{"paper":{"title":"An Empirical Analysis of Calibration and Selective Prediction in Multimodal Clinical Condition Classification","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Selective prediction degrades performance in multimodal clinical condition classification due to class-dependent miscalibration.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Farah E. Shamout, L. Juli\\'an Lechuga L\\'opez, Tim G. J. Rudner","submitted_at":"2026-03-03T08:16:44Z","abstract_excerpt":"As artificial intelligence systems move toward clinical deployment, ensuring reliable prediction behavior is fundamental for safety-critical decision-making tasks. One proposed safeguard is selective prediction, where models can defer uncertain predictions to human experts for review. In this work, we empirically evaluate the reliability of uncertainty-based selective prediction in multilabel clinical condition classification using multimodal ICU data. Across a range of state-of-the-art unimodal and multimodal models, we find that selective prediction can substantially degrade performance desp"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"selective prediction can substantially degrade performance despite strong standard evaluation metrics. This failure is driven by severe class-dependent miscalibration, whereby models assign high uncertainty to correct predictions and low uncertainty to incorrect ones, particularly for underrepresented clinical conditions.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that the tested unimodal and multimodal models and the specific ICU dataset splits are representative enough to generalize the miscalibration failure mode to broader clinical deployment scenarios.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Selective prediction degrades performance in multimodal clinical condition classification due to class-dependent miscalibration that assigns high uncertainty to correct predictions and low uncertainty to incorrect ones.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Selective prediction degrades performance in multimodal clinical condition classification due to class-dependent miscalibration.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"facc65be53f068b0b2c03ac239f1c93aa09a1c003ee33e31d115b666ef5950e7"},"source":{"id":"2603.02719","kind":"arxiv","version":4},"verdict":{"id":"4de4ad46-b29d-43af-aa5a-77ee3c29c9b7","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T16:37:55.521039Z","strongest_claim":"selective prediction can substantially degrade performance despite strong standard evaluation metrics. This failure is driven by severe class-dependent miscalibration, whereby models assign high uncertainty to correct predictions and low uncertainty to incorrect ones, particularly for underrepresented clinical conditions.","one_line_summary":"Selective prediction degrades performance in multimodal clinical condition classification due to class-dependent miscalibration that assigns high uncertainty to correct predictions and low uncertainty to incorrect ones.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that the tested unimodal and multimodal models and the specific ICU dataset splits are representative enough to generalize the miscalibration failure mode to broader clinical deployment scenarios.","pith_extraction_headline":"Selective prediction degrades performance in multimodal clinical condition classification due to class-dependent miscalibration."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2603.02719/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"827e4d79f4d02844342a709227af034737b21c21aa3e60147cb640f5a4e39f26"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}