Association of CT Features with Radiologic Patterns in Interstitial Lung Disease: Multinomial Analysis in CARE-PF

Background Clinical practice guidelines define radiologic pattern categories based on the integration of multiple imaging features. However, the relative importance of each feature is unknown. Purpose To determine the relative weights of imaging features in distinguishing radiologic patterns and identify incongruencies between guideline-defined and radiologist-assigned patterns in fibrotic interstitial lung disease (ILD). Materials and Methods In this secondary analysis of the prospective Canadian Registry for Pulmonary Fibrosis (CARE-PF), consecutive patients were evaluated in multidisciplinary discussion between January 2021 and March 2022, with documentation of features and their extent at high-resolution CT. Radiologic pattern was identified according to American Thoracic Society clinical guidelines (guideline-defined pattern) and radiologist impression (radiologist-assigned pattern). Receiver operating characteristic curves were used to evaluate how continuous features distinguished usual interstitial pneumonia (UIP), fibrotic hypersensitivity pneumonitis (fHP), and nonspecific interstitial pneumonia (NSIP) patterns. Logistic regression was used to evaluate discordance between guideline-defined and radiologist-assigned patterns. A multinomial model was used to quantify the association of individual radiologic features with radiologist-assigned patterns. Results A total of 1498 patients (mean age, 66 years ± 12 [SD]; 753 male patients) were included, with radiologist-assigned patterns of UIP (36%; 544 of 1498), fHP (17%; 250 of 1498), NSIP (33%; 499 of 1498), and "no confident pattern" (14%; 205 of 1498). More honeycombing, less total ground-glass opacity (GGO), and less pure GGO distinguished UIP from non-UIP (area under the receiver operating characteristic curve [AUC], 0.75, 0.81, and 0.77, respectively); more hypoattenuating lung differentiated fHP from non-fHP (AUC, 0.84); and more total GGO and less honeycombing distinguished NSIP from non-NSIP (AUC, 0.71 and 0.70, respectively), with features with greater than 10% lung involvement commonly demonstrating specificities of 90% or greater. Certain features led radiologists to disagree with guideline-defined UIP (eg, admixed GGO, central component of disease), guideline-defined fHP (consolidation, peripheral and basal distributions), and guideline-defined NSIP (honeycombing, reticulation). Features most helpful for pattern identification according to experienced radiologists (odds ratio ≥3 or ≤0.33) were related to distribution or distinctive findings (eg, three-density sign, axillary lymphadenopathy). Conclusion Key continuous features with greater than 10% lung involvement commonly demonstrated good specificity, and radiologists emphasized disease distribution and distinctive dichotomous features when determining fibrotic ILD patterns. © RSNA, 2026 <i>Supplemental material is available for this article.</i> See also the editorial by Czum in this issue.