The Impact of Care Delays on Traumatic Brain Injury Outcomes in Tanzania: Descriptive Analytics and Machine Learning

Background: Traumatic brain injury (TBI) is the leading cause of trauma related death and disability worldwide. Poor TBI outcomes disproportionately affect low- and middle-income countries (LMICs). Treatment delays may contribute to poor TBI outcomes in LMIC emergency departments (EDs). A prognostic model is a low-cost, user-friendly solution to optimizing patient care in low-resource hospitals. The aim of this study was twofold: (1) assess associations between care delays and TBI patient outcomes, and (2) build a prognostic model that uses care delays to predict TBI patient outcomes.

Methods: This study uses a 3209 de-identified TBI patient registry from Kilimanjaro Christian Medical Center (KCMC) ED in Moshi, Tanzania. We created nine variables representing delays to care and assessed their association with poor outcomes (Glasgow Coma Score (GCS) < 4) using logistic regression. We then constructed a prognostic model that predicts TBI patient outcomes dichotomized as good (GCS ≥ 4) and poor (GCS < 4). Predictors included socio-demographics, injury characteristics, vital signs, and care delays.

Results: Associations between care delays and TBI outcomes were not significant. However, care delays were top predictors of a poor outcome in our prognostic model. Our model achieved an area under the receiver operating curve of 89.5% (95% CI: 88.8, 90.3).

Conclusion: Our TBI prognostic model demonstrates the predictive value of care delay information. Time to care data is easy to collect. A prognostic model that uses time to care data allows healthcare providers to update patient prognosis as patients progress through their hospital stay.