Designing and evaluating large language model-enabled clinical decision support for heart failure: a modular and risk-tiered framework - Scorecard - MDSpire
Advertisement
Designing and evaluating large language model-enabled clinical decision support for heart failure: a modular and risk-tiered framework
Clinical Scorecard: Developing and Assessing a Modular, Risk-Based Framework for Clinical Decision Support in Heart Failure Using Large Language Models
At a Glance
Category
Detail
Condition
Key Mechanisms
Utilization of large language models (LLMs) for clinical decision support through data synthesis and patient-specific reasoning.
Target Population
Care Setting
Key Highlights
HF care involves a sequence of decisions under uncertainty.
LLMs can process free-text questions and summarize electronic health records.
The Heart Failure Intelligent Agent (HF-IA) is proposed as a modular framework.
Different clinical tasks require distinct data inputs and error tolerances.
Evaluation of HF-IA should focus on clinical decisions supported and potential harm from errors.
Guideline-Based Recommendations
Diagnosis
Diagnosis depends on symptoms, examination, natriuretic peptides, imaging, and exclusion of mimics.
Management
Guideline-directed medical therapy (GDMT) safety and optimization must consider patient-specific factors.
Monitoring & Follow-up
Monitoring should include laboratory surveillance and assessment of worsening HF.
Risks
Risk of missing critical conditions such as hyperkalemia or worsening kidney function.
Patient & Prescribing Data
Patients with various types of heart failure including HFrEF, HFmrEF, and HFpEF.
Treatment optimization requires consideration of blood pressure, kidney function, and patient preferences.
Clinical Best Practices
Implement a modular design for clinical decision support systems.
Ensure data exchange adheres to health-system standards like FHIR.
Define decision context and required data elements before making recommendations.
