{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.23301", "kind": "paper", "url": "https://arxiv.org/abs/2606.23301"}], "authors": ["Yitong Qiao", "Lei Liu", "Yue Shen", "Jian Wang", "Jinjie Gu", "Zhixuan Chu", "Kui Ren"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.23301", "source_url": "https://arxiv.org/abs/2606.23301"}, "categories": ["Agents & Tool Use", "Biology & Drug Discovery", "Reasoning"], "description": {"en": "EHR-Complex evaluates clinical agent performance on interactive reasoning over MIMIC-IV electronic health records. It consists of about 52K tasks across six clinical intents, requiring agents to execute SQL or Python in a sandboxed environment to answer patient- and population-level queries. Scoring is based on exact-match accuracy against expected outcomes."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.23301", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.23301", "title": "EHR-Complex"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.23301", "modality": null, "name": "EHR-Complex", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "High", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "unknown", "origin_record_id": "bm_ehr-complex_08f2ee7c", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "2b7fcf3cb2f5fcd4bd12458564bf938f5c4a27cd9c2b8f799e178a5e89bcc7ae", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.23301"}, "publisher": null, "released": "2026-06-22", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.23301", "source_url": "https://arxiv.org/abs/2606.23301"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-23301", "source": "claire_radar", "source_benchmark_id": "2606.23301", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["Biology & Drug Discovery"], "area": "Agents & Tool Use", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": null, "githubScope": null, "githubStars": null, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": null, "hfPaperUrl": "https://huggingface.co/papers/2606.23301", "observedAt": "2026-10-04T03:02:42.754940Z", "signalStatus": {"githubStars": {"state": "not_applicable"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "state": "unavailable"}}, "status": "unavailable"}, "capabilities": ["Reasoning"], "confidence": "High", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "6a7a4bcc87590e1d877d8fb6b3ed8b222b7c1bf3f1e614ed94e9d017e0713e3c", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "EHR-Complex evaluates clinical agent performance on interactive reasoning over MIMIC-IV electronic health records. It consists of about 52K tasks across six clinical intents, requiring agents to execute SQL or Python in a sandboxed environment to answer patient- and population-level queries. Scoring is based on exact-match accuracy against expected outcomes.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence", "evaluation protocol evidence"], "snippet": "In this work, we introduce EHR-Complex, a large-scale benchmark designed for interactive clinical database reasoning."}, "familyId": "bmf_68a1189baa1d", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_ehr-complex_08f2ee7c", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Pharma & Biotech"], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2606.23301", "paper": "https://arxiv.org/abs/2606.23301", "pdf": "https://arxiv.org/pdf/2606.23301", "project": null, "report": "https://arxiv.org/abs/2606.23301"}, "metrics": [], "motivation": "Clinical agents promise to democratize access to electronic health records (EHRs), yet existing benchmarks fail to reflect the complexity of practical EHR analysis, e.g., often operating on idealized, clean EHRs via static SQL generation rather than interactive execution.", "name": "EHR-Complex", "oneLine": "EHR-Complex evaluates clinical agent performance on interactive reasoning over MIMIC-IV electronic health records. It consists of about 52K tasks across six clinical intents, requiring agents to execute SQL or Python in a sandboxed environment to answer patient- and population-level queries. Scoring is based on exact-match accuracy against expected outcomes.", "paperTitle": "EHR-Complex: Benchmarking Medical Agents for Complex Clinical Reasoning", "primaryDomain": "Biology & Drug Discovery", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.23301", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-22", "source": {"authors": ["Yitong Qiao", "Lei Liu", "Yue Shen", "Jian Wang", "Jinjie Gu", "Zhixuan Chu", "Kui Ren"], "categories": ["cs.AI"], "id": "2606.23301", "title": "EHR-Complex: Benchmarking Medical Agents for Complex Clinical Reasoning", "type": "arxiv", "url": "https://arxiv.org/abs/2606.23301"}, "sourceUpdatedAt": "2026-06-22T13:14:21Z", "topics": ["Reasoning"], "venueAttempts": [], "whyItMatters": "Existing clinical benchmarks often rely on simplified, static SQL generation, failing to reflect real-world EHR complexity. EHR-Complex introduces interactive, multi-step reasoning tasks with compositional queries, revealing that state-of-the-art models achieve only 62.3% accuracy and exhibit fragility under repeated sampling, highlighting significant room for improvement in robust clinical reasoning."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
