Agentic discovery of blood biomarker from distilled private health records
The paper releases a scorer meant to let AI agents test CBC biomarker ideas without exposing patient records.
The scorer was trained inside Clalit Health Services’ data boundary on records covering more than 5.4 million patients. It predicts case-control AUC for candidate CBC expressions across 13 immune-mediated diseases, so an agent can propose, score, and refine hypotheses against distilled evidence. In external validation, the agent-discovered expressions beat their literature-seeded starting points by a median 4.18 AUC percentage points. The authors also report that reranking candidates from three frontier research tools improved on their first choices in most comparisons across three independent cohorts. HF Daily Papers' note
The scorer was trained inside Clalit Health Services’ data boundary on records covering more than 5.4 million patients. It predicts case-control AUC for candidate CBC expressions across 13 immune-mediated diseases, so an agent can propose, score, and refine hypotheses against distilled evidence. In external validation, the agent-discovered expressions beat their literature-seeded starting points by a median 4.18 AUC percentage points. The authors also report that reranking candidates from three frontier research tools improved on their first choices in most comparisons across three independent cohorts. HF Daily Papers' note
score 5