AI-Enhanced Spirometry Interpretation in Primary Care: Evidence from a Randomized Controlled Trial

Highlights

• AI decision support software improves diagnostic accuracy of spirometry interpretation among primary care clinicians.
• Greatest benefit observed in cases of chronic obstructive pulmonary disease (COPD), with a 15.9% improvement over standard care.
• AI assistance also enhances differential diagnosis and technical quality assessment, but does not significantly increase clinicians’ confidence or improve pattern interpretation.
• AI may help address long-standing issues of under- and misdiagnosis of respiratory diseases in primary care.

Study Background and Disease Burden

Chronic respiratory diseases, including chronic obstructive pulmonary disease (COPD) and asthma, are leading causes of morbidity and mortality worldwide. Spirometry is the gold standard for diagnosing and monitoring these conditions, yet studies consistently reveal that both the quality of spirometry and clinicians’ confidence in interpretation are highly variable in primary care. This variability contributes to underdiagnosis, overdiagnosis, and misdiagnosis—resulting in missed treatment opportunities, inappropriate therapy, and increased healthcare costs. While artificial intelligence (AI) decision support systems have shown promise in enhancing lung function interpretation in specialist settings, their utility in the frontline context of primary care, where clinician expertise and resources are more variable, has remained unclear.

Study Design

This parallel-group, randomized, controlled superiority trial (ClinicalTrials.gov identifier, NCT05933694) aimed to determine whether AI decision support software can improve the performance of primary care clinicians in interpreting spirometry results. The trial enrolled clinicians (general practitioners, nurses, and others involved in spirometry) who were randomized to assess 50 real-world spirometry cases via an online platform, either with (intervention) or without (control) the aid of AI decision support software. The primary endpoint was the agreement between clinicians’ preferred diagnosis and the reference diagnosis established by expert pulmonologists. Key secondary endpoints included accuracy in differential diagnosis, assessment of technical quality of spirometry, interpretation of spirometric patterns, and clinicians’ self-rated confidence.

Key Findings

A total of 400 clinicians were screened, of whom 234 were randomized and 133 (67 in intervention, 66 in control) completed the full assessment. The cohort was predominantly female (73%), with 42% general practitioners and 50% nurses, reflecting the real-world profile of spirometry users in primary care.

The addition of AI decision support software resulted in a statistically significant improvement in diagnostic performance. The mean difference in preferred diagnosis agreement with the expert reference was 9.0 percentage points higher in the AI-assisted group (95% CI, 4.5 to 13.3%; P=0.001). Notably, in the planned subgroup analysis of COPD cases, the improvement was even more pronounced, with a mean difference of 15.9 percentage points (95% CI, 9.0 to 22.7%; P<0.001).

Secondary outcomes revealed that AI support improved clinicians’ ability to provide accurate differential diagnoses and assess the technical quality of spirometry tracings. However, AI support did not significantly impact pattern recognition (e.g., distinguishing between obstructive and restrictive patterns) or self-reported confidence in interpretation.

The results underscore both the clinical and statistical significance of AI support, particularly in improving the recognition of COPD—a condition that is frequently underdiagnosed in primary care settings. The software’s impact on technical quality assessment is also notable, given that suboptimal test quality undermines the entire diagnostic pathway of spirometry.

Expert Commentary

These findings align with previous specialist-based studies, extending the benefits of AI to the primary care domain, where the need for decision support is arguably even greater. As highlighted by recent global guidelines (e.g., GOLD and GINA), correct interpretation of spirometry is foundational to respiratory disease management, yet is often lacking in community settings due to limited training and experience. The trial’s pragmatic approach—using real-world cases and a diverse clinician population—enhances the external validity of its findings.

However, several limitations merit consideration. First, the online study format does not replicate the full complexity of clinical workflow, patient interaction, or real-time decision making. Second, the absence of improvement in clinician confidence may indicate a need for more integrated, educational AI tools rather than solely diagnostic support. Finally, the relatively modest absolute improvement in diagnostic performance suggests that AI should augment, not replace, clinician judgment and that ongoing training in spirometry remains essential.

Conclusion

This randomized controlled trial provides robust evidence that AI-assisted spirometry interpretation can significantly improve diagnostic accuracy among primary care clinicians, particularly for COPD. By reducing misdiagnosis and enhancing technical assessment, AI has the potential to address critical gaps in respiratory disease management at the primary care level. Future research should explore the integration of AI tools into routine clinical workflows, their impact on patient outcomes, and strategies to enhance clinician engagement and confidence.

References

1. AI-Assisted Spirometry Interpretation in Primary Care: A Randomized Controlled Trial. NEJM AI. 2025;2(8). DOI: 10.1056/AIoa2400804.
2. Global Initiative for Chronic Obstructive Lung Disease (GOLD) 2024 Report. https://goldcopd.org.
3. Culver BH, et al. Recommendations for a Standardized Pulmonary Function Report. An Official American Thoracic Society Technical Statement. Am J Respir Crit Care Med. 2017;196(11):1463-1472.