Clinical evaluation of artificial intelligence supported software in the interpretation of spirometry results in a primary care setting
Rationale: Few primary care physicians perform spirometry tests or interpret spirometry results themselves. This results in assessments of chronic respiratory disease diagnoses not in line with guidelines and might lead to less time- and cost-effective secondary care visits. Since the interpretation of spirometry tests heavily relies on pattern recognition, AI-based software could potentially provide support to GPs performing spirometry testing. While previous studies demonstrated the added value of AI for the interpretation of full pulmonary function testing (PFT), this study aims to investigate the accuracy of GP assessment of spirometry curves alone with and without the use of the provided AI-supported software, ArtiQ.Spiro.
Methods: A total of 6 primary care practices in Belgium used ArtiQ.Spiro, during a 3-month period alongside existing referral structures.
The accuracy of the diagnostic hypothesis based on the clinical evaluation and interpretation of spirometry, supported by the AI software, was established by comparison to the diagnosis made by an expert panel of 3 pulmonologists. This panel reviewed the same data and all relevant information from the patient record 3 months after the spirometry session.
Results: In 77% of cases, GPs agreed with the diagnosis proposed by ArtiQ.Spiro. Combined with spirometry and AI support, GPs labeled patients correctly in 67% of patients. When comparing the AI-proposed diagnosis with the pulmonologists’ diagnosis, 82% of diagnoses were correct.
Conclusion: GPs often don’t perform and interpret spirometry consistently. AI has been proven to provide highly accurate diagnosis support as compared to pulmonologist opinion. As such, the investigated AI-based software should increase the diagnostic accuracy of chronic respiratory diseases in a primary care setting. Further research should focus on how to further improve the GP’s use of and collaboration with AI tools.
Authors: Authors: Willaert S.1, Maes J.2, De Vos M. 1,3, Topalovic M.2, Verbakel J. Y.1,4
Affiliations:
1. Department of Public Health and Primary Care, KU Leuven, KU Leuven, Leuven, Belgium
2. ArtiQ NV, Leuven, Belgium
3. Department of Electrical Engineering, KU Leuven, Leuven, Belgium
4. NIHR Community Healthcare Medtech and IVD cooperative, Nuffield Department of Primary Care Health Sciences, University of Oxford, UK
