processing workflow to use the one or more state or temporary variables.
8. The method of claim 1 , wherein the sensor data represents a movement of the subject.
9. The method of claim 1 , wherein the sensor data represents a physiological attribute of the subject.
10. 10. The method of any one of claims 1 to 9, wherein the predicted hemophilia-related time course comprises a predicted level of a factor, therapeutic agent or active ingredient for each of a set of time points, and the hemophilia-related prediction comprises another predicted level of the factor, therapeutic agent or active ingredient associated with another time point.
11. 10. The method of any one of claims 1 to 9, wherein the predicted hemophilia-related time course comprises, for each of a set of time points, a predicted clotting time or a probability that an abnormal bleeding event will occur, and the hemophilia-related prediction comprises another predicted clotting time or another probability that an abnormal bleeding event will occur associated with another time point.
12. 12. The method of any one of claims 1 to 11, wherein the results indicate a recommendation to reduce the subject's activity level.
13. 12. The method of any one of claims 1 to 11, wherein the results indicate a recommendation to treat the subject in view of an increased probability of abnormal bleeding.
14. The method of claim 1 , wherein the result corresponds to a recommendation to seek medical attention for bleeding displayed in the photograph.
15. 15. The method of any one of claims 1 to 14, wherein the subject-specific data set comprises data identified in an electronic medical record corresponding to the subject.
16. 16. The method of claim 1, wherein the results are output on the subject's device via a chatbot.
17. 1. A system comprising: one or more data processors; 17. A system comprising: a non-transitory computer-readable storage medium containing instructions that, when executed on the one or more data processors, cause the one or more data processors to perform the method of any one of claims 1 to 16.
18. A computer program comprising instructions adapted to cause one or more data processors to carry out a method according to any one of claims 1 to 16.
Citation Information
Patent Citations
Physiological sensing and therapeutic delivery systems and methods - Patents.com
JP2020509465A
Drug Monitoring Tools
JP2020510899A
Population pharmacokinetics tools and uses thereof
US20160306945A1
