Staton, Catherine LynchHuo, Lily2023-06-082023https://hdl.handle.net/10161/27816<p>Background: Traumatic brain injury (TBI) has a disproportionate burden on low- and middle-income countries (LMICs) and cost-effective and culturally relevant measures are necessary to improve TBI care. This study aims to characterize emergency healthcare providers’ decision making when treating TBI patients, develop a machine learning-based model to predict TBI patient outcome, and conduct a decision curve analysis (DCA) to evaluate model clinical applicability. Methods: This study is twofold: 1) a secondary analysis of a TBI data registry with 4142 patients and 2) a survey examining physicians decision-making in treating 50 TBI patients in real time. Results: Five machine learning models were developed with AUCs ranging from 70.86% (Single C5.0 Ruleset) to 85.67% (Ensemble Model). DCA showed that all models exhibited a greater net benefit over ranges of clinical thresholds. The survey collected information on 50 patients providing insight on tools used by physicians in real-time when treating TBI patients as well as the unmet need patients at KCMC faced. Conclusions: This study is the first to use machine learning modeling and DCA in the context of TBI prognosis in Sub-Saharan Africa. Prognostic models have great potential within the decision-making process for treating TBI patients in LMIC health systems and such utility can be expanded through determining different threshold probabilities for various interventions.</p>NeurosciencesEpidemiologyPublic healthDecision curve analysisLMICsMachine learningNeurosurgeryPrognostic modelTraumatic brain injuryMaster's thesis