{"id":"W4416185868","doi":"10.1038/s44277-025-00042-z","title":"Using large language models as a scalable mental status evaluation technique","year":2025,"lang":"en","type":"article","venue":"NPP—Digital Psychiatry and Neuroscience","topic":"Mental Health via Writing","field":"Psychology","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"","keywords":"Mental health; Scalability; Language model; Identification (biology); Anxiety; Natural language; Random forest; Computational linguistics; Session (web analytics)","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":[],"consensus_categories":[],"category_scores_codex":[0.002558644,0.00163809,0.0007452811,0.001202082,0.0004401223,0.001634678,0.001354669,0.001096556,0.003007032],"category_scores_gemma":[0.01136171,0.0005076196,0.00137357,0.0005854052,0.0003814688,0.00187876,0.001229553,0.002471565,0.002350937],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001070391,"about_ca_system_score_gemma":0.00101018,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007177877,"about_ca_topic_score_gemma":0.01102931,"domain_scores_codex":[0.9987119,0.0006140911,0.0001068158,0.0003466404,0.0001446574,0.00007583703],"domain_scores_gemma":[0.9950926,0.00367108,0.0002472625,0.0003994287,0.0004815311,0.0001081143],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001143334,0.0006915602,0.01284614,0.0003238882,0.0004838345,0.0008290519,0.0004860405,0.4571197,0.01509359,0.005947927,0.01799565,0.4870393],"study_design_scores_gemma":[0.00002615431,0.00005630563,0.0004629908,0.00001042069,0.00002788443,0.00004709491,0.00004046252,0.9918611,0.001638734,0.005002798,0.0008120223,0.00001402956],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1239371,0.001177623,0.848729,0.002081991,0.0002904274,0.0004453769,0.003931654,0.01581327,0.00359361],"genre_scores_gemma":[0.7706859,0.0002838779,0.2173409,0.0006974746,0.0001679181,0.0004470273,0.006671857,0.0004636912,0.00324134],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.007177877,"threshold_uncertainty_score":0.01427221,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05727910150430604,"score_gpt":0.4208321513042883,"score_spread":0.3635530497999823,"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."}}