Risk prediction for early adverse events after intervention in adults with bicuspid aortic valve stenosis with or without concomitant aortic surgery: a retrospective cohort study

Abstract

Early adverse events after intervention for bicuspid aortic valve stenosis remain clinically important, whereas prediction models tailored to this population are limited. We aimed to develop and internally validate a risk model for early adverse events. We conducted a retrospective single-center cohort study of 634 adults treated between January 2011 and December 2023. Completion of the index intervention was defined as the prediction landmark, and 54 candidate predictors available at or before this time point were considered. The primary outcome was a composite of early adverse events occurring during the index hospitalization or within 30 days after the intervention. Predictor selection was performed using least absolute shrinkage and selection operator (LASSO) logistic regression, followed by multivariable logistic regression. Model performance was assessed using discrimination, calibration, bootstrap internal validation, decision-curve analysis, and sensitivity analyses. Early adverse events occurred in 154 patients (24.3%), including 16 deaths (2.5%). LASSO retained six predictors: operative duration, red blood cell count, creatinine, myoglobin, pro-B-type natriuretic peptide, and procalcitonin. The final post-LASSO logistic model had an apparent area under the curve (AUC) of 0.826 (95% confidence interval, 0.785–0.867). Full-development-pipeline bootstrap validation yielded an optimism-corrected AUC of 0.790 and calibration slope of 0.807, whereas bootstrap validation of the fixed final model yielded an optimism-corrected AUC of 0.820 and calibration slope of 0.965. Sensitivity analyses produced broadly similar discrimination, and decision-curve analysis indicated greater net benefit than treat-all and treat-none strategies across threshold probabilities of approximately 7.5–60%. A six-predictor post-LASSO logistic model based on information available by completion of the index intervention showed moderate-to-good discrimination for early adverse events. External validation is required before the model can be considered for routine clinical application.