AI-Assisted Spirometry Interpretation in Primary Care: A Randomized Controlled Trial

Le résumé fourni par la source

BackgroundSpirometry quality and confidence in spirometry interpretation are highly variable in primary care, contributing to underdiagnosis, overdiagnosis, and misdiagnosis of chronic respiratory diseases worldwide. Artificial intelligence (AI) decision support software has been shown to improve the accuracy of lung function interpretation in specialist care, but its applicability to primary care remains unknown.MethodsWe conducted a parallel-group, randomized, controlled superiority trial to evaluate whether AI decision support software improves spirometry interpretation performance by primary care clinicians. Clinicians working in primary care who refer for, perform, or interpret spirometry assessed 50 real-world patient spirometry records, providing the most likely diagnosis (“preferred diagnosis”) through an online platform either with (intervention) or without (control) AI decision support software. The primary outcome was the preferred diagnosis prediction performance, measured as the percentage of cases in which the preferred diagnosis agreed with the reference diagnosis predetermined by expert pulmonologists. A planned subgroup analysis focused on cases with a diagnosis of chronic obstructive pulmonary disease (COPD). Secondary outcomes were performance in differential diagnosis prediction, technical quality assessment, pattern interpretation, and self-rated confidence in interpretation.ResultsOut of 400 clinicians assessed for eligibility, 234 were randomly assigned, with 133 (57%) completing the assessment (intervention, n=67; control, n=66) — 73% female, 42% general practitioners, and 50% nurses. Compared with the control group, the addition of AI decision support software led to improvements in preferred diagnosis prediction performance in all cases (mean difference, 9.0; 95% confidence interval [CI], 4.5 to 13.3%; P=0.001) and in COPD cases (15.9; 95% CI, 9.0 to 22.7%; P<0.001). Differential diagnosis prediction and technical quality assessment performance also improved with intervention, but not pattern interpretation or clinician confidence levels.ConclusionsIn primary care clinicians, the adjunctive use of AI spirometry decision support software improved diagnosis prediction performance, which may help address the suboptimal interpretation of spirometry in primary care. (Funded by the National Institute for Health and Care Research and others; ClinicalTrials.gov identifier, NCT05933694.)

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.