{"id":"W2970473468","doi":"10.1111/bdi.12829","title":"An examination of the quality and performance of the Alda scale for classifying lithium response phenotypes","year":2019,"lang":"en","type":"article","venue":"Bipolar Disorders","topic":"Bipolar Disorder and Treatment","field":"Medicine","cited_by":54,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University","funders":"","keywords":"Interpretability; Scale (ratio); Confounding; Reliability (semiconductor); Biomarker; Computer science; Psychology; Clinical psychology; Machine learning; Statistics; Medicine; Mathematics; Genetics; Biology","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":[0.0005814339,0.00009880476,0.0001802082,0.00003700879,0.00009560447,0.000005492336,0.0001156852,0.00006738315,0.00000993738],"category_scores_gemma":[0.00006839342,0.00005644616,0.00009370295,0.0001492808,0.0001438502,0.0000805273,0.00003768552,0.00008163187,0.000001051837],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001994846,"about_ca_system_score_gemma":0.00006628365,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008505921,"about_ca_topic_score_gemma":0.0002131031,"domain_scores_codex":[0.9991009,0.0001954052,0.0002123206,0.0001632831,0.0002041783,0.0001239337],"domain_scores_gemma":[0.9991979,0.0001282667,0.0001382722,0.0004389627,0.00006738216,0.00002920293],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006403215,0.0002125725,0.8958538,0.0002775088,0.00003388761,4.026242e-9,0.001141121,0.000004211259,0.01828949,0.0001497266,0.000004341572,0.08339303],"study_design_scores_gemma":[0.001302378,0.0004617615,0.9830016,0.00006553555,0.00007935487,0.000001115196,0.0005324587,0.0004902366,0.008919373,0.00006237671,0.005019734,0.00006405306],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9929818,0.005163811,0.00003596634,0.0006984332,0.00009435494,0.0008066017,0.00002655753,0.0000107185,0.0001817632],"genre_scores_gemma":[0.9990466,0.0004592418,0.0001658308,0.00005395502,0.000009593896,0.00001814533,0.000006082176,0.00001373036,0.0002268362],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08714783,"threshold_uncertainty_score":0.2301807,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01966661821205363,"score_gpt":0.2878961674425157,"score_spread":0.2682295492304621,"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."}}