Academic

One Panel Does Not Fit All: Case-Adaptive Multi-Agent Deliberation for Clinical Prediction

arXiv:2604.00085v1 Announce Type: new Abstract: Large language models applied to clinical prediction exhibit case-level heterogeneity: simple cases yield consistent outputs, while complex cases produce divergent predictions under minor prompt changes. Existing single-agent strategies sample from one role-conditioned distribution, and multi-agent frameworks use fixed roles with flat majority voting, discarding the diagnostic signal in disagreement. We propose CAMP (Case-Adaptive Multi-agent Panel), where an attending-physician agent dynamically assembles a specialist panel tailored to each case's diagnostic uncertainty. Each specialist evaluates candidates via three-valued voting (KEEP/REFUSE/NEUTRAL), enabling principled abstention outside one's expertise. A hybrid router directs each diagnosis through strong consensus, fallback to the attending physician's judgment, or evidence-based arbitration that weighs argument quality over vote counts. On diagnostic prediction and brief hospita

Y
Yuxing Lu, Yushuhong Lin, Jason Zhang · · 1 min read · 3 views
#cs.AI #cs.CL #cs.MA

arXiv:2604.00085v1 Announce Type: new Abstract: Large language models applied to clinical prediction exhibit case-level heterogeneity: simple cases yield consistent outputs, while complex cases produce divergent predictions under minor prompt changes. Existing single-agent strategies sample from one role-conditioned distribution, and multi-agent frameworks use fixed roles with flat majority voting, discarding the diagnostic signal in disagreement. We propose CAMP (Case-Adaptive Multi-agent Panel), where an attending-physician agent dynamically assembles a specialist panel tailored to each case's diagnostic uncertainty. Each specialist evaluates candidates via three-valued voting (KEEP/REFUSE/NEUTRAL), enabling principled abstention outside one's expertise. A hybrid router directs each diagnosis through strong consensus, fallback to the attending physician's judgment, or evidence-based arbitration that weighs argument quality over vote counts. On diagnostic prediction and brief hospital course generation from MIMIC-IV across four LLM backbones, CAMP consistently outperforms strong baselines while consuming fewer tokens than most competing multi-agent methods, with voting records and arbitration traces offering transparent decision audits.

Executive Summary

This article proposes a novel approach to clinical prediction using large language models, addressing the issue of case-level heterogeneity. The Case-Adaptive Multi-agent Panel (CAMP) dynamically assembles a specialist panel tailored to each case's diagnostic uncertainty, enabling principled abstention outside one's expertise. CAMP outperforms strong baselines while consuming fewer tokens than most competing multi-agent methods. The approach provides transparent decision audits through voting records and arbitration traces.

Key Points

  • ▸ CAMP addresses the issue of case-level heterogeneity in clinical prediction using large language models.
  • ▸ The approach dynamically assembles a specialist panel tailored to each case's diagnostic uncertainty.
  • ▸ CAMP enables principled abstention outside one's expertise and provides transparent decision audits.

Merits

Strength in Addressing Case-Level Heterogeneity

CAMP effectively addresses the issue of case-level heterogeneity in clinical prediction, enabling more accurate and reliable predictions.

Efficient Token Consumption

CAMP consumes fewer tokens than most competing multi-agent methods, making it a more efficient approach to clinical prediction.

Demerits

Limited Generalizability

The performance of CAMP may be limited to the specific datasets and models used in the study, and further research is needed to evaluate its generalizability to other clinical scenarios.

Expert Commentary

The article presents a well-researched and novel approach to clinical prediction using large language models. The use of multi-agent systems and dynamic assembly of specialist panels is a promising direction for addressing case-level heterogeneity. However, further research is needed to evaluate the generalizability of CAMP to other clinical scenarios and to investigate its potential implications for policy and regulation. The provision of transparent decision audits through voting records and arbitration traces is a significant strength of the approach, addressing the need for explanation and transparency in AI decision-making in healthcare.

Recommendations

  • ✓ Further research is needed to evaluate the generalizability of CAMP to other clinical scenarios and to investigate its potential implications for policy and regulation.
  • ✓ The approach may be extended to other areas of healthcare, such as diagnosis and treatment planning, to further improve the accuracy and reliability of clinical predictions.

Sources

Original: arXiv - cs.AI

Related Articles