{"paper":{"title":"RISED: A Pre-Deployment Safety Evaluation Framework for Clinical AI Decision-Support Systems","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Clinical AI models that pass standard accuracy tests can fail on input stability and threshold sensitivity.","cross_cats":["cs.AI","cs.CY","stat.AP"],"primary_cat":"cs.LG","authors_text":"Rohith Reddy Bellibatlu","submitted_at":"2026-05-13T02:17:13Z","abstract_excerpt":"Aggregate accuracy metrics dominate the evaluation of clinical AI decision-support systems but do not detect deployment-phase failures of input reliability, subgroup equity, threshold sensitivity, or operational feasibility. We propose the RISED Framework: a five-dimension pre-deployment evaluation covering Reliability, Inclusivity, Sensitivity, Equity, and Deployability, in which each dimension is operationalized through formal sub-criteria, pre-specified pass/fail thresholds, and bias-corrected accelerated (BCa) bootstrap 95% confidence intervals combined under a Holm-Bonferroni family-wise "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"A classifier satisfying conventional high-discrimination benchmarks can simultaneously fail input-encoding stability and threshold-shift sensitivity checks, while subgroup AUC parity remains statistically inconclusive, pointing to deployment risks that aggregate evaluation alone cannot detect.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the five chosen dimensions and their operationalized sub-criteria with pre-specified thresholds adequately capture the primary pre-deployment risks for clinical AI systems across diverse datasets and use cases.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"RISED is a structured pre-deployment safety framework that flags failures in clinical AI systems across reliability, inclusivity, sensitivity, equity, and deployability using pre-specified criteria and statistical corrections, even when aggregate accuracy looks strong.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Clinical AI models that pass standard accuracy tests can fail on input stability and threshold sensitivity.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"40024834475abee06200c61a1fba006570cb8387589939b4d1abaf9e5be2acf3"},"source":{"id":"2605.12895","kind":"arxiv","version":1},"verdict":{"id":"bef42c7b-3f83-4c07-9247-75ff3c92a55f","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:09:19.495359Z","strongest_claim":"A classifier satisfying conventional high-discrimination benchmarks can simultaneously fail input-encoding stability and threshold-shift sensitivity checks, while subgroup AUC parity remains statistically inconclusive, pointing to deployment risks that aggregate evaluation alone cannot detect.","one_line_summary":"RISED is a structured pre-deployment safety framework that flags failures in clinical AI systems across reliability, inclusivity, sensitivity, equity, and deployability using pre-specified criteria and statistical corrections, even when aggregate accuracy looks strong.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the five chosen dimensions and their operationalized sub-criteria with pre-specified thresholds adequately capture the primary pre-deployment risks for clinical AI systems across diverse datasets and use cases.","pith_extraction_headline":"Clinical AI models that pass standard accuracy tests can fail on input stability and threshold sensitivity."},"references":{"count":105,"sample":[{"doi":"","year":null,"title":"and Torkamani, Ali and Dias, Raquel and Gianfrancesco, Milena and Arnaout, Rima and Kohane, Isaac S","work_id":"f0895a9b-4adb-4c6c-a04f-339328cc7e53","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"and Porras, Antonio R","work_id":"0f1f71b2-07f9-4ecc-a9ba-bb60ba4e31b4","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"and Calvert, Melanie J","work_id":"5e8b5a88-2d28-4f3a-a5e5-23969f6014d0","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"and Denniston, Alastair K","work_id":"fd2a7a32-7c79-4c7e-a2a0-8186385a7928","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Journal of the American Statistical Association , year =","work_id":"acc72481-3709-44d8-9ed8-4ab9f814ce86","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":105,"snapshot_sha256":"54728d5b38602b16403c9cb6d4206edbbbf8e5d0892a8493f1c6e803c57ce0ee","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}