Developing New Prognostic Models for Predicting Outcomes in Severe Traumatic Brain Injury

Abstract

Globally, the annual incidence rate of traumatic brain injury (TBI) in all ages is 349 per 100,000 person-years. The incidence of TBI varies across regions, populations, regulations, and health systems; but in general, the rate is expected to be higher in low and middle-income countries (LMICs). As LMICs usually have poor pre-hospital care, delays in patient transfer, lack of facilities and well-trained staff; these make the burden of TBI more devastating and a pressing public health issue. In this dissertation, we focused on several approaches to determine the value of prognostic research in adult patients with severe TBI, test model feasibility and develop new prediction models for TBI population in LMICs, specifically in terms of predicting mortality and functional outcome. In our narrative review, a search was conducted using PubMed database using the terms “prognostic model” combined with “traumatic brain injury.” The search was limited to articles that included human subjects that were published during 2006 - 2019. We found 20 studies that met the stated search criteria. The results of our review also point out that the majority of current models (85%) were developed from participants in high-income countries (HICs). Additionally, through use of the Reporting Recommendations for Tumor Marker Prognostic Studies (REMARK) checklist as a tool to assess the quality of reporting prognostic research. We found published articles fell short of describing the flow of patients through the study, explaining the type of biological materials used and their handling methods as well as presenting estimated effects of key variables with confidence intervals. These findings reveal the gaps in knowledge and also provide recommendations to refine future TBI prognostic research. We, then, examined the applicability and external validity of the International Mission for Prognosis and Clinical Trials in Traumatic Brain Injury (IMPACT) and Corticosteroid Randomization after Significant Head Injury (CRASH) models for outcome prediction following TBI using a South-American cohort. A total of 550 patients with severe TBI were enrolled in the Benchmark Evidence from South American Trials: Treatment of Intracranial Pressure (BEST-TRIP) study and a simultaneously conducted observational study. Although we found the overall performance of all IMPACT and CRASH models was adequate when applied in this dataset, the IMPACT lab model is preferable as it presents a higher discriminative capacity than the CRASH models. Our findings showed the discriminative performance is still reasonable in the dataset of the contemporary clinical studies indicate the IMPACT and CRASH models continuing relevance. Next, we develop a new prognostic model for severe TBI based on patient admission characteristics and compare this model to the performance of the set of variables in the IMPACT and CRASH core models. We found patient pre-injury comorbidity, anisocoria, glucose level, and motor response post-resuscitation were independent predictors of in-hospital mortality. For functional outcome measured by GOS, respiratory rate at hospital admission, hematocrit, and Glasgow Coma Scale (GCS) score post-resuscitation were independent predictors. Overall, the Thai model for predicting in-hospital mortality had the best performance. The Thailand study reveals the prognostic value of readily available, commonly collected variables like patient vital signs, hematocrit, and glucose level. However, since the data used in the present analysis came from a single site, future external validation should be conducted to confirm model generalizability before applying them more broadly in patients with severe TBI in LMICs.