{"id":"W4213065772","doi":"10.1115/1.4053888","title":"Use of Real-World Data for Enhancing Model Credibility: Applications to Medical Device Development","year":2022,"lang":"en","type":"article","venue":"Journal of Medical Devices","topic":"Total Knee Arthroplasty Outcomes","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Spinologics (Canada)","funders":"","keywords":"Credibility; Computer science; Relevance (law); Process (computing); Data science; Data modeling; Data mining; Risk analysis (engineering); Reliability engineering; Engineering; Medicine; Software 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":[],"consensus_categories":[],"category_scores_codex":[0.004336606,0.000161541,0.0007250396,0.0003432725,0.0001638577,0.00001910395,0.001000478,0.00007433782,0.0006554669],"category_scores_gemma":[0.004468234,0.0001288308,0.0001278968,0.000528764,0.0001185898,0.0002335611,0.0009472477,0.0006016882,0.000006582814],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001943164,"about_ca_system_score_gemma":0.003390688,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007671244,"about_ca_topic_score_gemma":0.003561266,"domain_scores_codex":[0.9942236,0.00009693405,0.0013723,0.0003213506,0.003680653,0.0003051524],"domain_scores_gemma":[0.9959099,0.001346937,0.0006161932,0.0006094477,0.0003458351,0.00117164],"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":[0.004183269,0.006357417,0.1781328,0.004514286,0.00282782,0.0006361952,0.005021559,0.004121894,0.002803589,0.004602848,0.1117298,0.6750685],"study_design_scores_gemma":[0.003573541,0.0008756472,0.03046328,0.001048856,0.0004611736,0.0005939773,0.001396892,0.06287716,0.001048407,0.0001293465,0.8971595,0.0003722623],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7970722,0.0005725549,0.1301405,0.06981516,0.0005688627,0.001405756,0.0001669654,0.00005575037,0.0002022997],"genre_scores_gemma":[0.8998874,0.0001173414,0.09176504,0.0070645,0.000497447,0.00009800492,0.0000781796,0.00004193542,0.0004501622],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7854297,"threshold_uncertainty_score":0.7176903,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1432862036920433,"score_gpt":0.4003832022937769,"score_spread":0.2570969986017336,"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."}}