BehaviorSFT: Behavioral Token Conditioning for Clinical Agents Across the Proactivity Spectrum

2025-05-27

AI Summary

Clinical LLM agents suffer from behavioral adaptation imbalance: strong passive responsiveness (e.g., diagnostic reasoning) but deficient proactive intervention (e.g., identifying unmentioned critical information gaps or risks). To address this, we propose “behavioral tokens”—explicit, structured markers that dynamically guide models along a clinical proactivity spectrum, enabling clinically appropriate balance between initiative and restraint. Our method integrates behavioral-token-conditioned supervised fine-tuning, a multi-granularity evaluation framework (BehaviorBench), and a blinded clinical expert assessment protocol. Experiments show our model achieves 97.3% Macro F1 on BehaviorBench; Qwen2.5-7B-Ins attains 96.5% accuracy on proactive tasks, and blinded expert evaluation confirms its interaction style closely mirrors real-world clinical practice. This work pioneers modeling behavioral controllability as a structured token learning problem, establishing a novel paradigm for behavior alignment in clinical LLM agents.

Technology Category

• Cognitive Modeling & Cognitive Systems: Adaptive Behavior
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Intelligent Robots: Behavior Learning & Control

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Agentic search
• User Modeling, Personalization and Recommendation: Metrics for user behavior and evaluating success

Abstract

Large Language Models (LLMs) as clinical agents require careful behavioral adaptation. While adept at reactive tasks (e.g., diagnosis reasoning), LLMs often struggle with proactive engagement, like unprompted identification of critical missing information or risks. We introduce BehaviorBench, a comprehensive dataset to evaluate agent behaviors across a clinical assistance spectrum, ranging from reactive query responses to proactive interventions (e.g., clarifying ambiguities, flagging overlooked critical data). Our BehaviorBench experiments reveal LLMs' inconsistent proactivity. To address this, we propose BehaviorSFT, a novel training strategy using behavioral tokens to explicitly condition LLMs for dynamic behavioral selection along this spectrum. BehaviorSFT boosts performance, achieving up to 97.3% overall Macro F1 on BehaviorBench and improving proactive task scores (e.g., from 95.0% to 96.5% for Qwen2.5-7B-Ins). Crucially, blind clinician evaluations confirmed BehaviorSFT-trained agents exhibit more realistic clinical behavior, striking a superior balance between helpful proactivity (e.g., timely, relevant suggestions) and necessary restraint (e.g., avoiding over-intervention) versus standard fine-tuning or explicit instructed agents.

Problem

• LLMs struggle with proactive clinical engagement.
• Need for dynamic behavioral adaptation in clinical agents.
• Improving balance between proactivity and restraint in LLMs.

Innovation

• Behavioral tokens condition LLMs dynamically
• BehaviorBench dataset evaluates clinical agent behaviors
• BehaviorSFT improves proactive and reactive performance