Designing and evaluating large language model-enabled clinical decision support for heart failure: a modular and...

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Decision Support Systems
• Cardiology

Background:

• Heart failure (HF) management involves complex, sequential clinical decisions.
• Current clinical workflows present challenges in synthesizing patient data and evidence.
• Large language models (LLMs) offer potential for data synthesis and patient-specific reasoning.

Purpose of the Study:

• To present a conceptual framework for an LLM-enabled clinical decision support system for heart failure care.
• To propose a modular, risk-tiered approach for integrating LLM functions.
• To outline requirements for the design, evaluation, and governance of such systems.

Main Methods:

• Conceptual framework development for the Heart Failure Intelligent Agent (HF-IA).
• Proposal of a modular, risk-tiered architecture for LLM functions.
• Recommendation of a multi-faceted evaluation strategy including node-level tests, case replay, prospective validation, and monitoring.

Main Results:

• The HF-IA framework is presented as a conceptual model, not a clinically validated tool.
• The framework details distinct requirements for different agent functions regarding data, standards, risk, and validation.
• A comprehensive evaluation approach is suggested to ensure safety and efficacy.

Conclusions:

• LLM-based clinical decision support for HF should be modular and risk-tiered, not a single autonomous agent.
• The proposed HF-IA framework clarifies design and evaluation needs for future LLM applications in HF care.
• Rigorous, multi-stage evaluation is crucial for developing effective and safe LLM-enabled HF decision support systems.