Emotion-Adaptive Large Language Model-Driven Clinical Decision Support: User Evaluation of the Empathic Clinical...

Area of Science:

• Artificial Intelligence in Healthcare
• Computational Behavioral Modeling
• Affective Computing

Background:

• Increasing cannabis use necessitates predictive models for usage patterns and health impacts.
• Opaque AI systems hinder user trust and interpretation of clinical decision support outputs.
• Existing explainable AI methods are often too technical for clinicians, causing frustration.

Purpose of the Study:

• To propose and evaluate an empathic clinical decision support system (empathic-CDSS).
• To integrate large language models (LLMs) with emotion recognition and explainability for transparent, adaptive, and emotionally attuned AI explanations.
• To enhance user confidence and trust in AI-assisted clinical decision-making.

Main Methods:

• Developed an empathic-CDSS integrating explainable AI, causal inference, and affective computing within an LLM framework.
• Inferred user affective states using the circumplex model of affect (valence and arousal).
• Dynamically adjusted explanation tone, structure, style, and complexity based on user emotions for personalized, plain-language outputs.

Main Results:

• The empathic-CDSS generated personalized, transparent explanations, revealing causal reasoning and enhancing user trust.
• Affect-based feedback allowed dynamic adaptation of explanation delivery to individual user needs.
• Participants reported significantly improved usability, clarity, satisfaction, and trust compared to a baseline system.

Conclusions:

• Introduced a transparent, trustworthy, and emotionally adaptive AI framework for clinical decision support.
• Uniting causal reasoning, affective sensing, and LLM explanations offers a novel direction for emotionally intelligent, user-centered AI.
• Potential applications include behavioral monitoring and personalized interventions, particularly for cannabis use and related health domains.