DIAGNOSIS AND EFFECTIVENESS OF MONITORING ATTENTION DEFICIT HYPERACTIVITY DISORDER

Abstract

A method and a system are provided for taking biomarker measurements of patients who have ADHD. Mathematical analysis (e.g., pattern recognition, machine learning and AI algorithms) of the biomarker measurements is used to create a unique personal prediction model and data set for an individual patient. The unique personal data set is used to diagnose and monitor a particular problem of the individual patient associated with ADHD, or to recommend a treatment for a particular problem of the individual patient associated with ADHD, or to predict an outcome of a treatment for a particular problem of the individual patient associated with ADHD.

Claims

1. A method for dealing with Attention Deficit Hyperactivity Disorder (ADHD) comprising: taking biomarker measurements of patients who each have ADHD; using said biomarker measurements to create a cluster of ADHD patients, wherein the patients in said cluster have similar attributes according to at least one of sex, age range and comorbidity) and biomarker measurements, so that within said cluster a variance in said biomarker measurements used to create said cluster is smaller than a variance in biomarker measurements of patients not in said cluster; and processing differences between said biomarker measurements of the patients in said cluster to create a unique personal data set for an individual patient in said cluster; and using the unique personal prediction model and data set to diagnose and monitor a particular problem of said individual patient associated with ADHD, or to recommend a treatment for a particular problem of said individual patient associated with ADHD, or to predict an outcome of a treatment for a particular problem of said individual patient associated with ADHD.

2. The method according to claim 1, wherein the step of processing differences is done by pattern recognition, machine learning or AI algorithms.

3. The method according to claim 1, comprising steps as follows: assigning other patients to one of various patients' cluster depending on biomarker measurements of said other patients; and using pattern recognition of differences between the biomarker measurements of said other patients to create a unique personal data set for an individual patient of said other patients.

4. The method according to claim 3, wherein the unique personal data set further comprises calibration and calculation of a personal baseline.

5. The method according to claim 1, further comprising performing ongoing treatment, including delivering personal analysis of medical treatment effect and predictions using a personal pattern based on the unique personal data set.

6. The method according to claim 1, further comprising performing ongoing cluster calibration using automation machine learning and shifts between clusters.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The present invention will be understood and appreciated more fully from the following detailed description, taken in conjunction with the drawings in which:

[0015] FIGS. 1 and 2 are simplified illustrations of personal calibration weighting used in a non-limiting embodiment of the methods and system of the present invention, wherein FIG. 1 shows the Pearson correlation between a subset of possible features, showing that the markers are not correlated, hence are valuable predictors, and FIG. 2 shows the contribution of each feature to the first discriminant vector, illustrating the personal model calibration per specific patient. In FIG. 2, IBI is inter-bit interval (between heart rate peaks), EDA is galvanic skin response and acceleration is that of the patient or a portion of the patient.

[0016] FIG. 3 is a simplified illustration of using a personal calibration model to setup a personal baseline for monitoring medical treatment effectiveness.

[0017] FIG. 4 is a simplified illustration of implementation of the Patient 1 personal model on Patient 2, both from the same cluster, and shows that this proved to be incapable of separating the significance of the biomarkers and incapable of setting up a personal baseline without personal calibration.

[0018] FIG. 5 is a simplified illustration of assigning a personal calibration to the search pattern for Patient 2, which proved successful in setting up a personal baseline in accordance with the physician diagnosis.

[0019] FIGS. 6 and 7 are simplified illustrations of two different examples of personal, ongoing analysis using biomarkers, in accordance with non-limiting embodiments of the invention.

DETAILED DESCRIPTION OF EMBODIMENTS

[0020] As mentioned above, in contrast to the prior art, the present invention uses mathematical analysis of biomarker measurements to define clusters of patients and, for each patient, to create a personal profile, baseline and prediction pattern. The unique personal model is calibrated by machine and deep learning techniques which are specifically assigned per patient, for the provision of output that assess the clinical and behavioral indicators of ADHD.

[0021] The personal method may include the followings steps:

[0022] A: Cluster Allocation: [0023] Using patients' measured biomarker data to build an ensemble of predictive models for Diagnostic and Statistics manual-V (DSM-V) compliant clinical ADHD diagnosis scores used by physicians. This step uses biomarker data measured during the DSM-V compliant clinical diagnosis process, and would use a host of pattern recognition, machine learning and AI algorithms. [0024] Using pattern recognition, machine learning and AI algorithms to cluster patients based on external features (such as sex, age, comorbidities) and internal features (biomarker data), and to modify the predictive model to each cluster.

[0025] B: Personal Baseline Calibration and Continues Monitoring: [0026] Using pattern recognition, machine learning and AI algorithms to calibrate model features (e.g., biomarkers weighting and/or selecting the importance of a particular physiological feature) and tailor a predictive model to each individual, based on its cluster association and on its unique combination of external and internal features. [0027] Using the trained personalized predictive model to automatically and passively monitor patients for their response to ADHD treatment.

