Platelet-to-white blood cell ratio tracks mortality risk in ICU patients with HFpEF
Critically ill patients with heart failure with preserved ejection fraction often arrive in the ICU with multiple competing threats: hemodynamic stress, systemic inflammation, organ dysfunction and a narrow margin for recovery. In that setting, a ratio derived from a routine blood count has attracted attention because it may capture two biologic forces at once. Platelets may reflect vascular integrity and hemostatic reserve, while white blood cells signal inflammatory burden. In this retrospective analysis, a higher platelet-to-white blood cell ratio was associated with lower short-term mortality in ICU patients with HFpEF, but the gain in prediction beyond standard severity markers remained small.
The cohort from MIMIC-IV included 5,774 adult ICU patients with HFpEF, and 632 died within 28 days, a mortality rate of 10.95%. PWR was calculated from the first platelet and white blood cell measurements obtained within 24 hours of ICU admission, making it an early, readily available marker rather than a specialized assay. The association with mortality was not linear. Patients in higher PWR quartiles had progressively lower adjusted risk than those in the lowest quartile: Q2 had a hazard ratio of 0.79, Q3 had 0.58, and Q4 had 0.61, all versus Q1 after full adjustment, with a significant trend across quartiles. Restricted cubic splines supported the nonlinear shape of the relationship.
One sentence captures the signal plainly: higher PWR tracked with better survival.
Why this simple ratio may matter in HFpEF
HFpEF is not just a syndrome of congestion and stiff ventricles. It is also a disorder shaped by microvascular dysfunction, chronic inflammation, endothelial activation and repeated physiologic stress. Platelets and leukocytes sit close to that biology. A low ratio may indicate that inflammatory activation is dominating the hematologic profile, while a higher ratio may reflect a less inflammatory state or better preserved platelet availability relative to leukocyte burden. In the ICU, where HFpEF patients may be destabilized by infection, respiratory failure, renal dysfunction or fluid shifts, such a ratio can function as a compact proxy for global stress biology.
The study did not treat PWR as a standalone prognostic oracle. Investigators compared it with broader models built from established clinical variables and severity indicators. They also applied Boruta feature selection, machine learning methods, SHAP interpretation and repeated cross-validation to see whether the ratio added something beyond conventional bedside data. The answer was restrained. PWR contributed useful statistical information, but its clinical lift over the rest of the model was limited.
Machine learning improved ranking, not bedside certainty
Among five machine-learning approaches, random forest performed best in the internal holdout set, with an AUROC of 0.813 and an AUPRC of 0.387. A conventional unpenalized logistic regression model using the same 15 predictors reached an AUROC of 0.792 and an AUPRC of 0.389. The difference was modest. In other words, algorithmic complexity did not dramatically outperform a simpler clinical model, which is an important signal for intensivists who need tools that can travel from software to bedside use without much friction.
When the locked model was refitted in the full MIMIC-IV cohort and then applied to eICU, which contributed 685 patients and an in-hospital mortality rate of 13.3%, the AUROC was 0.767. Adding nonlinear PWR to the clinical baseline improved overall model fit on likelihood-ratio testing, but cross-validated discrimination barely moved, from 0.763 to 0.764. Calibration stayed comparable, and decision-curve analysis did not show a consistent net-benefit advantage. The ratio therefore behaved like a biologically plausible marker with real statistical associations, yet only limited incremental clinical utility once broad severity information was already in the model.
The prognostic signal is real, but the bedside gain is modest
This matters because ICU clinicians already rely on a dense web of data at admission: age, comorbidity burden, organ failure scores, blood pressure, renal function and respiratory status. A new marker must do more than correlate with outcome; it has to improve decisions. PWR did not clear that bar in a dramatic way. It may still have value as a fast, inexpensive complement when a clinician is trying to gauge risk early, especially if other information is incomplete or evolving. But it should not be mistaken for a replacement for established severity assessment.
The nonlinear curve also suggests that the ratio may behave differently at low versus high ranges, which is biologically plausible in HFpEF. Very low ratios may indicate a pronounced inflammatory state or consumption pattern linked to worse outcomes, while higher values may mark less deranged physiology. The available data do not prove causation, and they do not explain whether the ratio is simply a marker of illness severity or a direct reflection of pathophysiology relevant to mortality. Retrospective design leaves room for residual confounding, even after multivariable adjustment.
Practical implications for ICU risk assessment
Because PWR comes from routine blood counts, it can be computed without new ordering, extra cost or delay. That accessibility is the study’s strongest practical feature. In ICUs that already generate complete blood count data on admission, the ratio could be integrated into automated risk dashboards or used as an adjunct when clinicians are triaging patients with HFpEF for closer monitoring. It may be especially useful as a low-friction signal in systems where sophisticated biomarkers are unavailable.
Still, the external assessment was exploratory rather than endpoint-matched validation, since eICU lacked uniform post-discharge follow-up for a true 28-day mortality endpoint. The external dataset therefore used in-hospital mortality instead. That constraint limits direct comparability between the two cohorts and narrows how far the findings can be generalized. Even so, the direction of the association remained consistent, which lends some credibility to the biological signal.
The study also focused on patients who survived and remained in the ICU for at least 24 hours, because the ratio was defined from blood counts collected within that first day. Patients who died extremely early or were discharged very quickly were not part of the analytic frame. The result is a clinically relevant but still selected ICU population.
In HFpEF, prognosis often hinges on small shifts in inflammatory burden and physiologic reserve. This analysis places PWR among the short list of simple hematologic ratios that may help quantify that burden, while also showing how hard it is for any single marker to outperform the full clinical picture once severity is already well characterized.