Longitudinal profile of routine biomarkers for mortality prediction using unsupervised clustering algorithm in severely burned patients: a retrospective cohort study with prospectively collected data

• Published in: Annals of surgical treatment and research Vol. 104; no. 2; pp. 126 - 135
• Main Authors: Yoon, Jaechul, Kym, Dohern, Hur, Jun, Cho, Yong-Suk, Chun, Wook, Yoon, Dogeon
• Format: Journal Article
• Language: English
• Published: Korea (South) 대한외과학회 01-02-2023; The Korean Surgical Society
• Subjects: Original; 일반외과학; Cluster analysis; Biomarkers; Longitudinal studies; Burns; Mortality
• ISSN: 2288-6575; 2288-6796
• DOI: 10.4174/astr.2023.104.2.126

Burn injury has high clinical heterogeneity and worse prognosis in severely burned patients. Clustering algorithms using unsupervised methods to identify groups with similar trajectories in heterogeneous disease patients can provide insight into mechanisms of disease pathogenesis. This study analyzed routinely collected biomarkers to evaluate mortality prediction, find clinical meanings for these or their subtypes, and evaluate patterns.
This retrospective cohort study included patients aged >18 years, between July 2012 and June 2021. All eligible patients received fluid resuscitation and survived for at least 7 days. Characteristics of clinical interest to the physician at 4 clinically important time points were evaluated.
Eligible patients were divided into 4 subgroups according to these time points: from 1st week to 4th week. Total of 1,249 patients admitted within 2 days after burns and receiving fluid resuscitation were included. Mean Harrell's C-index of pH was the highest (0.816), followed by platelets (0.807), creatinine (0.796), red cell distribution width (RDW, 0.778), and lactate (0.759). Longitudinal profiles among biomarkers were different.
The main predictors were pH, platelets, creatinine, RDW, and lactate. Creatinine and RDW showed consistent patterns. The other markers varied according to patient condition. Thus, these markers could provide clues into underlying mechanisms and predict mortality.