Research Article: Development and internal validation of a prediction model for hypoxic hepatitis after coronary artery bypass grafting with cardiopulmonary bypass
Abstract:
Hypoxic hepatitis (HH) after on-pump coronary artery bypass grafting (CABG) is associated with poor outcomes, but dedicated prediction tools are lacking. We developed and internally validated a parsimonious model for severe postoperative aminotransferase elevation compatible with HH.
This retrospective single-center study included 626 consecutive adults who underwent CABG with cardiopulmonary bypass at Beijing Anzhen Hospital between June 2020 and June 2023. The outcome was peak postoperative alanine aminotransferase or aspartate aminotransferase ?800 U/L after exclusion of alternative causes of acute liver injury during data extraction. Missing predictor data were addressed with random-forest multiple imputation (20 datasets). Prespecified predictors were screened with LASSO penalized logistic regression in each imputed dataset using 10-fold cross-validation; stably selected variables entered the final logistic model. Internal validation used bootstrap optimism correction, with temporal split and complete-case analyses as secondary analyses.
The outcome occurred in 57/626 patients (9.1%) and was associated with higher in-hospital mortality (49.1% vs. 4.6%; P < 0.001). Final predictors were total sternotomy count (OR 3.533, 95% CI 1.724–7.241), cardiopulmonary bypass duration (OR 1.047 per minute, 95% CI 1.035–1.059), and peak intraoperative lactate (OR 1.403 per mmol/L, 95% CI 1.251–1.574). Apparent and optimism-corrected AUCs were 0.907 and 0.903; temporal validation AUC was 0.955.
A three-predictor model estimated the risk of severe postoperative aminotransferase elevation compatible with HH after on-pump CABG and may support intraoperative or early postoperative risk awareness. External validation is required before routine use.
Introduction:
Hypoxic hepatitis (HH) after on-pump coronary artery bypass grafting (CABG) is associated with poor outcomes, but dedicated prediction tools are lacking. We developed and internally validated a parsimonious model for severe postoperative aminotransferase elevation compatible with HH.
Read more