Association between the changing trends of platelet distribution width and in-hospital mortality in critically ill patients with sepsis: a multicenter study based on machine learning.

Background: Sepsis remains the leading cause of in-hospital mortality in critically ill patients. Platelet distribution width (PDW), an indicator of platelet activation and variability, is associated with inflammation and coagulation dysfunction during sepsis. However, dynamic changes in PDW and their association with patient outcomes remain unexplored. This study investigated the relationship between changing PDW trends and in-hospital mortality in critically ill patients with sepsis using machine learning techniques for robust analysis.

Methods: In the model development cohort, inpatient admissions fulfilling the sepsis 3.0 criteria in the Intensive Care Unit of Shenzhen People's Hospital were analyzed. PDW measurements were obtained at six-time points: First Day (D1), Second Day (D2), Third Day (D3), and the last three days before discharge (LD-3, LD-2, and LD-1). PDW was compared between survivors and non-survivors. Group-based trajectory modeling identified distinct PDW trajectory groups, and patient characteristics and outcomes were analyzed. The model was externally validated at a second hospital using identical inclusion criteria.

Results: A total of 1,090 and 429 patients with sepsis were included in the development and validation cohorts, respectively. Four distinct PDW trajectory groups emerged in the development cohort: "PDW Rapidly Increasing Group" (n = 174; 15.96%), "Low PDW Stable Group" (n = 416; 38.17%), "Moderate PDW Stable Group" (n = 421; 38.62%), and "High PDW Group" (n = 79; 7.25%). Subjects in the "PDW Rapidly Increasing Group" were the oldest, exhibiting the highest levels of inflammatory markers, including Interleukin 6 (IL-6), Procalcitonin, and C-reactive protein, and the highest hospital mortality rate of 55.2%. Conversely, the "Low PDW Stable Group" included the youngest patients, with the lowest inflammatory marker levels and a 16.6% mortality rate. Comparable trajectory groups, patient characteristics, and outcomes were observed in the validation cohorts.

Conclusions: Based on PDW trajectories, we identified and validated four distinct sepsis subphenotypes, each characterized by significant variations in inflammatory marker levels and clinical outcomes.