B23-09 Comparing Survival Outcomes Between Radiomic UIP and Visual UIP Classification in Fibrotic ILD

Rationale The usual interstitial pneumonia (UIP) pattern is the hallmark chest CT pattern for diagnosing idiopathic pulmonary fibrosis (IPF) and several other fibrotic interstitial lung diseases (ILD). Emerging therapies have shown remarkable potential in delaying fibrosis progression, further underscoring the need to reliably identify this UIP pattern earlier. Our group has previously developed and validated a deep learning (DL)-based radiomic classifier for UIP, called IQ-UIP, which recently received clearance from the U.S. Food and Drug Administration (FDA) in January 2025. In this study, we compare transplant-free survival between IQ-UIP and visual UIP classification in a large healthcare cohort of individuals with fibrotic ILD. Methods A retrospective cohort analysis was performed at the University of Michigan including all adult individuals presented at multidisciplinary ILD conference between 2016-2022 with a confirmed diagnosis of fibrotic ILD (n = 823). All baseline chest CTs in these individuals were processed through 4DMedical’s IQ-UIP tool to produce a binary DL-based UIP classification. Chest CTs identified as either “probable” or “definite” UIP by thoracic radiologists at ILD conference were combined, making up the visual UIP positive group. Transplant-free survival (TFS) was compared between UIP classification approaches using multivariable Cox proportional hazards regression models adjusted for age, sex, race, smoking history, autoimmune disease, and baseline forced vital capacity (FVC), with censoring at 36 months from the CT scan. Results 280 individuals were classified as IQ-UIP positive (34%) and over half (51%) of the cohort was identified as being visual UIP positive. IQ-UIP positive individuals had a 1.8-fold increased hazard of death or transplant (HR 1.84; 95% CI 1.40-2.41), whereas visual UIP classification was not associated with an increased hazard of death or transplant (HR 1.26, 95% CI 0.94-1.68). Agreement between IQ-UIP and visual UIP was fair (κ = 0.35), with 203 IQ-UIP negative individuals classified as visual UIP positive and 65 IQ-UIP positive individuals classified as visual UIP positive. Figure 1 shows Kaplan Meier plots of TFS according to both IQ-UIP and visual UIP classification. Individuals identified as IQ-UIP positive had significantly reduced TFS, regardless of their visual UIP classification. Conclusions IQ-UIP positive individuals had significantly decreased short-term TFS, regardless of visual UIP classification. Important next steps in this work include validating these preliminary findings prospectively to identify potentially high-risk IQ-UIP positive individuals that may benefit from earlier initiation of novel anti-fibrotic therapies.