Dynamic Perioperative Risk Stratification of In-Hospital Mortality After Elective Isolated On-Pump CABG: A Single-Center Firth Penalized Logistic Regression Analysis.

Background: Accurately assessing perioperative risk after elective on-pump coronary artery bypass grafting (CABG) is challenging, especially when it comes to identifying patients at high risk for early mortality. We explored whether including perioperative biochemical and hemodynamic variables could improve established preoperative risk evaluation. Methods: We analyzed 451 consecutive patients who underwent elective on-pump isolated CABG over two years, focusing on in-hospital mortality as the primary endpoint. The predictors were selected based on univariable analysis, clinical relevance, and multicollinearity assessment. To address the low number of mortality events, we used Firth-penalized logistic regression. The final model was assessed using discrimination, calibration, and internal validity using receiver operating characteristic analysis, calibration measures, and 1000-bootstrap resampling. A dataset-derived risk threshold was explored for risk stratification. Results: Eleven patients (2.44%) died during hospitalization. The final multivariable model included three variables independently associated with in-hospital mortalityEuroSCORE II (OR1.57, CI 95%: 1.13-2.14, p = 0.011), early postoperative vasoactive-inotropic score (VIS) (OR1.017, CI 95%: 1.008-1.033, p < 0.001) and preoperative CK-MB/CK ratio (OR21.06, CI 95%: 1.06-355, p = 0.046). The final model demonstrated good discrimination with an apparent AUC of 0.908 and an optimism-corrected AUC of 0.892. A dataset-derived risk-stratification threshold of predicted mortality ≥2.35% identified the highest-risk quintile, which contained 10 of the 11 observed deaths. Observed mortality was 11.1% in this quintile versus 0.28% among the remaining patients. Inflammatory indices showed limited predictive performance. Additionally, the use of an intra-aortic balloon pump was not independently associated with in-hospital mortality after adjusting for baseline risks. Conclusions: A dynamic model integrating preoperative risk assessment (EuroSCORE II and preoperative CK-MB/CK ratio) with early postoperative hemodynamic data (VIS) showed encouraging internal discrimination and calibration for mortality prediction after isolated on-pump CABG. Given the limited number of events and substantial uncertainty surrounding the CK-MB/CK ratio estimate, these findings should be considered exploratory and hypothesis-generating until externally validated.