Abstract. This article discusses the project of a web-based Medical Decision Support System (CBSA) designed to predict the risk of death in ICU patients. The proposed solution combines modern approaches to machine learning with asynchronous web architecture and intelligent interactive interface. The system uses a microservice approach and uses Django and WebSocket channels. This approach provides an excellent user interface and allows you to process a large number of parallel connections in real time. The clinical data from the MIMIC-IV kit served as the basis for training the analytical core of the system. It includes a multi-stage data processing pipeline with skip imputation, feature engineering, and ensemble modeling based on LightGBM gradient boosting. The experimental results showed that the model has a high predictive efficiency (AUC-ROC 0.982) while maintaining the correct calibration of probabilistic estimates. The interpretation of forecasts by the SHAP method increased the confidence of clinicians and explained the key clinical factors. Special attention is paid to the support of multimodal input data, such as medical documents in PDF and Excel formats, as well as text messages, which makes the system more suitable for clinical operations.

Keywords: MIMIC-IV, web application architecture, Django channels, machine learning, mortality prediction, interpretability of models, WebSocket, medical informatics, clinical decision support system.