B23-17 Associations Between Changes in High-resolution Computed Tomography Scores and Baseline Predictive Score in Patients With Idiopathic Pulmonary Fibrosis

Abstract Rationale The Single Time Point (STP) prediction score derived from baseline high-resolution CT (HRCT) has been developed to predict worsening disease on follow-up HRCT. Predicting progression using a single diagnostic HRCT scan would be valuable for precision medicine. Our aim is to test the association between HRCT changes and the baseline STP score, a machine learning-based radiomic HRCT biomarker, in an evaluation cohort of patients with idiopathic pulmonary fibrosis (IPF). Methods The Imaging Signature IPF (IS-IPF) study is a retrospective evaluation cohort that characterized CT-based imaging markers in 250 IPF subjects collected between March 2004 and October 2019. The baseline mean (±SD) age was 64.6 (±8.3) years; 75% were male. The mean percent predicted FVC and DLCO were 72.9% (±16.8) and 59.3% (±18.8), respectively. The mean STP score was 30.6% (±12.3%). Among the 250 subjects, 164 had follow-up HRCT scans (mean ±SD: 12.0 ±4.5 months). A previously developed machine learning algorithm was used to quantify the fibrosis burden and total ILD, represented by the quantitative lung fibrosis (QLF) and quantitative ILD (QILD) scores. Using a pre-determined STP threshold of 30%, we compared changes in QLF scores between groups using a two-sample t-test to evaluate the performance of the STP biomarker. Results Among the 164 patients with paired HRCT scans, 48 had an STP ≥30% and 116 had an STP 30%. The mean (SE) change in QLF was 5.4% (1.23) for the STP ≥30% group and −1.2% (0.70) for the STP 30% group. The mean follow-up duration for HRCT scans was approximately 12 months for both groups (SD: 5.5 months for STP ≥30% and 4.0 months for STP 30%). The baseline STP score was significantly associated with changes in quantitative CT scores (p 0.0001), indicating its predictive value for radiographic worsening. Similarly, total quantitative ILD (QILD) scores showed a comparable association and mean change (Mean (SE): 5.6% (1.5) for STP ≥30% and −2.1% (1.0) for STP 30%, respectively; p 0.0001). Conclusions The radiomic STP score derived from baseline HRCT predicts radiological worsening in IPF patients, as assessed by changes in QLF. The STP biomarker may be useful for enriching patient cohorts in clinical trials and guiding treatment decisions. Prospective validation of these findings is warranted. This abstract is funded by: Boehringer Ingelheim