{"id":"W4385694112","doi":"10.1159/000531054","title":"Clinical Validation of Novel Digital Measures: Statistical Methods for Reliability Evaluation","year":2023,"lang":"en","type":"review","venue":"Digital Biomarkers","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"Bayer (Canada)","funders":"","keywords":"Reliability (semiconductor); Categorical variable; Computer science; Reliability engineering; Measure (data warehouse); Data mining; Machine learning; Engineering","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","scholarly_communication"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.06167578,0.0005668849,0.00262006,0.0004745824,0.0001233904,0.00130571,0.001379832,0.0005596306,0.0000828314],"category_scores_gemma":[0.1803502,0.0003900129,0.001954429,0.001535236,0.0005897849,0.001093106,0.0003196466,0.0002822809,0.0002026309],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003257395,"about_ca_system_score_gemma":0.001090089,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000003927189,"about_ca_topic_score_gemma":9.766052e-7,"domain_scores_codex":[0.9850199,0.001829355,0.006162271,0.00193215,0.004531032,0.000525333],"domain_scores_gemma":[0.9589745,0.03463363,0.002061641,0.001654561,0.002396875,0.0002788373],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00006865538,0.0004398741,0.0001490302,0.001815288,0.000365222,2.796203e-7,0.00001589115,0.0000109416,5.633703e-7,0.0001511631,0.003194521,0.9937886],"study_design_scores_gemma":[0.0007828067,0.0003158268,0.00009940819,0.00204107,0.0008267511,0.000002285778,0.0001436173,0.001449234,0.000004172221,0.0202075,0.9736063,0.0005209899],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"review","genre_scores_codex":[0.0000733605,0.4048096,0.5729237,0.0001025374,0.003019441,0.008316054,0.006974623,0.0001395519,0.003641152],"genre_scores_gemma":[0.000834067,0.9095527,0.07903821,0.00004014415,0.000645147,0.001697532,0.00708194,0.0002505633,0.0008597254],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.9932676,"threshold_uncertainty_score":0.9998552,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6501770228660378,"score_gpt":0.6044108355425473,"score_spread":0.04576618732349058,"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."}}