{"id":"W4412822443","doi":"10.1001/jamanetworkopen.2025.30220","title":"Reporting Guideline for Chatbot Health Advice Studies","year":2025,"lang":"en","type":"article","venue":"JAMA Network Open","topic":"Digital Mental Health Interventions","field":"Psychology","cited_by":21,"is_retracted":false,"has_abstract":true,"ca_institutions":"Ottawa Hospital; University of British Columbia; McMaster University; Women's College Hospital; Hamilton Health Sciences; University of Toronto","funders":"","keywords":"Checklist; Computer science; Guideline; Interpretability; Data science; Medicine; Psychology; Artificial intelligence","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":["metaresearch"],"domain":"reporting","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"reporting","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005718023,0.0001492428,0.0005065002,0.00005784569,0.0003537183,0.0001411081,0.0004967448,0.00007283048,0.0001326543],"category_scores_gemma":[0.0009028987,0.0001427911,0.0001281205,0.000373697,0.00003981078,0.0002185236,0.0005331328,0.0001488347,0.00009352082],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001940563,"about_ca_system_score_gemma":0.0001358399,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005386546,"about_ca_topic_score_gemma":0.0005811686,"domain_scores_codex":[0.995661,0.0001548104,0.003052312,0.0004437979,0.00009344823,0.0005946141],"domain_scores_gemma":[0.9969762,0.0003375631,0.001889585,0.000489524,0.0002050332,0.0001020726],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000168586,0.0001331923,0.001375225,0.0001356457,0.0001901293,0.000005259395,0.0002307689,0.00005059571,9.494821e-8,0.04132507,0.8004593,0.1559261],"study_design_scores_gemma":[0.001379708,0.0004716124,0.006163296,0.001522099,0.00002421286,0.00001173729,0.001568207,0.0001707152,0.000004665858,0.03583058,0.9526976,0.000155591],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.009887843,0.03348393,0.01100173,0.3767401,0.02081444,0.009488101,0.0001511644,0.0005282475,0.5379044],"genre_scores_gemma":[0.4783539,0.000128555,0.07761028,0.1280185,0.002826228,0.004901479,0.0004024747,0.0001204031,0.3076381],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.4684661,"threshold_uncertainty_score":0.582285,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1735591945722361,"score_gpt":0.5589418649843204,"score_spread":0.3853826704120843,"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."}}