5418. Quantitative CT Biomarkers for Predicting Overall Survival in Patients with Non-IPF Fibrotic Interstitial Lung Disease (fILD)
Authors * Denotes Presenting Author
  1. Steven Rothenberg *; The University of Alabama at Birmingham
  2. Scott Grumley; The University of Alabama at Birmingham
  3. Teja Kulkarni; The University of Alabama at Birmingham
  4. Ryan Rauch; The University of Alabama at Birmingham
  5. John Eddins; The University of Alabama at Birmingham
  6. Seth Lirette; The University of Mississippi Medical Center
  7. Andrew D. Smith; The University of Alabama at Birmingham
To compare the accuracy of fully-automated and semi-automated quantitative CT (QCT) biomarkers for predicting transplant-free survival (TFS) in participants with non-IPF fibrotic interstitial lung disease (fILD).

Materials and Methods:
For this retrospective multi-institutional study, high-resolution CT (HRCT) chest images and survival outcomes were gathered from 42 centers and 414 participants with non-IPF fILD, who were prospectively enrolled in the Pulmonary Fibrosis Foundation Patient Registry. A fully automated artificial intelligence (AI) algorithm that was trained on external data was applied to the baseline axial HRCT images and used to quantify fibrotic extent, including whole lung percentage and rind percentage (FE-WL and FE-R, respectively). Fibrotic extent (FE) is defined as the quantity of ground glass opacity, reticulation, and honeycombing findings on CT. The pulmonary surface irregularity (PSI) score was measured in a semi-automated manner using the same baseline axial HRCT images. The gender-age-physiology (GAP) score, comprised of gender, age, and pulmonary function test results, was also available at baseline. Each of the biomarkers were categorized into three groups. Cut points for the QCT biomarkers were derived from exploratory data analysis and Youden J statistics. Kaplan-Meier plots with log rank test, cox hazards models, and concordance index were used to assess the accuracy of each biomarker for predicting TFS.

The cohort included 42% men and were 80% White. The mean age was 63 years. When comparing the highest to lowest stages for each biomarker, the HR for TFS was 6.1x for FE-WL (HR:6.1, <em>p</em> < .001), 3.8x for FE-R (HR:3.8, <em>p</em> < .001), 15.6x for the combination of FE-WL/FE-R (HR:15.6, <em>p</em> < .001), 3.9x for PSI score (HR:3.9, <em>p</em> < .001), 16.9x for the combination of FE-WL/FE-R/PSI score (HR:16.9, <em>p</em> < .001), 5.3x for GAP score (HR:5.3, <em>p</em> < .001), and 17.6x for the combination of FE-WL/FE-R/PSI score/GAP score (HR:17.6, <em>p</em> < .001). Concordance statistics for predicting TFS were 0.65 for FE-WL, 0.65 for FE-R, 0.71 for the combination of FE-WL/FE-R, 0.64 for PSI score, 0.71 for the combination of FE-WL/FE-R/PSI score, 0.65 for GAP score, and 0.68 for the combination of FE-W/FE-R/PSI score/GAP score.

A fully-automated artificial intelligence algorithm designed to quantify fibrotic extent on HRCT had higher accuracy for predicting TFS than the semi-automated PSI score and the GAP score in a multi-institutional cohort patients with non-IPF fILD. fILDs are progressive and fatal fibrotic lung diseases with variable disease course. Patients are universally evaluated by HRCT, and a QCT biomarker obtained from existing HRCT images with no additional patient cost or radiation can be valuable in clinical practice. In this study, we report the prognostic value of fully-automated and semiautomated quantitative CT biomarkers in comparison to the GAP score, a prognostic clinical marker.