{"id":"W2105852486","doi":"10.1186/1472-6963-8-277","title":"Reliability of the interRAI suite of assessment instruments: a 12-country study of an integrated health information system","year":2008,"lang":"en","type":"article","venue":"BMC Health Services Research","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":677,"is_retracted":false,"has_abstract":true,"ca_institutions":"Homewood Research Institute; University of Waterloo","funders":"University of Waterloo; Curtin University of Technology","keywords":"Health informatics; Reliability (semiconductor); Medicine; Nursing research; Cohen's kappa; Kappa; Suite; Health care; Health administration; Nursing; Family medicine; Public health; Statistics","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":[0.0445347,0.0001830821,0.0007992351,0.0007941765,0.00054222,0.00006614435,0.002208464,0.00009120582,0.00005470052],"category_scores_gemma":[0.0006047749,0.000106522,0.0001080067,0.003513541,0.0003095359,0.0009829114,0.0005730197,0.0005437569,0.000008744917],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007132391,"about_ca_system_score_gemma":0.002786337,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02217947,"about_ca_topic_score_gemma":0.00641921,"domain_scores_codex":[0.9793022,0.007465601,0.003884483,0.0005111161,0.008233674,0.000602919],"domain_scores_gemma":[0.9918459,0.0006854472,0.001824899,0.002084461,0.003295711,0.0002635986],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000383511,0.002080101,0.9487247,0.00464759,0.00002404433,1.254046e-7,0.03534415,0.0009892086,0.00004543488,0.000124894,0.0001608749,0.007475389],"study_design_scores_gemma":[0.001036573,0.00260266,0.8290715,0.0006726564,0.00000268614,0.000001876834,0.1558597,0.009736136,0.0001032368,0.000104893,0.0007397592,0.00006833016],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9953597,0.00003887503,0.0002426449,0.0002146578,0.0002855801,0.003340838,0.00009046346,0.00001772861,0.0004094983],"genre_scores_gemma":[0.9988275,0.00005020455,0.0008285342,0.0001578911,0.00001830429,0.00007439555,0.00001545145,0.000006900932,0.00002083],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1205156,"threshold_uncertainty_score":0.9843319,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1940493002678006,"score_gpt":0.4783379459711168,"score_spread":0.2842886457033162,"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."}}