Scenario modeling of the drug prescription process for children: application of machine learning methods

Objective: determining the most appropriate machine learning method to solve the problem of drug prescribtion for children, evaluating its performance and potential for implementation into scenario modeling systems of the pharmaceutical care structure.

Material and methods. The study was based on data on drug prescription for children from medical information systems of Moscow clinics for the period from January to December 2023 including information about patients, the date of treatment, diagnoses, prescribed medications and the doctor's specialty. Preliminary data processing enabled to extract additional features and define the process as a multi-label classification task. The following model architectures were developed and validated: fully connected neural network (FCNN), convolutional neural network (CNN), One-vs-Rest (OvR) classifier, eXtreme gradient boosting classifier (XGBC), and RandomForestClassifier (RFC). The models were evaluated using area under curve (AUC) of receiver operating characteristic (ROC), F1-measure metrics and Custom Accuracy metrics.

Results. The XGBC model showed the best results for all tasks and metrics. After optimizing the model and dataset, the AUC ROC reached 0.9993, the F1-measure was 0.8318, and Custom Accuracy metric was 0.8548. The model effectively predicted the prescription of drugs with similar pharmacological effects, allowing us to evaluate the structure of pharmaceutical care within a specific scenario. Optimization of the data and model increased the accuracy of predictions up to 85%.

Conclusion. The XGBC model proved to be the most appropriate for solving the problem of scenario modeling of drug prescribtion. The identified problems with predicting similar drugs validate the demand for further improvement of the model and data. Concurrently, the results obtained attest the potential of integrating machine learning methods into scenario modeling systems for pharmaceutical care.

1. Romanov I.A. Machine learning as a competitive advantage of the company. Moscow Economic Journal. 2022; 7 (3): 42 (in Russ.). https://doi.org/10.55186/2413046X_2022_7_3_141.

2. Ksenofontov D.M. Scenario modeling of the economic policy epidemiological effects. Scientific Papers: Institute of National Economic Forecasting of RAS. 2020; 18: 542–65 (in Russ.). https://doi.org/10.47711/2076-318-2020-542-565.

3. Tsatsulin A.N., Tsatsulin B.A. Scenario approach to building predictive models for the development of regional health systems. St. Petersburg State Politechnical University Journal. Economics. 2021; 14 (2): 115–36. https://doi.org/10.18721/JE.14208.

4. Aksenova E.S., Evdokimov D.S., Katasonova K.A. An improved agentbased model with scenario modeling functional and digital twin properties for forecasting socioepidemiological-economic processes in the regions of Russia. Artificial Societies. 2023; 18 (4) (in Russ.). https://doi.org/10.18254/S207751800028782-9.

5. Komkov A.A., Mazaev V.P., Ryazanova S.V., et al. Application of the program for artificial intelligence analytics of paper text and segmentation by specified parameters in clinical practice. Cardiovascular Therapy and Prevention. 2023; 21 (12): 3458 (in Russ.). https://doi.org/10.15829/1728-8800-2022-3458.

6. Gusev A.V., Novitskiy R.E., Ivshin A.A., Alekseev A.A. Machine learning based on laboratory data for disease prediction. FARMAKOEKONOMIKA. Sovremennaya farmakoekonomika i farmakoepidemiologiya / FARMAKOEKONOMIKA. Modern Pharmacoeconomics and Pharmacoepidemiology. 2021; 14 (4): 581–92 (in Russ.). https://doi.org/10.17749/2070-4909/farmakoekonomika.2021.115.

7. Koledachkin A.A. Using simulation and simulation in testing: prospects with use of AI. Vesynik nauki / Bulletin of Science. 2024; 5 (9): 513–40 (in Russ.).

8. Orji U., Ukwandu E. Machine learning for an explainable cost prediction of medical insurance. Machine Learn App. 2024; 15: 100516. https://doi.org/10.1016/j.mlwa.2023.100516.

9. Narkevich A.N., Vinogradov K.A., Paraskevopulo K.M., Grjibovski A.M. Intelligent data analysis in biomedical research: artificial neural networks. Ekologiya cheloveka / Human Ecology. 2021; 28 (4): 55–64 (in Russ.). https://doi.org/10.33396/1728-0869-2021-4-55-64.

10. Golounina O.O., Belaya Zh.E., Voronov K.A., et al. Machine learning methods in differential diagnosis of ACTH-dependent hypercortisolism. Problems of Endocrinology. 2024; 70 (1): 18–29 (in Russ.). https://doi.org/10.14341/probl13342.

11. Firyulina M.A., Kashirina I.L., Gafanovich E.Y. Using of machine learning methods in prescribing hypertension therapy. Modeling, Optimization and Information Technology. 2020; 8 (4): 4 (in Russ.). https://doi.org/10.26102/2310-6018/2020.31.4.025.

12. Silva P., Rivolli A., Rocha P., et al. Machine learning for drugs prescription. In: Yin H., Camacho D., Novais P., Tallón-Ballesteros A. (Eds.) Intelligent Data Engineering and Automated Learning – IDEAL 2018. Part I. Springer; 2018: 548–55. https://doi.org/10.1007/978-3-030-03493-1_57.