Application Of Multi-Inflammatory Index To Predict 28-Day Mortality In ICU Patients With Heart Failure: A Retrospective Machine Learning Study Based On The MIMIC-IV Database

Background Heart Failure(HF) is one of the leading cardiovascular diseases, and its high mortality rate in the Intensive Care Unit (ICU) has drawn increasing attention.HF mortality prediction is critical for developing individualized prevention and treatment plans.The objective of this study is to evaluate how effectively the Multi-Inflammatory Index—I, II, and III (MII-1, MII-2, MII-3)—can forecast mortality rates among ICU patients with heart failure. Methods A retrospective analysis was conducted on patients from the MIMIC-IV database, during which clinical data and laboratory findings were gathered. Feature selection was carried out utilizing the Boruta algorithm.We developed four machine learning models: XGBoost, Decision Tree,LightGBM and Naive Bayes.The performance of these models was assessed through five-fold cross-validation.To examine feature importance and model interpretability, SHAP values were employed. Results A total of 904 patients with HF were included in the final cohort for this study,with a median age of 71 (61–79)years.The ICU 28-day mortality for patients with HF was 23.23%.The Multi-Inflammatory Index of nonsurvival patients with HF in ICU was significantly higher than that of survival patients.Fourteen variables most associated with death from HF were selected using Boruta algorithm.In comparison,the LightGBM model demonstrated superior predictive performance among the four models, achieving an area under the curve (AUC) of 0.735. On the other hand, the DesicionTree machine exhibited limited generalization ability with an AUC of 0.575, indicating relatively poorer performance in prediction accuracy.The SHAP technique uncovers the 15 most significant predictors of HF based on their ranking in terms of importance, with the duration of shock index, age, and Multi-Inflammatory Index as the foremost predictor variable. Conclusion The LightGBM model demonstrated adequate sensitivity and accuracy. Multiple inflammatory Index play an important role in the prediction of death in ICU patients with heart failure.Consequently, this facilitates the development of improved treatment strategies and efficient allocation of resources for optimal patient care.