Prognostic value of platelet-to-white blood cell ratio in critically ill patients with non-traumatic subarachnoid hemorrhage: a machine learning-driven study of the public database and the hospital information system.
Liu L, Huang Z, et al. • Frontiers in neurology • 2026
PWR is an independent predictor of short- and long-term mortality in ICU patients with NSAH, and machine learning models incorporating PWR significantly improve risk stratification.
Key Findings
Results
Higher platelet-to-white blood cell ratio (PWR) quartiles were independently associated with lower risk of 30-day mortality in critically ill non-traumatic subarachnoid hemorrhage patients.
Analysis based on 1,176 patients with critically ill NSAH enrolled from MIMIC-IV
Patients were predominantly male (58.8%) with a median age of 62 years
In multivariable Cox models, higher PWR quartiles were associated with HR = 0.55 (95% CI 0.35, 0.86, p = 0.009) for 30-day mortality after full adjustment
Restricted cubic splines (RCS) analysis suggested a linear association between PWR and mortality risk
Results
Higher PWR quartiles were independently associated with lower risk of 365-day mortality in critically ill NSAH patients.
In multivariable Cox models, higher PWR quartiles were associated with HR = 0.53 (95% CI 0.37, 0.77, p = 0.001) for 365-day mortality after full adjustment
The association was linear as suggested by RCS analysis
No significant interaction was observed by subgroup analysis across all subgroups examined
Results
The top-performing machine learning model incorporating PWR achieved an AUC of 0.868 for internal validation in predicting 30-day mortality.
Features were selected using Boruta and LASSO algorithms
SHAP (Shapley Additive exPlanations) analysis revealed a correlation between decreased PWR and an elevated predicted risk of mortality
Internal validation was conducted using the MIMIC-IV dataset
ROC curve and AUC were used to evaluate model performance
Results
External validation of the machine learning model demonstrated generalizability across two independent databases.
External validation in the eICU-Collaborative Research Database (eICU-CRD) yielded an AUC of 0.874
External validation in the Hospital Information System (HIS) of the First Affiliated Hospital of Harbin Medical University yielded an AUC of 0.763
The two external validation cohorts represent both a large multi-center US database and a Chinese hospital information system
Results
SHAP analysis identified decreased PWR as correlated with elevated predicted risk of mortality in the machine learning model.
SHAP (Shapley Additive exPlanations) method was utilized to provide model interpretability
The analysis confirmed the directionality of the PWR-mortality relationship observed in Cox regression analyses
Clinical features were selected alongside PWR using Boruta and LASSO algorithms for model development
What This Means
This research suggests that a simple blood test ratio — platelets divided by white blood cells (called PWR) — can help predict survival outcomes in critically ill patients hospitalized after non-traumatic subarachnoid hemorrhage (a type of brain bleed not caused by trauma). Analyzing data from 1,176 patients in a large intensive care database, researchers found that patients with higher PWR values had significantly lower risks of dying within 30 days and within one year, even after accounting for other health factors. The relationship was consistent across different patient subgroups.
The researchers also built machine learning models that combined PWR with other clinical information to predict 30-day mortality. These models performed well internally (AUC of 0.868) and were validated in two external datasets — one large multi-center US database and one Chinese hospital system — with AUC values of 0.874 and 0.763 respectively, suggesting the approach generalizes reasonably well across different healthcare settings. An interpretability analysis (SHAP) confirmed that lower PWR values were associated with higher predicted mortality risk.
This research suggests that PWR — a readily available and inexpensive measurement derived from routine blood tests — could serve as a useful prognostic tool for clinicians managing subarachnoid hemorrhage patients in the ICU. Incorporating PWR into machine learning-based risk models may support earlier identification of high-risk patients and more personalized care decisions, though further prospective studies would be needed to confirm clinical utility.
Liu L, Huang Z, Gui B, Xu L, Liu Y, Wang Y, et al.. (2026). Prognostic value of platelet-to-white blood cell ratio in critically ill patients with non-traumatic subarachnoid hemorrhage: a machine learning-driven study of the public database and the hospital information system.. Frontiers in neurology. https://doi.org/10.3389/fneur.2026.1864288