Derivation and validation of a prediction model for histopathologic fibrotic hypersensitivity pneumonitis

• DOI: 10.1016/j.rmed.2021.106598
• Publication Date: 2021-08-31T00:14:23Z
• Authors: Federica De Giacomi; Darin White; Paul A. Decker; Laszlo T. Vaszar; Nathan Sandbo; Augustine S. Lee; Jay H. Ryu; Teng Moua
• Keywords: Hypersensitivity pneumonitis; Usual interstitial pneumonia

Clinical differentiation of fibrotic hypersensitivity pneumonitis (f-HP) remains challenging given variable and overlapping presentations with other fibrotic interstitial lung disease (f-ILD).We derived a multivariable model for predicting histopathologic f-HP to better inform multidisciplinary team discussion (MDD) diagnosis, particularly when biopsy may be unsafe or cannot be achieved.Patients with histopathologically-defined f-HP and other overlapping f-ILD were reviewed for distinguishing clinical and radiological variables. Using elastic net logistic regression, a penalized regression approach to minimize overfitting, a clinical model built on non-invasive assessments was derived for the prediction of histopathologic f-HP. This model was then validated in an independently derived external cohort from three sites.The derivation and validation cohorts consisted of 248 (84 cHP and 164 other f-ILD) and 157 (82 f-HP and 75 other f-ILD) histopathologically-defined patients, respectively (total study N = 405). Variables retained from the elastic net model included age in years (regression coefficient 0.033), male sex (-1.109), positive exposure history (1.318), percent predicted forced vital capacity (-0.021), radiologic peribronchovascular axial ILD distribution (0.199), mid (-0.22) or lower lobe (-0.839) craniocaudal or patchy (0.287) ILD distribution, upper (1.188) or equivalent upper and lower lobe (0.237) traction bronchiectasis, mosaic attenuation (1.164), and centrilobular nodules (2.045). Bias corrected AUC was 0.84 (standard error = 0.02) for the derivation cohort and 0.80 (CI 0.73-0.87) for the validation cohort.This multivariable model demonstrated good predictive performance for delineating histopathologically-defined f-HP from other f-ILD as a means of avoiding or justifying biopsy and supporting MDD diagnostic confidence.