Related Experiment Videos
Predicting Outcomes of Traumatic Brain Injury Using Machine Learning Models Among Patients at Kilimanjaro Christian
William Nkenguye1,2,3, Edwin Joseph Shewiyo1,2,3, João Vitor Perez De Souza3,4
1Department of Epidemiology and Biostatistics, School of Public Health KCMC University Moshi Tanzania.
Health Science Reports
|August 5, 2026
Summary
Machine learning models accurately predict traumatic brain injury (TBI) outcomes in low-resource settings. Random Forest and Decision Tree models show strong performance, aiding clinical decisions and resource allocation in neurocritical care.
Area of Science:
- Neuroscience
- Medical Informatics
- Public Health
Background:
- Traumatic brain injury (TBI) presents a significant global health challenge, particularly in low- and middle-income countries (LMICs) with limited neurocritical care access.
- Accurate prognostic models are essential for timely clinical decisions and resource optimization in TBI management.
- This study focuses on developing and evaluating machine learning (ML) models for predicting TBI outcomes in a resource-limited setting.
Purpose of the Study:
- To develop and validate machine learning models for predicting outcomes in adult TBI patients.
- To assess the performance of various supervised ML algorithms using trauma registry data.
- To identify key predictors of TBI outcomes in a LMIC context.
Main Methods:
- A retrospective cohort study of 4596 adult TBI patients from Kilimanjaro Christian Medical Centre (KCMC) trauma registry (2013-2024).
- Supervised ML algorithms (Random Forest, Decision Tree, Logistic Regression, SVM, ANN) were trained on 70% of data after imputation and SMOTE.
- Model performance was evaluated on a 30% test set using AUC, accuracy, sensitivity, specificity, and predictive values.
Main Results:
- 26.2% of patients experienced poor outcomes (Glasgow Outcome Scale 1-3).
- Random Forest (RF) and Decision Tree (DT) models achieved the highest predictive performance (AUCs of 0.83 and 0.82, respectively).
- Key predictors of poor outcomes included TBI severity, pupil non-reactivity, low oxygen saturation, lack of CT scan, alcohol use, and abnormal vital signs.
Conclusions:
- Machine learning models, especially RF and DT, demonstrate robust predictive capabilities for TBI outcomes using routinely collected data in LMICs.
- These interpretable models, relying on admission data, can facilitate real-time triage and risk stratification.
- External validation and integration into clinical decision-support systems are recommended for wider implementation in similar settings.
