P3 Integrating AI-assisted spirometry: validation and user experience in a COPD diagnostic pathway

Abstract

<h3>Introduction</h3> Accurate interpretation of spirometry is essential for diagnosing chronic obstructive pulmonary disease (COPD). AI-based tools have the potential to enhance diagnostic accuracy and streamline workflow. ArtiQ.Spiro is an AI-based software that provides spirometry quality feedback and diagnostic support. As part of efforts to integrate ArtiQ.Spiro into our COPD diagnostic pathway in Glasgow, we conducted validation studies comparing its performance with clinician interpretation and assessed user experience. <h3>Methods</h3> We retrospectively analysed spirometry data from two separate cohorts totalling 248 patients who attended for direct-access COPD diagnostic pathway in 2022 and 2024. The diagnostic support outputs of ArtiQ.Spiro were compared to interpretations by ARTP-registered clinicians and final diagnostic pathway outcomes to determine concordance. Additionally, we conducted a prospective validation where ArtiQ.Spiro’s interpretations were obtained by assistant physiologists during test acquisition and initial interpretation for 125 consecutive tests. User experience feedback was collected from both trainee and senior assistant physiologists. <h3>Results</h3> ArtiQ. Spiro demonstrated excellent concordance with pathway outcomes in the retrospective cohort. There was near-complete agreement between a ‘normal’ AI interpretation and a ‘normal’ clinician-reported spirometry and pathway outcome (NPV = 0.942), with good agreement by Artiq.Spiro alone for COPD identification (PPV = 0.69). In the prospective validation, 90% of tests were correctly interpreted by both assistant physiologists and ArtiQ.Spiro. Notably, 10% of initial assistant physiologist interpretations were incorrect but were correctly identified by ArtiQ.Spiro. Trainee assistant physiologists reported significant benefits from ArtiQ.Spiro’s quality control and reporting components, highlighting its value in improving learning regarding test selection, interpretation, and reversibility decision-making. Senior assistants provided positive feedback, valuing continued access to the software and expressing willingness to issue reports for normal spirometry when spirometry was technically adequate, and assistant and ArtiQ.Spiro physiological interpretations were concordant. <h3>Conclusions</h3> Based on these validation results and positive user experience across different clinician groups, we will proceed with implementing ArtiQ.Spiro in our COPD diagnostic pathway, which currently manages over 600 patients per month. AI-assisted interpretation has the potential to significantly support trainee and experienced clinicians, improve diagnostic accuracy, optimise resource utilisation, and enhance early accurate diagnosis of COPD, while releasing senior clinician time for complex cases.

Ähnliche Arbeiten