The predictive value of a novel inflammatory index for short-term mortality risk in elderly patients with severe heart failure:an interpretable analysis based on the shap algorithm
ZHANG Ting
CHEN Rui
CAO Wenzhai
Abstract：Objective To construct an interpretable machine learning model to predict the short-term mortality risk of elderly patients with severe heart failure,and to explore the predictive value of the new inflammatory index.Methods A retrospective analysis was conducted on elderly patients with heart failure diagnosed in the MIMIC-IV 3.1 database.The study participants were randomly assigned to either the training cohort(70%)or the validation cohort(30%).Six algorithms including logistic regression(LR),decision tree(DT),random forest(RF),adaptive boosting(AdaBoost),light gradient boosting machine(LightGBM),and gaussian naive bayes(NB)were used to construct the prediction model.The discriminatory ability of the predictive algorithm was assessed through analysis of ROC curves,precision-recall(P-R)curves,and calibration curves.SHAP was used for model interpretation and to select the core inflammatory index.The optimal cutoff value was determined using the ROC curve.Results A total of 1994 elderly patients with severe heart failure were included.Among them,253 patients died within 28 days(12.7%),and 1741 patients survived(87.3%).After initial screening,65 clinical features were included in the construction of the machine learning model.The LightGBM model showed the best predictive performance,with an Area Under the Curve(AUC)of[0.897(0.881-0.909)],an average precision(AP)of 0.86 on the P-R curve,and a calibration curve showing that the predicted probability was consistent with the actual observation results.SHAP value analysis revealed that acute physiology score III(APS III),Glasgow coma scale(GCS),monocyte to lymphocyte ratio(MLR),respiratory rate(RR),age,blood urea nitrogen,Oxford acute severity of illness score(OASIS),ACEI,neutrophil-to-lymphocyte ratio(NLR),and nutritional risk index(NRI)were important influencing factors.The AUC of MLR and NLR on the ROC curve were 0.682 and 0.667,respectively.With a cutoff value of 0.426 and 7.083,the sensitivity was 0.747 and 0.751,and the specificity was 0.529 and 0.503.Conclusion The LightGBM model can better predict the short-term mortality risk of elderly patients with severe heart failure.The new inflammatory indices,such as NLR and MLR,have potential clinical application value for the short-term mortality risk stratification of elderly patients with heart failure.
Keywords：heart failuremortality riskmonocyte-to-lymphocyte rationeutrophil-to-lymphocyte ratioSHAP algorithm
Publication Date：2026-01-30
Online Publishing Date：2026-01-27(First online date of this platform, not the publication date of the document)
Pages：6( 5-10 )