B23-08 Deep Learning-based HRCT Assessment of Baseline Disease Severity in the Deupirfenidone Elevate IPF Trial: Comparison With a Well-Characterised IPF Population

Abstract Rationale The ELEVATE IPF trial was a randomized, placebo-controlled, dose-finding Phase 2b study of deupirfenidone (LYT-100) in patients with Idiopathic Pulmonary Fibrosis (IPF). 257 participants were randomized in a ratio of 1:1:1:1 to receive either 550 mg deupirfenidone, 825 mg deupirfenidone, 801 mg pirfenidone or placebo orally three times a day (TID) for 26 weeks. High-resolution CT (HRCT) scans were collected at screening to determine eligibility. Deep learning-based quantitative analysis of these HRCT scans enables objective assessment of disease severity and comparison of the enrolled cohort with well-characterised IPF populations, to determine whether the cohort is representative of real-world IPF patients. Methods The Qureight platform automated the segmentation of CT scans using 3D Convolutional Neural Network-based algorithms which quantified the total lung volume (Lung8) and percent fibrosis extent (Fibr8). 93% (240/257) patients had HRCT scans that passed quality criteria for running the models. Scan-derived features were correlated with baseline lung function test results using the Spearman’s rank correlation. Fibrosis extent was compared with baseline automated HRCT features from a prospectively recruited IPF cohort (PROFILE) using nonparametric tests. A p-value 0.05 was considered significant. Results Total lung volume correlated strongly with baseline forced vital capacity (FVC; r = 0.82) (Fig 1A). There was a moderate negative correlation shown between fibrosis volume (ml) and FVC (r = -0.56), and between percent fibrosis extent and percent predicted FVC (r = -0.49). Fibrosis extent was well matched when comparing the ELEVATE IPF and baseline scans of the PROFILE cohort (p-value = 0.77, median 13.5% vs 14.1%; rank biserial correlation coefficient = 0.01) (Fig1B). Conclusion The randomized cohort in the ELEVATE IPF trial is comparable to a well-characterised real world population of patients with IPF. No indication that the baseline disease severity of the cohort was different from real-world IPF patients was identified. Deep learning-based quantitative HRCT provides a robust tool for comparing recruited trial populations with other historic, well-characterised IPF cohorts. These deep learning models (with the addition of airway and vascular quantification) will be used for analysis of biomarker change and to ensure a representative cohort in the upcoming Phase 3 study of deupirfenidone 825 mg TID compared to pirfenidone 801 mg TID (SURPASS-IPF). This abstract is funded by: PureTech Health and Qureight Ltd