COTCAgent: Preventive Consultation via Probabilistic Chain-of-Thought Completion

2026-05-14 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Medical Devices & Health Technology · Depth: Advanced, quick

Summary

COTCAgent, a Probabilistic Chain-of-Thought Completion Agent, is introduced as a hierarchical reasoning framework designed to improve longitudinal electronic health record (EHR) analysis. Current large language models struggle with EHRs due to issues like hallucinating clinical trends from implied quantitative evidence and failing to capture long-range temporal dependencies from non-uniform time series and scarce labels. COTCAgent addresses these by using a Temporal-Statistics Adapter for standardized trend output, a Chain-of-Thought Completion layer with a symptom-trend-disease knowledge base for risk evaluation, and a bounded completion module for structured evidence acquisition. This framework decouples statistical computation, feature matching, and language generation, reducing reliance on complex multi-modal inputs and lowering computational overhead. Powered by Baichuan-M2, COTCAgent achieved 90.47% Top-1 accuracy on a self-built dataset and 70.41% on HealthBench, surpassing existing medical agents and mainstream LLMs.

Key takeaway

For AI Scientists and Machine Learning Engineers developing clinical decision support systems, COTCAgent offers a robust approach to overcome current LLM limitations in longitudinal EHR reasoning. You should consider adopting its hierarchical framework, particularly its decoupled statistical computation and probabilistic chain-of-thought completion, to improve diagnostic accuracy and reduce computational overhead in your medical AI applications.

Key insights

COTCAgent enhances longitudinal EHR reasoning by decoupling statistical computation and leveraging a probabilistic chain-of-thought.

Principles

• Decouple statistical computation from language generation.
• Utilize knowledge bases for weighted disease risk scoring.
• Employ iterative scoring constraints for rigorous reasoning.

Method

COTCAgent uses a Temporal-Statistics Adapter for trend output, a Chain-of-Thought Completion layer with a knowledge base for risk evaluation, and a bounded completion module for structured evidence acquisition via iterative scoring.

In practice

• Integrate a Temporal-Statistics Adapter for EHR trend analysis.
• Develop symptom-trend-disease knowledge bases for risk assessment.
• Implement bounded completion for structured evidence gathering.

Topics

• COTCAgent
• Longitudinal EHR Reasoning
• Probabilistic Chain-of-Thought
• Clinical Decision Support
• Temporal-Statistics Adapter

Code references

• FrankDengAI/COTCAgent

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Beyond Visual Cues: CoT-Enhanced Reasoning for Semi-supervised Medical Image Segmentation — Integrating Chain-of-Thought reasoning with LLM-generated linguistic knowledge improves semi-supervised medical image segmentation by resolving visual-semantic mismatches.
• Reasoning models struggle to control their chains of thought, and that’s good — Low CoT controllability in frontier models enhances the trustworthiness of AI safety monitoring systems.
• AI-assisted Protocol Information Extraction For Improved Accuracy and Efficiency in Clinical Trial Workflows — AI-assisted RAG significantly boosts clinical trial protocol data extraction accuracy and efficiency over standalone LLMs.
• NTS-CoT: Mitigating Hallucinations in LLM-based News Timeline Summarization with Chain-of-Thought Reasoning — Chain-of-Thought reasoning effectively mitigates hallucinations in LLM-based news timeline summarization by structuring information extraction and causal inference.
• ChronoMedKG: A Temporally-Grounded Biomedical Knowledge Graph and Benchmark for Clinical Reasoning — Temporal information is critical for accurate clinical reasoning, a gap ChronoMedKG addresses with its temporally-grounded biomedical knowledge graph.
• Free Energy Heuristics: Fast-And-Frugal Cognition as Active Inference Under Uncertain Precision — High meta-uncertainty causes CoT to degrade LLM accuracy by manufacturing false confidence.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.