ClinLens: Towards Long-Horizon Coding Agents for Longitudinal Multimodal Clinical Data Science
arXiv:2607.26155
ClinLens introduces a benchmark of 200 executable tasks that link multiple MIMIC data modalities—structured records, notes, ECGs, chest X‑rays, and echocardiograms—to evaluate long‑horizon coding agents that generate auditable clinical analyses over patient timelines.
Abstract
Clinical data-science agents must transform heterogeneous longitudinal records into auditable analyses, yet existing benchmarks largely isolate medical question answering, structured-table reasoning, or generic scientific repositories. We introduce CLINLENS, a benchmark of 200 executable tasks over five linked MIMIC resources spanning structured electronic health records, notes, electrocardiograms, chest radiographs, and echocardiograms. A 4 x 5 taxonomy crosses four patient-time scopes with five analysis capabilities. Program-first reverse synthesis pairs each bounded semi-raw package with an evaluator-private reference workflow and checks required artifacts, cohort and temporal semantics, and the final answer. On a fixed 126-task suite, the strongest of 24 standardized model-scaffold configurations achieves 56.3% scope-macro STRICTPASS despite 100% EXECSUCCESS. For reference, a separately configured coding agent solves 83 of 126 tasks, while five biomedical systems adapted to GPT-4o-mini reach at most 2.9% scope-macro STRICTPASS. These results expose a substantial gap between runnable submissions and correct clinical analyses.