ChronoMedKG: A Temporally-Grounded Biomedical Knowledge Graph and Benchmark for Clinical Reasoning

2026-05-21 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Biomedical Informatics · Depth: Expert, quick

Summary

ChronoMedKG is a novel temporal biomedical knowledge graph designed to address the static nature of existing KGs like PrimeKG and Hetionet, which lack crucial temporal information for clinical reasoning. It comprises 460,497 evidence-linked triples covering 13,431 diseases, with each association tied to temporal components such as onset window or progression stage, backed by PMID-traceable evidence and a multi-signal credibility score. The graph is constructed via a disease-autonomous multi-agent pipeline where frontier LLMs extract knowledge from PubMed and PMC literature, retaining only relations supported by multi-model consensus, credibility filtering, and ontology alignment. ChronoMedKG achieved 92.7% agreement with Orphadata and adds temporal grounding for 6,250 diseases, including 1,657 Orphanet-coded rare diseases. A new benchmark, ChronoTQA, with 3,341 questions, reveals frontier LLMs lose 30 points on temporal tasks, while ChronoMedKG retrieval rescues 47-65% of their long-tail failures, significantly outperforming HPOA-RAG's 17-29%.

Key takeaway

For AI Scientists and NLP Engineers developing clinical reasoning systems, existing static knowledge graphs are insufficient for longitudinal patient data. You should consider integrating ChronoMedKG to provide the crucial temporal axis needed for accurate disease progression and symptom relevance. Furthermore, leverage the ChronoTQA benchmark to rigorously evaluate your models' temporal reasoning capabilities, as it demonstrates significant improvements over traditional RAG approaches.

Key insights

Temporal information is critical for accurate clinical reasoning, a gap ChronoMedKG addresses with its temporally-grounded biomedical knowledge graph.

Principles

• Temporal context is essential for clinical reasoning.
• Multi-model consensus improves knowledge graph accuracy.

Method

ChronoMedKG is built using a disease-autonomous multi-agent pipeline where frontier LLMs extract knowledge from PubMed/PMC, followed by multi-model consensus, credibility filtering, and ontology alignment.

In practice

• Integrate ChronoMedKG into retrieval-augmented clinical systems.
• Utilize ChronoTQA to benchmark temporal clinical reasoning capabilities.

Topics

• Biomedical Knowledge Graphs
• Clinical Reasoning
• Temporal AI
• Retrieval Augmentation
• Large Language Models
• ChronoTQA

Best for: AI Scientist, Research Scientist, NLP Engineer

Related on AIssential

• ClinicalMC: A Benchmark for Multi-Course Clinical Decision-Making with Large Language Models — Evaluating LLMs in evolving multi-course clinical scenarios is crucial for real-world healthcare deployment.
• Chronological Blindness: Benchmarking Temporal Reasoning in Vision-Language Models with CHRONOSIGHT — Vision-language models exhibit "chronological blindness," struggling with temporal reasoning compared to human perception.
• COTCAgent: Preventive Consultation via Probabilistic Chain-of-Thought Completion — COTCAgent enhances longitudinal EHR reasoning by decoupling statistical computation and leveraging a probabilistic chain-of-thought.
• DYNA : Dynamic Episodic Memory Networks for Augmenting Large Language Models with Temporal Knowledge Graphs in Continuous Learning — DYNA augments frozen LLMs with temporal knowledge graphs to mitigate forgetting and improve temporal recall without retraining.
• Text Knows What, Tables Know When: Clinical Timeline Reconstruction via Retrieval-Augmented Multimodal Alignment — Multimodal alignment of clinical text and EHR data significantly improves temporal precision in patient timeline reconstruction.
• SPOKE: A massive biomedical knowledge graph for precision health and drug discovery — SPOKE integrates vast biomedical data into a knowledge graph for drug discovery and precision medicine.

Open in AIssential →

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