Clustering identifies endotypes of traumatic brain injury in an intensive care cohort- a CENTER-TBI study
Accepted version
Peer-reviewed
Repository URI
Repository DOI
Change log
Authors
Abstract
Background 26 Whilst the Glasgow coma scale (GCS) is one of the strongest outcome predictors, the current 27 classification of traumatic brain injury (TBI) as ‘mild’, ‘moderate’ or ‘severe’ based on this fails 28 to capture enormous heterogeneity in pathophysiology and treatment response. We 29 hypothesized that data-driven characterization of TBI could identify distinct endotypes and give 30 mechanistic insights. 31 32 Methods 33 We developed an unsupervised statistical clustering model based on a mixture of probabilistic 34 graphs for presentation (<24 hours) demographic, clinical, physiological, laboratory and 35 imaging data to identify subgroups of TBI patients admitted to the intensive care unit in the 36 CENTER-TBI dataset (N=1,728). A cluster similarity index was used for robust determination 37 of optimal cluster number. Mutual information was used to quantify feature importance and for 38 cluster interpretation. 39 40 Results 41 Six stable endotypes were identified with distinct GCS and composite systemic metabolic 42 stress profiles, distinguished by GCS, blood lactate, oxygen saturation, serum creatinine, 43 glucose, base excess, pH, arterial partial pressure of carbon dioxide, and body temperature. 44 Notably, a cluster with ‘moderate’ TBI (by traditional classification) and deranged metabolic 45 profile, had a worse outcome than a cluster with ‘severe’ GCS and a normal metabolic profile. 46 Addition of cluster labels significantly improved the prognostic precision of the IMPACT 2 47 (International Mission for Prognosis and Analysis of Clinical trials in TBI) extended model, for 48 prediction of both unfavourable outcome and mortality (both p<0.001). 49 50 Conclusions 51 Six stable and clinically distinct TBI endotypes were identified by probabilistic unsupervised 52 clustering. In addition to presenting neurology, a profile of biochemical derangement was found 53 to be an important distinguishing feature that was both biologically plausible and associated 54 with outcome. Our work motivates refining current TBI classifications with factors describing 55 metabolic stress. Such data-driven clusters suggest TBI endotypes that merit investigation of 56 identify bespoke treatment strategies to improve care.