Group-based trajectory modeling of platelet dynamics in sepsis: from phenotypic identification to the exploration of prognostic mediators

TL;DR

This study identifies distinct platelet count patterns in sepsis patients and shows how these patterns are linked to survival rates and clinical outcomes.

Contribution

The novel use of group-based trajectory modeling to identify platelet dynamics subphenotypes and their prognostic implications in sepsis.

Findings

Three distinct platelet count trajectories were identified: persistent low level, high-level decline, and rebound rise.

Patients with persistent low platelet counts had significantly higher 28-day mortality compared to the other groups.

Platelet trajectories' effect on mortality is partially mediated through changes in SOFA score and APTT.

Abstract

Sepsis continues to impose a substantial global burden, with persistently high rates of hospitalization and mortality. A major challenge in advancing sepsis therapeutics lies in its inherent clinical and biological heterogeneity. Recent research has increasingly shifted from static biomarker assessment toward the identification of dynamic trajectory subtypes to better capture this heterogeneity.

This study aims to characterize the temporal patterns of platelet count changes during the first three days after admission among 280 sepsis patients using group-based trajectory modeling (GBTM), assess their association with clinical outcomes, and explore potential mediating pathways.

We retrospectively analyzed data from 280 sepsis patients admitted to the Department of Critical Care Medicine at Jiangsu University Affiliated People's Hospital between September 2022 and December 2024. Baseline demographics, clinical characteristics, and serial platelet counts were collected. Missing data were addressed using multiple imputation techniques. GBTM was applied to identify distinct trajectory patterns of platelet counts. The association between each trajectory and 28-day all-cause mortality was evaluated using multivariable Cox proportional hazards regression models. Additionally, mediation analysis was performed to investigate potential mechanisms underlying these associations.

GBTM revealed three distinct platelet count trajectories: persistent low level (71.79%), high-level decline (20.00%), and rebound rise (8.21%). Patients in the persistent low-level group exhibited significantly higher 28-day all-cause mortality compared to the other two groups (71.63% vs. 55.36% vs. 13.04%, p < 0.05) and had the shortest median survival time. After adjusting for key confounders, Cox regression showed that, relative to the persistent low-level group, the high-level decline group had a 42% lower risk of death (HR = 0.58, 95% CI: 0.36–0.92, p = 0.02), while the rebound rise group demonstrated an 93% reduction in mortality risk (HR = 0.07, 95% CI: 0.02–0.26, p < 0.001). Mediation analysis indicated that the effect of platelet trajectories on 28-day mortality may be partially mediated through changes in Sequential Organ Failure Assessment (SOFA) score and log-transformed APTT.

Dynamic modeling of platelet count trajectories enables effective identification of clinically meaningful subphenotypes in sepsis patients, offering a robust framework for prognosis prediction. This approach supports refined risk stratification and personalized management strategies, thereby providing novel insights into the pathophysiology and clinical care of sepsis.

Figures