Platelet to white blood cell ratio and short term mortality in critically ill patients with HFpEF

Critically ill patients with heart failure with preserved ejection fraction (HFpEF) require early and practical risk assessment. The platelet-to-white blood cell ratio (PWR), derived from routine blood counts, may reflect the balance between platelet availability and leukocyte-driven inflammatory stress. Its prognostic value in ICU patients with HFpEF remains uncertain. Methods: We performed a retrospective cohort study using MIMIC-IV v3.1 for derivation and primary analyses and eICU v2.0.1 for exploratory external assessment. Eligible participants were adult ICU patients with HFpEF. PWR was defined from the first platelet and white blood cell measurements recorded within 24 h after ICU admission. The primary endpoint in MIMIC-IV was 28-day all-cause mortality. Because eICU did not provide uniform post-discharge follow-up for determining 28-day mortality, in-hospital mortality was used only for an exploratory external assessment and was not considered an endpoint-matched validation outcome. Analyses included Kaplan–Meier curves, Cox models, restricted cubic splines, subgroup testing, Boruta selection, machine learning and SHAP interpretation. Incremental prognostic value was evaluated using nested fixed-horizon logistic models, repeated cross-validation, discrimination, calibration, reclassification and decision-curve analysis. Results: MIMIC-IV contributed 5774 patients, including 632 deaths within 28 days (10.95%). After full adjustment, mortality risk was lower in higher PWR quartiles than in Q1: Q2, hazard ratio (HR) 0.79 (95% confidence interval (CI) 0.64–0.96, P = 0.020); Q3, HR 0.58 (95% CI 0.46–0.72, P < 0.001); and Q4, HR 0.61 (95% CI 0.49–0.76, P < 0.001; P for trend < 0.001). Restricted cubic splines supported a nonlinear pattern. Among the five machine-learning approaches, random forest showed the highest discrimination in the internal holdout test set, with an AUROC of 0.813 and an AUPRC of 0.387. Conventional unpenalized logistic regression using the same 15 predictors achieved an AUROC of 0.792 and an AUPRC of 0.389, indicating only a modest difference between the two modelling approaches. After the locked model specification was refitted using the complete MIMIC-IV cohort and applied to eICU (n = 685; in-hospital mortality, 13.3%), the AUROC was 0.767 (95% CI 0.715–0.817). Although adding nonlinear PWR to the clinical baseline improved full-sample model fit (likelihood-ratio P = 0.012), the cross-validated AUC changed only from 0.763 to 0.764. Calibration remained comparable, and decision-curve analysis showed no consistent gain in net benefit. Conclusions: Among ICU patients with HFpEF who survived and remained in the ICU for at least 24 h, higher PWR was independently associated with lower 28-day mortality, and the association was nonlinear. Its incremental predictive contribution beyond comprehensive severity indicators was modest. Because PWR is derived from routine blood counts, it may provide an accessible complementary risk signal.