Improving the Predictability of the Pediatric Resuscitation and Trauma Outcome (PRESTO) model in injury patients in Tanzania
Abstract
Background: Injuries are responsible for 950,000 deaths per year of children and adolescents under 18 years old. The use of trauma prediction scores has been shown to be helpful in determining the severity and prognosis of injury patients. The pediatric resuscitation and trauma outcome (PRESTO) score was developed as a feasible and simple score of short-term mortality prediction specific to pediatric populations in low and middle-income countries (LMICs). Using relevant variables available at the bedside in resource-limited settings, PRESTO has been validated in three LMICs including South Africa, Rwanda, and Tanzania. In Tanzania, our team found that these variables were present for the majority of pediatric injury patients and the model performed well in predicting mortality, though limited by small sample size.
Objective: We sought to improve the predictability of the PRESTO model for pediatric patients by increasing our sample size and including adult trauma patients from the same population at Kilimanjaro Christian Medical Centre (KCMC) in Moshi, Tanzania.
Design/Methods: Data was collected between November 2020 and February 2024 from a pediatric registry, and between April 2018 and February 2024 from an adult injury registry from KCMC. Missing data was addressed using multiple imputation with chained equations. The data was split into training and testing sets (75/25 ratio). Ten different machine learning algorithms were trained to predict in-hospital mortality. Clinical and demographic predictor variables collected in the Emergency Department were used as predictors (i.e. heart rate, systolic blood pressure, oxygen saturation, respiratory rate, Glasgow Coma Scale [GCS], age, sex), with adjustments for age-vital sign interactions. Model performance was evaluated through 10-fold cross-validation. Hyperparameters were optimized using grid search, and performance was assessed with the ROC-AUC, sensitivity and specificity.
Results: A total of 5,635 injury patients (911 pediatric and 4,724 adult) were included in our analysis. Pediatric in-hospital mortality was 6.8%, while adult mortality was 4.4%. Pediatric patients presented with worse GCS scores compared to adults (4% severe vs 1.4% severe, respectively). The three best-performing models overall were the Random Forest (RF), C5, and Extreme Gradient Boosting (XGB) with ROC-AUCs of 0.92, 0.90, and 0.81, respectively. In the test set, a consistent pattern across the models was observed where the ROC-AUC for pediatric patients was worse than for adult patients. In detail, a difference of 0.08, 0.06, and 0.05 in favor of the adult patients was noted for the XGB, RF, and C5 models, respectively.
Conclusions: Adding adult data did not improve the predictability of the PRESTO model, supporting the idea that children need to be considered separately, especially when it comes to vital signs and predicting mortality. These findings emphasize the importance of increasing the sample size of our pediatric injury registry to continue to collect locally relevant data to develop more accurate models for stratifying mortality risk. Such efforts will ensure that models are tailored to the unique needs of pediatric patients, particularly in LMICs, improving both predictions and resource allocation.