[0028] C: Ongoing Calibration: [0029] The predictive models are automatically updated and improved (continuous learning) with more biomarker data measured for each patient.

[0030] The above setups of the indicators are based on the criteria established by the American Medical Association (AMA) as described in the Diagnostic and Statistics manual-V (DSM-V).

[0031] The method also assesses the objective impact of other treatment options for ADHD patients, such as neurofeedback and various behavioral treatments.

[0032] The output of the method is achieved by demonstrating personal machine learning based prediction model and personal baseline of the individual with and without effective treatment, pertaining to the DSM-V criteria, including relation to each of the ADHD disorder aspects: attention, hyperactivity and impulsivity. This output supports physicians' evaluation of treatment success and provides additional, ongoing analysis which recommends possible advantageous treatment modifications.

[0033] The method uses pattern recognition, machine learning, AI algorithms, and other techniques. It is based on biomarker measurements (described below) and external information (such as sex, age, etc.). A predictive model is designed upon collection of all samples, then modified within each cluster of patients, and then further modified to match the specific personal pattern of every patient. Eventually, every individual patient is characterized by a unique, personalized, predictive model.

[0034] The following is a description of one non-limiting embodiment of the method and system of the invention.

[0035] First, a combination of the following of biomarkers' measurements may be automatically collected and recorded by a device, such as a wearable device:

TABLE-US-00001 PPG nano watt Inter beat interval Sec Heart rate (BPM) Sec Skin temperature Celsius ° C. Accelerometer gravity (g) (or motion sensor) Heart Rate Variability hertz (Hz) Gyro degrees per second (°/s) Respiratory rate BRA Blood pressure mmHg EEG μV Skin conductance Hz ECG mV/mS Blood oxygen saturation SaO.sub.2 Systemic vascular resistance PRU

[0036] Afterwards, patients may be assigned to a pre-defined cluster using the biomarkers and physiological attributes. For example, initial ongoing biomarkers measurement, together with patient's attributes (e.g., girl, 15 years old, Afro-American, with birth difficulties) may be used to assign a patient into a designated cluster. The method may start with a certain number of clusters, and in parallel there may be ongoing cluster calibration and addition of new clusters.

[0037] The methods for clustering and for building predictive models may be based on an ensemble of models, including linear models (e.g., Fisher Discriminant Analysis and Linear Discriminant analysis) and nonlinear ones (e.g., neural network classifiers, random forests). The ensemble of models may be adjusted and constantly re-trained whenever more patient data become available.

[0038] The method may also include measuring biomarkers in parallel to physician diagnostics. For example, the measurement may be carried out together with DSM-V based clinical evaluation to associate biomarkers with attention, hyperactivity and impulsivity.

[0039] The method may also include calculation of a personal calibration model. The method may assign a personal calculation model to setup baseline and ongoing prediction models. (Two examples are given below.)

[0040] The method may also include setting up personalized biomarkers weightings and significance level. For example, these may include personal biomarker weighting and/or selecting the importance of a particular feature while assessing non-linear models.

[0041] The method may also include both monitoring ongoing treatment and/or preliminary ADHD diagnosis. This may include utilizing personal prediction patterns and performing ongoing analysis using machine learning based process.

[0042] The method may also include ongoing cluster calibration and shifts between clusters. For example, there may be ongoing evaluation of the patient to decide whether he/she should shift between clusters (e.g., age group changed from 7-10 to 11-14), or there may be ongoing creation of new clusters or sub-clusters.

[0043] Personal Process—Samples Demonstrating the Significance of Personal Modeling:

[0044] The importance of personal modeling is now explained with reference to FIGS. 1-7. The figures relate to measurements of two ADHD patients, both from the same cluster (in which Fisher Discriminant Analysis was used as a search method). A personal baseline was created for both patients. The presented baseline calculations illustrate three stages of examination—without medication, with ineffective medication and with effective medication.

[0045] Patient 1:

[0046] Personal calibration weighting was assigned as seen in FIGS. 1 and 2. FIG. 1 shows the Pearson correlation between a subset of possible features. FIG. 2 shows the contribution of each feature to the first discriminant vector.

[0047] Using the personal calibration model, the personal baseline was set up in accordance with the physician DSM-V diagnosis (X represents medication effectiveness). This is shown in FIG. 3.

[0048] Patient 2:

[0049] After training the model and applying it on Patient 2 (Patients 1 and 2 are both from the same cluster), direct implementation of the Patient 1 personal calibration (weighting of biomarkers) on Patient 2 proved incapable of creating DSM-V compliant data clusters and incapable of setting up a personal baseline (as both patients are from the same cluster one can notice a slight, but not valid, separation). This is shown in FIG. 4.

