A multimodal model for predicting progressive pulmonary fibrosis in fibrotic hypersensitivity pneumonitis using clinical features, pulmonary function, and deep learning–based CT analysis

Background Hypersensitivity pneumonitis (HP) is an immune-mediated interstitial lung disease (ILD) that occurs in susceptible individuals exposed to inhaled antigens. The INBUILD study demonstrated that fibrotic HP can show progressive behavior similar to other progressive fibrosing ILDs. However, predictors of progression in fibrotic HP remain poorly defined. Recent guidelines have introduced the concept of progressive pulmonary fibrosis (PPF), and several studies have reported the utility of artificial intelligence and deep learning–based quantitative computed tomography (CT) analysis for evaluating ILD progression. Methods This single-center retrospective observational study included 120 patients diagnosed with fibrotic HP through multidisciplinary discussion. We evaluated clinical data, high-resolution CT findings, pulmonary function tests, peripheral blood leukocyte telomere length, and automated deep learning–based quantitative CT analysis of ILD. Factors associated with the development of PPF were assessed using multivariable analysis. Results Among the 120 patients, 48 (40%) met the criteria for PPF. Multivariable analysis identified six independent predictors of PPF: digital clubbing, modified Medical Research Council dyspnea score, cough, humidifier use, diffusing capacity of the lung for carbon monoxide (DLCO), and the percentage of lung volume classified as consolidation with traction bronchiectasis by automated deep learning–based CT analysis. The predictive model achieved an area under the receiver operating characteristic curve of 0.85 (95% CI: 0.78, 0.92). Conclusions Specific clinical features, pulmonary function, and automated quantitative CT-derived abnormalities were identified as significant predictors of PPF in patients with fibrotic HP. DL-based CT quantification may provide objective imaging information that complements conventional clinical and radiological assessment for identifying patients at risk of PPF.