A Clinical Decision Support System for Post-Surgical Cardiovascular Remote Monitoring.

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Decision Support Systems
• Cardiovascular Monitoring

Background:

• Post-surgical cardiovascular monitoring presents significant data integration challenges for clinical teams.
• Current clinical decision support systems inadequately synthesize patient history, intraoperative data, and monitoring streams.
• Lack of transparency in existing systems hinders trust and clinical adoption.

Purpose of the Study:

• To present a novel Retrieval-Augmented Generation (RAG) architecture for post-surgical cardiovascular monitoring.
• To focus on evidence traceability and practical workflow integration in clinical decision support.
• To develop a system offering visible and challengeable reasoning for clinical recommendations.

Main Methods:

• A three-layer RAG architecture was developed: retrieval, augmentation, and generation.
• The retrieval layer creates 768-dimensional representations of clinical scenarios.
• The augmentation layer uses ensemble methods for data integration and probability estimation.
• The generative layer employs a fine-tuned BERT classifier and Gemini 2.5 Pro for actionable recommendations.
• Components were prototyped using MIMIC-III and MIMIC-III-Ext-PPG datasets.

Main Results:

• The system architecture was successfully prototyped on de-identified public datasets.
• A structured evaluation framework was proposed to assess technical performance and clinical applicability.
• Key technical targets include >90% retrieval precision, <3s response time, and >85% clinical appropriateness rating.

Conclusions:

• A RAG architecture for post-surgical cardiovascular monitoring was described, linking recommendations to source documents for transparency.
• The proposed system enhances evidence traceability and clinical workflow integration.
• A structured evaluation framework is outlined to guide the system toward clinical validation and adoption.