Occurring widely with an estimated incidence of 69 million each year globally, traumatic brain injury (TBI) brings huge burdens to families of patients and to social economy (1). The mortality of TBI patients is high, especially severe TBI, with the mortality rate ranging from 23.0%–38.8% (2, 3). Accurate and convenient risk stratification of admitted TBI patients in the early phase is helpful for clinicians in making personalized treatments. The conventional tool of risk stratification for TBI is the Glasgow Coma Scale (GCS), which is evaluated based on clinical symptoms. However, single GCS could not comprehensively reflect the severity and progression of TBI patients due to the complex pathophysiological process of TBI and would be influenced by intubation and status of sedation. Identifying TBI phenotypes with different risks of clinical outcomes using laboratory biomarkers may make up for the insufficiency of GCS and guide clinicians in making specific treatment options based on pathophysiological processes.

Previous studies have explored phenotypes of some critically ill patients with acute kidney injury, sepsis, or acute respiratory distress syndrome using clustering methods (4–7). Composed of neutrophils, lymphocytes, monocytes, basophils, and eosinophils, leukocytes are recruited to the injured brain tissue and activated to participate in neuroinflammation after TBI (8–13). Previous studies have used the value of a single type of leukocytes or the neutrophil-to-lymphocyte ratio to reflect the inflammatory status and predict the prognosis of TBI (14–19). There is no study yet using blood values of all types of leukocytes to evaluate the prognosis of TBI. We designed this study to identify TBI phenotypes based on blood values of leukocytes using the K-means clustering method.

We found that the three TBI clusters had different inflammatory statuses through the clustering of peripheral leukocytes. Cluster 1 with a higher inflammatory status had a worse prognosis than cluster 2 with a lower inflammatory status. TBI could induce an increase in leukocytes soon after an initial injury. Neutrophils would be released from the bone marrow by stimulating cortisol and catecholamines after TBI (20). The activated microglia after TBI would then activate the endothelial cell and recruit the peripheral neutrophil to infiltrate the injured area by releasing inflammatory chemokines and cytokines (8, 9). Then, activated T cells and monocytes/macrophages would be recruited to the injured brain area and participate in the response of the adaptive immune system for the injured brain tissue (10–13). These cells would interactively activate and promote a secondary brain injury by aggravating neuroinflammation and immune response (21).

The relatively lower level of lymphocytes in the cluster and the higher level of lymphocytes in cluster 2 indicated that the peripheral immunosuppressive status might be related with a higher inflammatory status and worse outcome. Actually, previous studies found that TBI could cause the acute increase of cortisol levels both in serum and cerebrospinal fluid (22, 23). The increased plasma cortisol in the early phase after TBI would prevent the lymphocyte egress from the secondary lymphoid tissues, which may be an endogenous protective response to inhibit the excessive infiltration of T cells in the injured brain area, subsequently amplifying neuroinflammation (24). However, the peripheral immunosuppressive status is certainly associated with a higher risk of infection among extracranial organs.

Many previous studies used the value of a single type of leukocytes or neutrophil-to-lymphocyte ratio to evaluate the risk of poor prognosis of TBI (14–19), while not all types of leukocytes were included and analyzed to evaluate the prognosis risk, which was a limitation of these studies. We performed this study to identify inflammatory clusters of TBI patients with different mortality risks by comprehensively clustering multiple types of leukocytes, namely, neutrophils, lymphocytes, monocytes, basophils, and eosinophils. Three inflammatory clusters of TBI were identified and were consistent with their GCS. This classification is beneficial for physicians to stratify risks and make personalized treatment schedules. It may also be helpful to identify TBI subgroups benefiting from specific treatments targeted at inflammation.

Several limitations were unavoidable in this study. Firstly, a number of patients were excluded from this study mainly due to lack of records on leukocytes on the first day, causing a selection bias. The classification we identified should be further verified in other medical centers with more generalized TBI patients and a prospective design collecting levels of leukocytes at a fixed time period, as early as possible, after admission. Secondly, specific subgroups of lymphocytes were not analyzed including NK cells, B cells, cytotoxic T cells, and regulatory T cells due to lack of records from the MIMIC-III database. Thirdly, common inflammation markers such as C-reactive protein, interleukin-1, interlukin-6, TNF-α, and interferon-γ were not analyzed among the three clusters due to lack of records from MIMIC-III. Future studies could be performed to measure these markers among our identified clusters and evaluate the consistency between these markers and clusters with different inflammatory statuses. Fourthly, functional outcomes and recovery statuses were not recorded in MIMIC-III, so we were unable to analyze the relationship between the discovered three clusters and these outcomes. Future studies could be designed to verify the importance of the three clusters on functional outcomes and recovery statuses. Finally, agents with immunomodulatory effects or cytotoxicity were not particularly extracted and analyzed, which may limit the reliability of our classifications.

Three TBI clusters with different inflammatory statuses and prognoses were identified based on levels of blood leukocytes. This classification is beneficial for physicians in evaluating the prognosis and making personalized treatments for TBI patients.