[0050] After re-calibrating and assigning a unique personal model for Patient 2, the invention was able, as in Patient 1, to set up the personal baseline in accordance with the physician DSM-V diagnosis (X represents medication effectiveness). This is shown in FIG. 5.

[0051] In another clinical trial, a first set of patients were positively diagnosed with ADHD, and some were given an effective medical treatment and some were given no treatment. A second set of patients were positively diagnosed with ADHD and some were given an ineffective medical treatment and some were given no treatment. Both the first set and the second set of patients were analyzed by the system and method of the invention. The system and method correctly identified 100% of the time which patients in the first set were given an effective treatment and close to 100% of the time which ones were given no treatment. The system and method correctly identified close to 100% of the time which patients in the second set were given an ineffective treatment and close to 100% of the time which ones were given no treatment.

[0052] Technical Process:

[0053] All of the biomarker measurements are automatically collected and recorded by a wearable device. Data may be transformed to a centralized hub. Data may be stored within the wearable device memory for later downloading. Thereafter, the data may be transmitted or uploaded to a digital data repository (cloud or other) and analyzed by implementing machine and deep learning techniques.

[0054] System output is a translation of the processed analysis into medical indicators and status reports and is provided to physicians through a medical web application, and to the individuals (patient, child, parents etc.) through a mobile application.

[0055] Analysis is seamlessly transferred to the individual's smartphone (or to other communication device), or via a transmitting, add-on dedicated unit for real-time data transmission.

[0056] The web application provides the physicians the results of the analysis which include continuous diagnosis and monitoring of ADHD, which may include indication of medication and dosing effectiveness, actual duration of effect, long term patterns and trends and other clinical aspects (e.g., effectiveness of other clinical treatments, effect on the quality of sleep), predictions related to recommended treatment changes, level of patient medication usage, adherence and overall treatment effectiveness snapshots. Thus, the method provides medical professionals big data insights into real-life patient clinical and behavioral patterns.

[0057] In addition, the method may use patient clustering analysis to provide deep learning-based recommendations (e.g., preliminary type of recommended medication to be allocated to each cluster for the initial treatment process).

[0058] The application provides a clear view of the medication term or effectiveness during the day, and helps prepare and better manage daily activity of the individual, using predictions of ADHD symptoms' levels.

[0059] System's Elements/Components:

[0060] The system implementing the method is an IoT platform which may include a user-friendly wearable device. The wearable device may include, without limitation, a sensor hub that include sensors for all biomarkers mentioned in the above table, a user interface (e.g., including graphics, sound and vibration methods of interface), and a software application. For example, without limitation, the application may operate the related services and processes to read the sensors, perform the analysis and send the real time feedback through the user interface to the user. This software application may also communicate with the cloud-based software on a real time or a periodic basis, to update the data measured from the sensors in the device.

[0061] The system may further include a cloud-based SW tool or other data repository, which aggregates and analyzes the sensors data, updates related algorithms and generates output in dash boards and reports to stakeholders.

[0062] The system may further include a web application interface for medical professionals—providing output on diagnosis of ADHD, indication of medication and dosing effectiveness, predictions of recommended treatment changes, level of patient medication usage, adherence and overall treatment effectiveness snapshots.

[0063] The system may further include a mobile application interface for users (individuals who have ADHD and stakeholders (e.g. parents) providing a real time view on the individual's ADHD condition throughout the day, including medication cycle impact and phasing out timing, and predictions of ADHD personal patterns.

[0064] Method Outputs and Example of the Importance of Personal Analysis:

[0065] Tests of the methods of the invention have demonstrated their efficacy, resulting in outputs significantly correlated to the demonstrated clinical results, as indicated by physicians.

[0066] The following section provides examples of personal, ongoing analysis using biomarkers.

Examples: Output Samples

[0067] Example 1 (shown in FIG. 6): A case of a child treated with CONCERTA27, later switched to RITALIN LA20—classification outputs clearly demonstrate the different success levels of the two medications, aligned with physician's clinical observations (RITALIN: effective, CONCERTA: not). The personal baseline scale for this patient is: (1.5)—no medication, 1.5—effective medical treatment (and, thus, ˜0—ineffective medical treatment). The classification is based on multiple biomarkers measurements combined and a personalized formula determining the significance of each marker for this specific patient.

[0068] Example 2 (shown in FIG. 7): A description of the result of the processed biomarkers measurement pertaining to the duration of medication effectiveness demonstrating the point in time from which medication impact is drastically reduced. This was found highly correlated with clinical condition of the medication phasing out time, indicating a drastic increase of ADHD symptoms (the baseline for the initial 150 minutes is, as before, 1.5, later on deteriorating to (1.5)).

[0069] The method demonstrated significant correlation between the processed biomarkers measurements and the clinical results identified by physicians, in the subjects evaluated.