Early dynamic changes in platelet counts and 28-day mortality in sepsis patients: a retrospective cohort study using dynamic latent class model and generalized additive mixture model analysis.

Objective: This study investigates the relationship between early dynamic changes in platelet (PLT) counts and 28-day mortality in Chinese patients with sepsis, addressing the limitations of previous studies that focused on single baseline measurements.

Methods: In this retrospective cohort study, 266 sepsis patients admitted to Shenzhen Second People's Hospital from January 2023 to December 2024 were included. A dynamic latent class model analyzed the patterns of PLT count changes during the first week of hospitalization. The Cox proportional hazards regression model assessed the link between these dynamic changes and 28-day mortality, supported by sensitivity and subgroup analyses for robustness. The GAMM model compared PLT change trajectories over 7 days between the mortality and survival groups.

Results: After adjusting for various variables, participants with gradually increasing PLT counts (class 2), decreasing counts (class 3), and persistently low counts (class 4) had hazard ratios (HRs) for 28-day mortality of 1.687 (95% CI:0.380, 7.494), 3.710 (95% CI:1.124, 12.251), and 4.258 (95% CI:1.435, 12.636) respectively, compared to those with persistently high PLT counts (class 1). The GAMM model revealed that PLT counts for patients who died were significantly lower and had a downward trend, while the survival group's counts trended upward; the difference between the two groups generally exhibited an upward trend after admission, with a calculated average daily increase of 12.919 × 10/L.

Conclusion: Early dynamic changes in PLT counts (1-7 days) are independently associated with 28-day mortality in sepsis patients. Those with low and declining PLT counts are at a higher risk. By dynamically monitoring early changes in PLT may help identify high-risk patients and inform personalized treatment strategies, improving outcomes.