Biological age model using explainable automated CT-based cardiometabolic biomarkers for phenotypic prediction of longevity
Perry J. Pickhardt, Michael W. Kattan, Matthew H. Lee, B. Dustin Pooler, Ayis Pyrros, Daniel Liu, Ryan Zea, Ronald M. Summers, John W. Garrett
Abstract
We derive and test a CT-based biological age model for predicting longevity, using an automated pipeline of explainable AI algorithms that quantifies skeletal muscle, abdominal fat, aortic calcification, bone density, and solid abdominal organs. We apply these AI tools to abdominal CT scans from 123,281 adults (mean age, 53.6 years; 47% women; median follow-up, 5.3 years). The final weighted CT biomarker selection was based on the index of prediction accuracy. The CT model significantly outperforms standard demographic data for predicting longevity (IPA = 29.2 vs. 21.7; 10-year AUC = 0.880 vs. 0.779; p < 0.001). Age- and sex-corrected survival hazard ratio for the highest-vs-lowest risk quartile was 8.73 (95% CI,8.14-9.36) for the CT biological age model, and increased to 24.79 after excluding cancer diagnoses within 5 years of CT. Muscle density, aortic plaque burden, visceral fat density, and bone density contributed the most. Here we show a personalized phenotypic CT biological age model that can be opportunistically-derived, regardless of clinical indication, to better inform risk assessment. The authors develop a model derived from an automated pipeline of explainable AI body composition tools applied to abdominal CT. They provide a tool for the personalized phenotypic assessment of biological aging that can be opportunistically derived, regardless of clinical indication.