{"id":"W4321748937","doi":"10.1093/jamia/ocad009","title":"Reproducible variability: assessing investigator discordance across 9 research teams attempting to reproduce the same observational study","year":2023,"lang":"en","type":"article","venue":"Journal of the American Medical Informatics Association","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"Alnylam Pharmaceuticals; AstraZeneca","keywords":"Observational study; Interquartile range; Cohort; Cohort study; Computer science; Inclusion (mineral); Variable (mathematics); Baseline (sea); Medicine; Implementation; Psychology; Statistics; Mathematics; Surgery; Pathology","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":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4728644,0.001101738,0.001567295,0.004166225,0.00206053,0.004428727,0.003399789,0.001889906,0.001622778],"category_scores_gemma":[0.619253,0.001158971,0.002785829,0.003354153,0.004262228,0.002319503,0.008166512,0.001747356,0.0004954885],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0030779,"about_ca_system_score_gemma":0.005142828,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001237313,"about_ca_topic_score_gemma":0.001387437,"domain_scores_codex":[0.3829786,0.4262053,0.1038271,0.03330972,0.0509345,0.002744735],"domain_scores_gemma":[0.160266,0.575684,0.1040487,0.1070811,0.05037116,0.002549009],"domain_codex":"methods","domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.005963782,0.0008649871,0.7348984,0.004162969,0.004268738,0.00102144,0.05761364,0.006788154,0.009930574,0.007454625,0.00745395,0.1595786],"study_design_scores_gemma":[0.001990984,0.006037573,0.7836933,0.005466324,0.004402686,0.003131239,0.02235694,0.05397513,0.03840275,0.02952416,0.05005944,0.0009595862],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6636138,0.002244016,0.3071808,0.001405128,0.0009970608,0.01271816,0.002016802,0.00157094,0.008253207],"genre_scores_gemma":[0.8602802,0.0001736354,0.1299029,0.000605426,0.0001972238,0.006643464,0.001206152,0.0004649816,0.0005261581],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5271356,"threshold_uncertainty_score":0.650052,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7985450138999625,"score_gpt":0.6273149303618455,"score_spread":0.171230083538117,"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."}}