{"id":"W7104177710","doi":"10.2139/ssrn.5709030","title":"Evaluation of tools used to assess adherence to PRISMA 2020 reveals inconsistent methods and poor tool implementability: part I of a systematic review","year":2025,"lang":"","type":"preprint","venue":"SSRN Electronic Journal","topic":"Medication Adherence and Compliance","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Children’s Health Research Institute","funders":"","keywords":"PsycINFO; Systematic review; MEDLINE; Psychological intervention; Psychometrics; Reliability (semiconductor); Self-report study; Comparability","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":[0.5818092,0.004514333,0.01580161,0.01880037,0.00334247,0.009370131,0.008319617,0.005549064,0.002984154],"category_scores_gemma":[0.7632388,0.00577216,0.02718324,0.02151686,0.00648992,0.008773732,0.007754792,0.006025278,0.0006650774],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01152305,"about_ca_system_score_gemma":0.05319396,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01264582,"about_ca_topic_score_gemma":0.02706229,"domain_scores_codex":[0.242523,0.4477455,0.241146,0.01238379,0.05361909,0.002582671],"domain_scores_gemma":[0.175675,0.6136014,0.1004882,0.04737169,0.060788,0.002075658],"domain_codex":"methods","domain_gemma":"reporting","domain_candidate":"reporting","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.003373848,0.0002503242,0.01805403,0.7392383,0.11172,0.0001678555,0.0073193,0.0007870504,0.0007157904,0.002009855,0.006639333,0.1097244],"study_design_scores_gemma":[0.008475102,0.003613306,0.03987234,0.6456779,0.2594424,0.0004418764,0.004784345,0.002806904,0.003387732,0.004492526,0.02619707,0.0008085283],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.08479297,0.6196461,0.09340355,0.01409239,0.005887148,0.1570059,0.01719909,0.001719402,0.006253409],"genre_scores_gemma":[0.4335809,0.1075278,0.2604218,0.004937081,0.0004976007,0.1852386,0.006235597,0.0007152505,0.0008454022],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.4181908,"threshold_uncertainty_score":0.5157037,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2475618662032558,"score_gpt":0.4841161739677188,"score_spread":0.236554307764463,"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."}}