{"id":"W2022144328","doi":"10.1007/s40264-015-0265-0","title":"Pediatric Drug Safety Signal Detection: A New Drug–Event Reference Set for Performance Testing of Data-Mining Methods and Systems","year":2015,"lang":"en","type":"article","venue":"Drug Safety","topic":"Pharmaceutical studies and practices","field":"Medicine","cited_by":31,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Manitoba; Université de Montréal","funders":"ZonMw; Pfizer; Eli Lilly and Company","keywords":"Medicine; Adverse Event Reporting System; Drug; Data mining; Event (particle physics); Data set; Pharmacology; Artificial intelligence; Computer science","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":[],"category_scores_codex":[0.02514312,0.001754091,0.001665967,0.01152512,0.0008801667,0.003063075,0.002939356,0.001685486,0.003525011],"category_scores_gemma":[0.07393967,0.0005803364,0.002325436,0.006110983,0.0005457085,0.002954622,0.002811623,0.001676558,0.00148681],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001524181,"about_ca_system_score_gemma":0.003623353,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0042927,"about_ca_topic_score_gemma":0.003776571,"domain_scores_codex":[0.9805446,0.005979278,0.00432329,0.002868143,0.005899867,0.0003848445],"domain_scores_gemma":[0.9355984,0.03378963,0.006068455,0.009915546,0.01389275,0.000735233],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.002287216,0.002131052,0.1586201,0.003754846,0.00216306,0.001386692,0.0006926793,0.0728164,0.01598935,0.008068065,0.04426238,0.6878282],"study_design_scores_gemma":[0.0009540209,0.004468635,0.1022749,0.001393869,0.001692651,0.002809226,0.0009570084,0.6983756,0.04556082,0.01431719,0.1268062,0.000389907],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1900999,0.003743134,0.687422,0.001348752,0.0004232866,0.003984146,0.0831164,0.02226208,0.007600335],"genre_scores_gemma":[0.2280494,0.0008685524,0.6419846,0.0003328758,0.0001019008,0.004127861,0.1229351,0.000498076,0.001101721],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9748569,"threshold_uncertainty_score":0.1329711,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2930657232303961,"score_gpt":0.4570619595480828,"score_spread":0.1639962363176867,"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."}}