Development and Evaluation of a Comprehensive Machine Learning Tool Differentiating Idiopathic Pulmonary Fibrosis from Other Interstitial Lung Diseases in the Japanese nationwide registry (PROMISE study).

<bold>Background and Objective:</bold> Idiopathic Pulmonary Fibrosis (IPF) is characterized by a poor prognosis and a low rate of diagnostic consensus. Then, we have developed a comprehensive machine learning tool to differentiate IPF from other Interstitial Lung Diseases (ILDs) using non-invasive tests typically employed in ILD assessment. This study aimed to refine and evaluate the algorithm with the PROMISE study, a prospective, nationwide registry in Japan. <bold>Methods:</bold> From September 2020 to October 2022, newly diagnosed ILD cases from 133 respiratory hospitals across Japan were registered and reviewed by an online central Multidisciplinary Discussion (MDD). Our previously reported model (Furukawa et al. Respirology 2022;27:739-746.) was re-trained with MDD diagnoses from 97 institutions and evaluated on the remaining 36 institutions. <bold>Results:</bold> The study included 2,751 patients, average age 69.1 years, 63.5% male, with 652 diagnosed with IPF by MDD. Cases for interim analysis were randomly divided into training and testing datasets based on facility, with 1,782 and 655 cases. The algorithm previously reported an accuracy of 83.6% at a single institution, achieved 83.4% accuracy in this nationwide registry for MDD-IPF. <bold>Conclusion:</bold> Using non-invasive data, this comprehensive machine learning tool to differentially diagnose IPF from other ILDs showed efficient, accurate, and rapid diagnosis. This approach provides a notable advancement in the diagnostic process of ILDs, and a reliable alternative to traditional diagnostic methods.