Method for tracking medication intake and patient compliance with doctor's recommended course of therapy, implemented through digital platform

Abstract

FIELD: information technology.SUBSTANCE: method for monitoring a patient's compliance with a course of drug therapy prescribed by a doctor. A proposed method is implemented through a digital platform and includes the registration of doctors and patients using individual identifiers, the creation and completion of adaptable questionnaires, and the collection of information on the purchase and administration of medications. The digital platform uses artificial intelligence and machine learning technologies to automatically match patient-prescribed medications with physician prescriptions for composition, dosage, and timing, including partial substitutions and equivalents, and to predict the risk of non-adherence based on the patient's behavioral and medical data. If any violations are detected, the system generates a notification to the doctor. The patient receives a refund of a portion of the cost of the medications if all conditions are met.EFFECT: automated monitoring of therapy compliance through intelligent analysis of pharmacy transactions and medical data.1 cl

Description

[0001] The invention relates to the field of information technology and concerns a method for monitoring a patient's compliance with a course of drug therapy prescribed by a doctor.

[0002] Technical result

[0003] Ensuring automated monitoring of patient compliance with the prescribed course of drug therapy by comparing pharmacy transaction data with medical prescriptions.

[0004] The prior art includes digital platforms and solutions that enable patient medication monitoring and interaction with their physicians. For example, publication US20190018932A1 describes a consumption transaction analysis system that uses payment network data to evaluate drug purchases and provide recommendations. Also, US20150324525A1 presents a medical data management system where patients independently enter information, including completing surveys and uploading documents.

[0005] However, existing solutions lack a mechanism for intelligently matching pharmacy transactions with physician prescriptions, especially in cases of partial matches, selection of analogs, or drug substitutions. Predictive forecasting of the risk of non-adherence to therapy based on the analysis of patient behavioral and medical data is also not implemented. Static analysis algorithms without machine learning are used, which are not adaptable to individual patient behavior patterns. Assessment of therapy compliance is not accompanied by automatic alerts to the physician based on deviations calculated using intelligent methods. Therefore, existing solutions do not provide reliable, adaptive, and automated monitoring of patient compliance with physician recommendations based on AI technologies, limiting their applicability for dynamic pharmacovigilance and digital clinical support.

[0006] Disclosure of invention

[0007] The invention relates to the field of medical digital technologies and is designed to monitor patient compliance with prescribed medication therapy using a platform implementing artificial intelligence and machine learning methods. The technical result consists of automated compliance monitoring through intelligent comparison of pharmacy transaction data with physician prescriptions, as well as predicting the risk of non-compliance.

[0008] The digital platform on which the method is implemented includes:

[0009] - modules for registration of doctors and patients,

[0010] - mechanism for issuing and activating questionnaires,

[0011] - block for collecting data on the purchase of medicines (by scanning QR codes of receipts and pharmacy APIs),

[0012] - AI analysis component, including a model for matching purchased drugs with prescriptions,

[0013] - a model for predicting the risk of non-adherence to therapy based on behavioral and medical patterns.

[0014] The system automatically compares purchased medication data with the prescribed parameters: active ingredient, dosage, frequency of administration, and course duration. In the event of discrepancies (e.g., purchase of an equivalent, incomplete quantity, or delay in the prescribed course), an intelligent physician alert module is activated. Furthermore, based on survey history, missed purchases, response rate, and other factors, the platform predicts the likelihood of non-adherence in a specific patient. This allows the physician to adjust the course before clinical complications arise. The platform motivates patients to adhere to therapy through a rewards system: upon fulfilling all conditions (purchase, surveys, and prescribed course), the patient receives a partial refund of the medication cost.

[0015] Implementation of the invention

[0016] The method is implemented through a cloud-based digital platform containing a mobile app and a chatbot (e.g., in Telegram). The server component is hosted in the cloud. The software implementation is built on PHP and Python (server) and Flutter (client).

[0017] 1. Doctor registration. Doctors receive a unique QR code (based on a timestamp) issued by a platform representative. After registration, doctors gain access to profile creation and questionnaire selection.

[0018] 2. Questionnaire development. Questionnaires are selected from a set compiled by specialized medical experts. They are divided into stages: pre-therapy, during therapy, and post-therapy. Questions can be multiple-choice, scaled, or open-ended.

[0019] 3. Patient registration. The patient enters a digital code or scans the QR code provided by the doctor. This confirms the in-person visit. The patient gains access to the questionnaire and a list of prescribed medications.

[0020] 4. Medication purchase. The patient scans the QR code on the receipt, or the system receives information via the pharmacy chain's API. The composition, dosage, form, and quantity of the medication are analyzed.

[0021] 5. AI-powered assignment and purchase matching. The platform uses an AI model (e.g., based on a decision tree or a simple neural network), trained on transaction and assignment history, to determine:

[0022] - complete or partial replacement of the drug,

[0023] - dosage compliance,

[0024] - course duration,

[0025] - the most effective trade names within the designated INN,

[0026] - permissible deviations.

