{"id":"W4221077687","doi":"10.1038/s41598-022-06484-1","title":"Evaluation of domain generalization and adaptation on improving model robustness to temporal dataset shift in clinical medicine","year":2022,"lang":"en","type":"article","venue":"Scientific Reports","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":89,"is_retracted":false,"has_abstract":true,"ca_institutions":"Institute for Clinical Evaluative Sciences; SickKids Foundation; Hospital for Sick Children","funders":"","keywords":"Computer science; Receiver operating characteristic; Artificial intelligence; Robustness (evolution); False positive paradox; Machine learning; Predictive modelling; Data mining","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02128721,0.001745281,0.001161373,0.001263198,0.0005649434,0.001465351,0.001629846,0.001827881,0.0009764169],"category_scores_gemma":[0.04295465,0.0004164647,0.001653513,0.0007663351,0.001020183,0.002251792,0.001986381,0.00247741,0.0004671172],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001202493,"about_ca_system_score_gemma":0.001809778,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008256907,"about_ca_topic_score_gemma":0.004989468,"domain_scores_codex":[0.9942307,0.003235403,0.0005000018,0.001309707,0.0004615916,0.000262487],"domain_scores_gemma":[0.9762981,0.01593149,0.001558721,0.003443716,0.002218993,0.000549037],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00188776,0.000892434,0.06109688,0.0003249065,0.0009918445,0.0001719461,0.0002558805,0.7787338,0.003658777,0.0009603989,0.00352258,0.1475028],"study_design_scores_gemma":[0.00009399393,0.0009250753,0.01134915,0.00007704959,0.0001391409,0.0001012521,0.0001345129,0.9812893,0.00321767,0.001634687,0.0009958638,0.00004232895],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8682526,0.005029116,0.1151849,0.002760735,0.0004862811,0.0004877871,0.001398678,0.002941236,0.003458644],"genre_scores_gemma":[0.9597266,0.0004086858,0.03626038,0.0005337413,0.0001152558,0.0001228889,0.002088441,0.00009094743,0.0006531363],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02128721,"threshold_uncertainty_score":0.1125789,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1090369519650329,"score_gpt":0.3930394425137202,"score_spread":0.2840024905486873,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}