Abstract
BACKGROUND: Leukocytes play an important role in inflammatory response after a traumatic brain injury (TBI). We designed this study to identify TBI phenotypes by clustering blood levels of various leukocytes. METHODS: TBI patients from the Medical Information Mart for Intensive Care-III (MIMIC-III) database were included. Blood levels of neutrophils, lymphocytes, monocytes, basophils, and eosinophils were collected by analyzing the first blood sample within 24 h since admission. Overall, TBI patients were divided into clusters following the K-means clustering method using blood levels of five types of leukocytes. The correlation between identified clusters and mortality was tested by univariate and multivariate logistic regression analyses. The Kaplan-Meier method was used to verify the survival difference between identified TBI clusters. RESULTS: A total of 172 (cluster 1), 791 (cluster 2), and 636 (cluster 3) TBI patients were divided into three clusters with the following percentages, 10.8%, 49.5%, and 39.8%, respectively. Cluster 1 had the lowest Glasgow Coma Scale (GCS) and the highest Injury Severity Score (ISS) while cluster 2 had the highest GCS and the lowest ISS. The mortality rates of the three clusters were 25.6%, 13.3%, and 18.1%, respectively. The multivariate logistic regression indicated that cluster 1 had a higher mortality risk (OR = 2.211, p = 0.003) than cluster 2, while cluster 3 did not show a significantly higher mortality risk than cluster 2 (OR = 1.285, p = 0.163). Kapan-Meier analysis showed that cluster 1 had shorter survival than cluster 2 and cluster 3. CONCLUSION: Three TBI phenotypes with different inflammatory statuses and mortality rates were identified based on blood levels of leukocytes. This classification is helpful for physicians to evaluate the prognosis of TBI patients.