Levosimendan in Patients With Low Cardiac Ejection Fraction Undergoing Coronary Artery Bypass Grafting: Machine Learning Double Robust Causal Effect Analysis.

OBJECTIVES

The effectiveness of levosimendan in patients with severely impaired left ventricular function undergoing coronary artery bypass grafting (CABG) remains debated. While some early publications suggested a survival benefit, recent studies show conflicting results. This study used causal analysis to determine if preoperative levosimendan improves 30-day survival in this selected population.

DESIGN

Retrospective cohort study with machine learning double-robust causal effect analysis.

SETTING

Institutional cardiac surgery database.

PARTICIPANTS

A total of 973 patients with reduced left ventricular ejection fraction (LVEF ≤30%) undergoing CABG with or without concomitant valvular surgery.

INTERVENTIONS: Preoperative levosimendan administration versus no levosimendan.

MEASUREMENTS AND MAIN

RESULTS

The primary outcome was 30-day all-cause mortality. Mortality prediction was modeled using extreme gradient boosting with Shapley additive explanations to quantify predictor contributions. Levosimendan contributed minimally to mortality prediction; age, renal function, and surgical complexity were the dominant predictors. The estimated average treatment effect of levosimendan on 30-day mortality was negligible (0.12 percentage points; 95% confidence interval [CI], -0.84 to 1.56). Null effects were consistent across double-robust targeted maximum likelihood estimation (difference, 1.08 percentage points; 95% CI, -6.66 to 10.71) and overlap weighting (0.12 percentage points; 95% CI, -0.80 to 1.58).

CONCLUSIONS

In this study, levosimendan does not significantly impact 30-day mortality in patients undergoing CABG with severely reduced LVEF (≤30%). Machine learning double-robust causal inference demonstrated a lack of significant therapeutic effect in this population.