A biomedical knowledge graph that tracks disease symptoms over time

Existing biomedical knowledge graphs treat disease-symptom links as static facts, but clinical reality is temporal—a finding's relevance shifts across a patient's lifespan. ChronoMedKG adds time-stamped information to 460,497 disease associations (from 13M raw extractions), grounding each link to onset windows and disease stages with traceable evidence. When tested on temporal reasoning questions, frontier LLMs drop ~30 points from static baselines; retrieval from ChronoMedKG recovers 47–65% of failures, compared to 17–29% from existing sources. The graph covers 13,431 diseases including 1,657 rare diseases previously ungrounded, making it immediately useful for building better clinical decision-support systems.