Introduction: Quality and interpretation accuracy of spirometry are variable in primary care. We aimed to evaluate whether an AI decision support software (ArtiQ.Spiro) improves the diagnostic prediction of primary care clinicians.

Methods: A parallel, two-group, randomised controlled trial of primary care clinicians in the UK who refer for, or interpret, spirometry. Clinicians were randomised 1:1 to independently interpret fifty de-identified, real-world patient spirometry sessions through an on-line platform either with (AI+) or without (AI-) an AI decision support software report. The primary outcome was diagnostic prediction performance (proportion of the fifty spirometry sessions where clinicians’ preferred diagnosis matched the reference diagnosis, derived by consensus from independent review of primary and secondary care notes and investigations by 3 pulmonologists). A Welch t test was used for analysis. Planned subgroup analysis of cases with a reference diagnosis of COPD (20/50) was performed.

Results: 234 participants were recruited and randomised from June 2023 to April 2024, 133 completed the study (AI+ n=66, AI- n=67): 73% female, 42% general practitioner, 54% on national spirometry register. The mean (SD) correct diagnostic prediction performance was significantly better for AI+ than AI- (58.7 [7.0]% versus 49.7 [16.6]%; mean [95%CI] treatment effect: 9.0 [4.5-13.3]%, p<0.01), and for reference COPD cases (mean [95% CI] treatment effect: 15.9 [9.0-22.7] %, p<0.0001).

Conclusion: Addition of AI decision support software improved diagnostic prediction performance in primary care clinicians assessing real-world spirometry, especially for cases of COPD.