{"id":"W4376612861","doi":"10.2196/44549","title":"Evaluation of the EsteR Toolkit for COVID-19 Decision Support: Sensitivity Analysis and Usability Study","year":2023,"lang":"en","type":"article","venue":"JMIR Formative Research","topic":"Data Visualization and Analytics","field":"Computer Science","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Leibniz-Gemeinschaft; Bundesministerium für Bildung und Forschung","keywords":"Usability; Computer science; Test (biology); Stability (learning theory); Coronavirus disease 2019 (COVID-19); Decision support system; Machine learning; Artificial intelligence; Human–computer interaction; Medicine","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03296211,0.001827261,0.001523008,0.00201248,0.0005365257,0.002438275,0.002757879,0.001613781,0.00385332],"category_scores_gemma":[0.08737513,0.0008206379,0.001426625,0.0011657,0.0008004603,0.002506833,0.002766755,0.001567824,0.0006036705],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001182831,"about_ca_system_score_gemma":0.001117111,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002488947,"about_ca_topic_score_gemma":0.001953365,"domain_scores_codex":[0.9792354,0.01509448,0.001819017,0.001048497,0.002318478,0.0004841049],"domain_scores_gemma":[0.8105721,0.171972,0.002208529,0.006478657,0.00780433,0.0009643225],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.01889214,0.01541802,0.06447919,0.01850747,0.002498467,0.004060615,0.02151421,0.3540694,0.03741597,0.008172619,0.01248987,0.442482],"study_design_scores_gemma":[0.002187551,0.01566315,0.04604583,0.001880609,0.0009824778,0.0009383666,0.006055601,0.8723435,0.02723208,0.004605562,0.02138961,0.0006755637],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9369454,0.0003697895,0.05350521,0.000291468,0.00006329491,0.001806154,0.0008602613,0.003441985,0.002716359],"genre_scores_gemma":[0.8747353,0.0003064464,0.1190924,0.0001913321,0.00002974394,0.002200753,0.001592493,0.0005184844,0.001333024],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03296211,"threshold_uncertainty_score":0.1743223,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2211729651340077,"score_gpt":0.5383628778568699,"score_spread":0.3171899127228622,"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."}}