[0027] The model takes into account possible INN equivalents, therapeutic groups, and analogs. If there is a discrepancy, the system classifies the degree of deviation.

[0028] 6. Non-adherence risk prediction. Based on behavioral data (delays, omissions, incomplete survey completion, past course history) and medical characteristics (nosology, age, gender, severity of condition), a machine learning algorithm (e.g., logistic regression or gradient boosting) is used to predict the likelihood of non-adherence.

[0029] 7. Physician Notification. If risk thresholds are exceeded or recommendations are not followed, the physician will automatically receive a notification with details of the violation and risk prognosis.

[0030] 8. If the conditions are fully met, the patient will receive a partial refund of the cost of the medications, implemented through the bonus system integrated into the app.

[0031] The claimed method uses artificial intelligence and machine learning technologies to implement two key intelligent functions: automatic matching of pharmacy transactions with doctor's prescriptions, predicting the risk of patient non-compliance with therapy, and automatic matching of purchased drugs with medical prescriptions.

[0032] At the stage of receiving data on the purchase of drugs (by scanning the QR code of the cash register receipt or via the pharmacy chain API), the system must determine whether the drug purchased by the patient corresponds to the one previously prescribed by the doctor.

[0033] The model is based on a decision tree trained on a sample of pharmacy transactions and their matching to prescriptions (classification task). For more complex cases, gradient boosting (e.g., XGBoost) can be used.

[0034] Input parameters:

[0035] • INN (International Nonproprietary Name)

[0036] • Trade name

[0037] • Dosage (mg or mg / ml)

[0038] • Release form (tablets, capsules, solution, etc.)

[0039] • Number of packages

[0040] • Prescribed frequency and duration of administration

[0041] • Date of appointment and date of purchase

[0042] Logic example:

[0043] • If prescribed: Ibuprofen 200 mg, 3 times a day, 7 days

[0044] • Purchased: Ibuprofen 400 mg × 10 tablets

[0045] • The model compares the INN, the dosage exceeds the prescribed one - acceptable if the total daily dose is within the normal range, the quantity is not enough for the full course

[0046] • Conclusion: Partial match, warning to the doctor

[0047] Result:

[0048] The model assigns the transaction status:

[0049] • "A Complete Match"

[0050] • "Partial Match"

[0051] • "Does not match"

[0052] • "Replacement with a synonymous drug"

[0053] Data and sources used:

[0054] • Nomenclature of drugs (PharmRegister, RLS, GRLS)

[0055] • Synonyms and equivalents module (generated manually or based on ontologies)

[0056] • Patient's appointment and purchase history

[0057] Predicting the likelihood that a patient will violate therapy (not purchase the drug, not complete the questionnaire, interrupt the course, etc.)

[0058] Model type

[0059] A logistic regression model or a lightly trained neural network (MLP) with a constrained architecture is used. Random forest or CatBoost-based implementations are also possible, depending on the availability of the training set.

[0060] Logistic Regression Formula (Example):

[0061] CopyEdit

[0062] P(violation) = 1 / (1 + e^-(β0 + β1⋅X1 + β2⋅X2 + … + βn⋅Xn))

[0063] where

[0064] X1, X2 … Xn - patient characteristics,

[0065] β0, β1 … βn - trained model weights

[0066] Features:

[0067] • Age, gender, diagnosis

[0068] • History of previous deviations

[0069] • Frequency of doctor visits

[0070] • Number of questionnaires completed on time

[0071] • Time between appointment and purchase

[0072] • Missed or delayed purchases

[0073] • Cashback history

[0074] • Number of interactions with the system (chatbot, application)

[0075] Result:

[0076] • Probability value from 0 to 1

[0077] • Classification based on threshold (e.g. 0.75): low risk, medium risk, high risk of non-compliance

[0078] • When the threshold is exceeded, the doctor receives an automatic notification

[0079] The training set is formed using real, anonymized data from users who have already used the platform: interaction history, surveys, purchases, and violations. Models are trained externally and then exported to production as pre-trained models (.pkl, .joblib, .onnx). Predictions are executed on the server for each new event (survey, transaction, visit).

[0080] Practical Application Example

[0081] 1. The patient received a prescription: loratadine 10 mg 1 time per day, 10 days

[0082] 2. After 3 days I purchased: Cetirizine 10 mg, 5 tablets

[0083] 3. AI matching model:

[0084] • INN: does not match → check for therapeutic analogue

[0085] • Recognized: Partial Replacement

[0086] • Insufficient duration

[0087] • Classification: "partial compliance"

[0088] 4. Prediction Model: Patient has previously discontinued therapy. Signs: purchase delay, insufficient volume. Prediction: risk of non-adherence - high (0.82)

[0089] 5. The system notifies the doctor with details