Multimodal prediction of documented 1-year asthma exacerbation risk incorporating a static baseline CT digital-twin framework: a retrospective cohort study

Background To develop and conditionally evaluate a multimodal prediction model incorporating a static baseline CT digital-twin framework for documented 1-year asthma exacerbation risk. Methods This retrospective single-center study analyzed 505 adults with asthma. Baseline three-dimensional CT reconstruction supplied quantitative features; the static baseline CT digital-twin framework did not simulate physiology or update dynamically. A LASSO-selected 10-variable logistic score was evaluated using nested domain comparisons, calibration, Brier score, and 1,000 bootstrap refits conditional on the selected predictor set. An available-variable clinical comparator was evaluated with and without CT features. Results Seventy-two patients (14.3%) had a documented exacerbation. The final score had apparent AUC 0.880 (conditional bootstrap 95% CI, 0.836–0.917) and Brier score 0.086. Fixed-predictor bootstrap correction yielded AUC 0.856, calibration slope 0.875, intercept −0.182, and Brier score 0.094, without accounting for the original variable-selection step. CT increased AUC from 0.880 to 0.894 in the fully nested comparison (likelihood-ratio P = 0.070), and from 0.725 to 0.774 in the available-variable comparator ( P < 0.001). These were apparent, exploratory comparisons, not external validation. Conclusion The multimodal score showed conditional internal performance, with a modest CT increment in the fully nested comparison and a larger apparent increment in the restricted available-variable comparator. The CT contribution remains exploratory and comparator-dependent. External validation, full selection-aware validation, quantitative CT repeatability assessment, and prospective impact evaluation are required before clinical use.