Longitudinal biomarker trajectories and their prognostic utility for 21-day mortality in burn patients with sepsis: a retrospective cohort study.

OBJECTIVE: To characterize the longitudinal trajectories of multi-category biomarkers and evaluate their association with 21-day all-cause mortality in critically ill burn patients with sepsis. METHODS: In this retrospective single-center cohort study, we analyzed 943 adult burn patients with sepsis, defined per Sepsis-3.0 criteria. Serial measurements of 15 biomarkers across nutritional, immunoglobulin, lymphocyte subset, inflammatory, and other categories were collected over 21 days. We employed linear mixed-effects models (LME) to compare trajectories between survivors and non-survivors, Cox regression to assess associations with mortality, time-dependent ROC to evaluate predictive performance, and k-means clustering to identify patient phenotypes based on integrated ALB, IL-6, and IgG trajectories. RESULTS: The 21-day mortality was 17.92%. LME revealed significantly different trajectories for 11 biomarkers between survivors and non-survivors (P < 0.05). Univariate Cox analysis identified multiple significant biomarkers, with transferrin (HR = 0.985, P = 6.84 × 10⁻11) and IgM (HR = 0.284, P = 1.24 × 10⁻5) as strong protective factors, and mitochondrial DNA (HR = 1.002, P = 1.89 × 10⁻⁹) as a risk factor. In multivariate analysis, only the Burn Index remained an independent risk factor (HR = 1.066, P < 0.001). Time-dependent ROC showed peak predictive accuracy at Day 7 (albumin AUC = 0.729). Clustering identified three distinct phenotypes-"Rapid Recovery" (mortality 5.2%), "Persistent Inflammatory & Catabolic" (mortality 38.0%), and "Intermediate" (mortality 18.7%; P < 0.001)-with starkly different biomarker trends and clinical profiles. CONCLUSIONS: The dynamic patterns of multi-category biomarkers are strongly associated with short-term survival in burn sepsis. While burn severity is a dominant baseline risk factor, longitudinal trajectory analysis captures the essence of the host's recovery or failure, effectively stratifying patients into prognostically distinct subgroups. This trajectory-based phenotyping highlights the potential of monitoring the host response over time to improve risk assessment and guide personalized management.