A15-05 Screening for Interstitial Lung Disease Using Quantitative CT Thresholds

Abstract Rationale With the increasing availability of treatment options for fibrotic interstitial lung disease (ILD), identifying disease earlier on chest computed tomography (CT) is becoming critically important. Despite the widespread availability of chest CT to diagnose ILD, there remains no consensus on how quantitative CT (qCT) criteria should be used to screen for ILD and identify high-risk patients. The objective of this study was to create a qCT-based measure of ILD (qILD), test its associations with transplant-free survival (TFS) in two cohorts, and compare this qILD measure to visual assessment by radiology. Methods A retrospective cohort analysis was performed at the University of Michigan including adults 21 and older undergoing outpatient chest CT for clinical indications in 2017 (n = 3,537). Those with cancer or contrasted scans were excluded. Commercially available software (Computer-Aided Lung Informatics for Pathology Evaluation and Ratings or CALIPER) was used to generate qCT measures of pulmonary fibrosis and vascular volume. Iterative modeling using logistic regression was performed to develop a composite qILD measure dichotomized at Youden’s index. The association between qILD classification and five-year TFS was tested using multivariable Cox proportional hazard models. All CT reports were screened with natural language processing (NLP) for fibrotic keywords. The qILD measure was validated in a lung cancer screening cohort (n = 12,514), the National Lung Screening Trial (NLST), using low dose CTs. Results qILD classification was associated with 3-fold increased hazard of death or transplant (HR 3.11; 95% CI 2.62-3.69) in the University of Michigan cohort. Using NLP, visual ILD classification was also associated with 2-fold increased hazard of death or transplant (HR 2.38; 95% CI 1.97-2.87). Agreement between qILD and visual ILD was fair (κ = 0.40), with the worst survival outcomes in those identified as both qILD and visual ILD positive. Survival was similar for groups that displayed evidence of qILD or visual ILD, suggesting equivalence when either was present. In the NLST, qILD classification was also associated with 3-fold increased hazard of death or transplant (HR 3.17; 95% CI 2.12-4.75). Figure 1 shows Kaplan Meier plots of TFS according to (a) qILD classification, (b) both qILD and visual ILD classification in the University of Michigan cohort and (c) qILD classification in the NLST. Conclusions This qILD measure was associated with decreased TFS in healthcare and lung cancer screening cohorts. Future extensions include deploying a qCT measure of ILD prospectively to validate these definitions and to systematically screen for incident disease. This abstract is funded by: None