Association and Discriminative Performance of a Bilirubin–D-Dimer–Albumin Index for LVEF

Background: Accurate identification of reduced left ventricular ejection fraction (LVEF) is important for risk stratification and management of patients with heart failure (HF). Because HF involves multiple interrelated pathophysiological processes, a combination of biomarkers may provide information beyond that offered by individual markers. This study evaluated the association and discriminatory performance of the bilirubin–D-dimer–albumin (BDA) Index for identifying LVEF < 45% and compared its performance with that of NT-proBNP. Methods: In this single-centre retrospective study, 338 patients with HF were categorized according to echocardiographic LVEF as < 45% (n=149) or ≥ 45% (n=189). The BDA Index was calculated as ln(total bilirubin [μmol/L]) × ln(D-dimer [ng/mL]) / albumin [g/L]. Discrimination was assessed using receiver operating characteristic (ROC) analysis and DeLong’s test. Multivariable logistic regression was used to evaluate the association between the BDA Index and LVEF < 45%, with internal validation performed using 1000 bootstrap samples and 10-fold cross-validation. Decision curve analysis (DCA) was used to assess the potential clinical utility of the model. Results: The BDA Index was significantly higher in patients with LVEF < 45% (P< 0.001). Its AUC was 0.902 (95% CI, 0.870– 0.934), compared with 0.773 (95% CI, 0.723– 0.823) for NT-proBNP (DeLong’s test, P< 0.001). At the Youden-derived cutoff of 0.435, sensitivity and specificity were 87.9% and 78.3%, respectively. After multivariable adjustment, each 0.1-unit increase in the BDA Index was associated with LVEF < 45% (OR, 4.65; 95% CI, 2.94– 7.36; P< 0.001). Adding the BDA Index significantly improved model fit (likelihood-ratio χ 2 =73.44, df=1, P< 0.001) and increased the AUC from 0.873 to 0.934. Internal validation yielded an optimism-adjusted AUC of 0.908, while 10-fold cross-validation yielded a mean AUC of 0.901. DCA showed greater estimated net benefit for the model incorporating the BDA Index across threshold probabilities of 5%– 80%. Conclusion: In this single-centre retrospective cohort, the BDA Index showed good discriminatory performance for identifying LVEF < 45% and provided additional information beyond NT-proBNP and other clinical and laboratory variables. Prospective multicentre studies with independent external validation are needed to assess its reproducibility, calibration, and potential clinical utility.