{"id":"W4306163964","doi":"10.1186/s12874-022-01739-x","title":"The failure of four bootstrap procedures for estimating confidence intervals for predicted-to-expected ratios for hospital profiling","year":2022,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Sepsis Diagnosis and Treatment","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"Sunnybrook Hospital; University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; Heart and Stroke Foundation of Canada","keywords":"Confidence interval; Statistics; Percentile; Medicine; Logistic regression; Intraclass correlation; Econometrics; Mathematics","routes":{"ca_aff":true,"ca_fund":true,"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"],"consensus_categories":[],"category_scores_codex":[0.01806046,0.0001737549,0.0006726523,0.0002153652,0.0006712224,0.00002599522,0.0004903884,0.0001509581,0.0001267886],"category_scores_gemma":[0.3087704,0.000116003,0.0002920959,0.0003459235,0.0003608989,0.00003406836,0.0002572484,0.0004358992,8.677965e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001503861,"about_ca_system_score_gemma":0.002451475,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003985392,"about_ca_topic_score_gemma":0.0001777235,"domain_scores_codex":[0.9945744,0.001984745,0.0007686669,0.000567554,0.001301151,0.000803453],"domain_scores_gemma":[0.8993431,0.09839277,0.0002086659,0.0004019452,0.001235749,0.0004177651],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.04358529,0.007418239,0.0725856,0.02219269,0.004992372,0.0001322048,0.0165784,0.001262402,0.04888074,0.1545167,0.3607315,0.2671239],"study_design_scores_gemma":[0.05213193,0.1287569,0.01970764,0.00407118,0.001592213,0.0005472574,0.05781938,0.3542251,0.2313072,0.1028337,0.0454939,0.001513634],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1338693,0.0005904297,0.8284973,0.02409657,0.0005140335,0.01215159,0.0001988196,0.00005731772,0.00002462589],"genre_scores_gemma":[0.2744817,0.00002088364,0.6940477,0.0003257265,0.0004640575,0.03029319,0.0001065984,0.00004298794,0.0002171091],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.3529627,"threshold_uncertainty_score":0.6970522,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6109030780082313,"score_gpt":0.5756935993115068,"score_spread":0.03520947869672453,"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."}}