Real-world before-and-after evaluation of AI support for lung cancer diagnosis at three US lung nodule clinics
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
Health systems and payers require evidence that artificial intelligence (AI)-enabled decision support improves care delivery. Integrating AI into lung nodule management pathways may streamline workflows, improve identification and triage of patients with pulmonary nodules, and enable earlier lung cancer diagnosis. Yet real-world evidence of clinical utility remains limited. This study evaluated the impact of implementing a commercially available AI software to support the lung nodule and lung cancer care pathway across three US health systems.
Materials and Methods
A real-world before-and-after study compared clinic activity during matched 12.5-month pre-implementation and post-implementation periods at three US lung nodule clinics. Primary outcomes were the number of new patients and the number of invasive procedures performed. Secondary outcomes were numbers of lung cancers and early-stage (stages 0–II) lung cancers, and time from baseline CT to diagnosis. Analyses used Wilcoxon signed-rank and Mann-Whitney U tests, with medians and interquartile ranges (IQRs) reported for continuous measures.
Results
Across sites, new patients increased from 138 before to 232 after implementation, equivalent to 11.0 to 18.6 patients per month ( p =0.0002), with the median rising from 9 (IQR 7–14) to 16 (IQR 13–21) patients. Monthly invasive procedures increased from 5.0 to 7.8 ( p =0.0093). Monthly cancers increased from 3.0 to 3.8 ( p =0.3242), while monthly early-stage cancers rose from 1.4 to 2.4 ( p =0.0996). Median time to diagnosis was 58 days (IQR 33–153) before and 57 days (IQR 27–85) after implementation ( p =0.28), with IQR narrowing from 120 days to 58 days.
Conclusion
Implementation of the AI-enabled software was associated with significantly increased patient throughput resulting in an increase in lung cancer diagnoses. These results, observed prospectively post-implementation, suggest AI-assisted identification and triage can expand access to timely specialist review while maintaining diagnostic efficiency.
