{"id":"W1514773312","doi":"10.1111/pme.12588","title":"Assessment of Pain Intensity in Clinical Trials: Individual Ratings vs Composite Scores","year":2014,"lang":"en","type":"article","venue":"Pain Medicine","topic":"Musculoskeletal pain and rehabilitation","field":"Medicine","cited_by":34,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University; Lawson Health Research Institute","funders":"","keywords":"Generalizability theory; Medicine; Physical therapy; Clinical trial; Reliability (semiconductor); Physical medicine and rehabilitation; Psychology; Internal medicine","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"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1454779,0.0001864668,0.001545554,0.0002705269,0.00003641312,0.000005003997,0.0001224287,0.0001718903,0.0001362882],"category_scores_gemma":[0.05824644,0.0001218509,0.0002255604,0.0003118486,0.0004964618,0.00003688979,0.0000415396,0.0005175125,0.000003320512],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004841445,"about_ca_system_score_gemma":0.00008375443,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002105714,"about_ca_topic_score_gemma":0.00003994192,"domain_scores_codex":[0.9881014,0.008524102,0.00209899,0.0003616637,0.0006714115,0.0002424184],"domain_scores_gemma":[0.9839199,0.01466234,0.0005491556,0.0004041445,0.0002410422,0.0002234253],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00029165,0.0003195163,0.9145071,0.001251218,0.00008350771,0.000003211936,0.0007254003,0.000006691626,0.007008275,0.0004069237,0.005227777,0.07016876],"study_design_scores_gemma":[0.004793982,0.003617273,0.9678909,0.002314708,0.0001041439,0.000001483779,0.0006785435,0.01910964,0.00004930146,0.0004823501,0.0008463952,0.0001112326],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9587675,0.0002116694,0.02498201,0.01346677,0.0003288966,0.001018397,0.000001601352,0.00005065547,0.001172442],"genre_scores_gemma":[0.9864846,0.00005158904,0.005431254,0.007189354,0.0006574743,0.00002270215,0.00009384615,0.00001667602,0.00005243348],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08723146,"threshold_uncertainty_score":0.9496863,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07039708009525254,"score_gpt":0.4333057115680937,"score_spread":0.3629086314728411,"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."}}