Machine learning model and hemoglobin to red cell distribution width ratio evaluates all-cause mortality in pulmonary embolism.

The ratio of hemoglobin (Hb) to red blood cell distribution width (RDW), known as HRR, functions as an innovative indicator related to prognosis. However, whether HRR can predict the mortality for pulmonary embolism (PE) patients remains ambiguous. A retrospective cohort study was conducted using the MIMIC IV database (3.0), All patients were categorized into four groups based on the HRR. We investigated the association between HRR and PE mortality. Cox regression models were used to evaluate these associations, while restricted cubic spline (RCS) regressions assessed potential nonlinear relationships. In addition, six machine learning models, including random survival forest (RSF), conditional Inference Tree(ctree), gradient boosting machine (gbm), nearest neighbors (nn), and extreme gradient boosting (xgboost), were applied, with Shapley additive explanation (SHAP) are used to determine the importance of characteristics. 2,272 PE patients were eligible for analysis. Our study identified both age and HRR levels (both with OR > 1, P < 0.05) as significant predictors of 30-day and 365-day mortality in PE patients admitted to the ICU. In Cox regression analysis, both age and HRR (both with HR > 1, P < 0.05) also emerged as prognostic risk factors for 30-day and 365-day mortality in this patient population. KM analysis demonstrated that patients with PE who were older or had increased HRR levels while hospitalized or in the ICU exhibited considerably reduced survival rates in comparison to younger individuals or those with lower HRR levels (P < 0.0001). Additionally, the RCS analysis revealed a pronounced nonlinear association between HRR levels and the risk of mortality. Validation set, coxph (ROC: 0.772) demonstrated superior predictive accuracy for these endpoints. identifying HRR as a vital component of mortality. A lower HRR correlates with high mortality rate in patients with PE patients. This model could serve as a useful tool for guiding mortality, assisting in clinical decision-making and improving patient management outcomes.