Seven-day platelet-to-white blood cell ratio trajectories and mortality in sepsis: a multicenter retrospective study using MIMIC-IV and eICU-CRD

Sepsis involves closely linked inflammatory and coagulation disturbances. The platelet-to-white blood cell ratio (PWR) integrates these two routinely measured components, but most studies have evaluated PWR only at admission. We investigated whether 7-day PWR trajectories identify clinically distinct phenotypes associated with mortality. This retrospective cohort study included adults with sepsis from MIMIC-IV (derivation cohort) and eICU-CRD (external validation cohort). Daily PWR values from ICU Days 0–6 were clustered in MIMIC-IV using longitudinal K-means, and the resulting cluster centers were applied unchanged to eICU-CRD. The primary outcome was 28-day all-cause mortality in MIMIC-IV and 28-day in-hospital mortality in eICU-CRD; 90-day all-cause mortality was additionally evaluated in MIMIC-IV. Associations were assessed using multivariable Cox regression. Age- and sex-adjusted restricted cubic splines evaluated the association between baseline PWR and mortality, and time-dependent receiver operating characteristic analyses assessed discrimination at 7, 14, and 28 days. The analysis included 33,506 patients (22,291 in MIMIC-IV and 11,215 in eICU-CRD). Two trajectories were identified: a recovery phenotype with higher and increasing PWR and a nonresolving phenotype with persistently lower PWR. The nonresolving phenotype was associated with higher 28-day mortality after full adjustment in MIMIC-IV (hazard ratio [HR], 1.51; 95% confidence interval [CI], 1.33–1.71) and eICU-CRD (HR, 1.99; 95% CI 1.69–2.34). The associations remained after additional adjustment for Day 0 PWR. Baseline PWR showed a nonlinear association with mortality, with data-derived breakpoints of 9.7 for 28-day mortality and 10.2 for 90-day mortality in MIMIC-IV. The model combining PWR trajectory, SOFA, and APS III achieved a 7-day AUC of 0.762 in MIMIC-IV and 0.721 in eICU-CRD. A persistently low PWR trajectory was independently associated with mortality in two ICU cohorts and provided prognostic information beyond admission PWR and established severity scores. Dynamic PWR assessment may complement conventional risk stratification, although prospective validation is required before clinical implementation.