Abstract
<jats:p>Background: Chronic respiratory diseases remain underdiagnosed among smokers, particularly in resource-limited public healthcare settings where routine spirometry is not consistently available. Artificial intelligence-supported screening may improve early identification of high-risk individuals requiring confirmatory evaluation. Objective: To evaluate the diagnostic accuracy of an AI-supported screening tool for detecting undiagnosed chronic respiratory disease among smokers attending urban public clinics in Lahore, Pakistan. Methods: This cross-sectional diagnostic accuracy study included 500 adult current or former smokers aged 35–70 years with at least 10 pack-years of exposure and no prior chronic respiratory disease diagnosis. Participants underwent AI-supported screening using demographic, smoking-related, clinical, symptom, pulse oximetry, and digital auscultation variables, followed by post-bronchodilator spirometry as the reference standard. Diagnostic performance was assessed using sensitivity, specificity, predictive values, overall accuracy, ROC analysis, calibration testing, and Cohen’s kappa agreement. Results: Spirometry confirmed chronic respiratory disease in 160 participants (32.0%). The AI tool correctly identified 130 true-positive and 240 true-negative cases, with 30 false negatives and 40 false positives. Overall accuracy was 88.0%, sensitivity 81.3%, specificity 85.7%, positive predictive value 76.5%, and negative predictive value 88.9%. The AUC was 0.91, Cohen’s kappa was 0.78, and the Hosmer–Lemeshow p-value was 0.41. Conclusion: AI-supported screening showed high diagnostic accuracy and substantial agreement with spirometry, supporting its potential role as a triage tool for early respiratory disease detection among smokers in public clinic settings</jats:p>