Description
- Abstract:
- Background: Sepsis is the leading cause of child death globally, with low- and middle-income countries (LMICs) bearing a disproportionate burden of deaths. LMICs often have limited prognostic tools and critical care capacity, leading to delayed recognition of advanced sepsis. Traditional sepsis scoring systems, like pSOFA and Phoenix Criteria, require extensive lab tests and resources not always available. This study aimed to develop and evaluate machine learning (ML) models using readily accessible clinical data throughout Bangladesh to predict mortality in pediatric sepsis following the initial 24 hours post-suspicion of sepsis. Methods: A prospective observational study was conducted on children with suspected sepsis admitted to the International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b) from February to December 2022. Seven ML models—Decision Tree, Random Forest, Support Vector Machine (SVM), Kernel SVM, Naïve Bayes, K Nearest Neighbors (KNN), and Logistic Regression (LR)—were trained. Features of interest included a complete blood count, lactate, age, weight, vital signs, SpO2:FiO2, and Glasgow Coma Scale, while an alternative feature set excluded lab tests. Model performance was assessed using the area under the receiver operating characteristic curve (AUROC). Feature selection and dimensionality reduction techniques improved model interpretability and reduced overfitting. Results: Within our sample of 96 patients, LR demonstrated high predictive performance with AUROC of 0.918 (CI: 0.880-0.953) for clinical variables and 0.915 (CI: 0.865-0.962) for clinical variables without labs. The Phoenix and pSOFA criteria were broken into individual sub-scores as variables, achieving an AUROC of 0.894 (CI: 0.849-0.938) and 0.925 (CI: 0.895-0.954), respectively. Conclusion: Logistic Regression was shown to be the most effective model for predicting mortality in pediatric sepsis using clinical variables available in LMICs. Future research should concentrate on refining these models to improve their applicability and integration into clinical practice, potentially through digital tools. By utilizing ML models, we can enhance the timely recognition and management of pediatric sepsis, ultimately leading to a reduction in mortality in LMICs.
- Notes:
- Scholarly concentration: Medical Education
Access Conditions
- Use and Reproduction
- CC0 1.0 Universal (CC0 1.0)
- Rights
- No Copyright - United States
- Restrictions on Use
- No Rights Reserved
Citation
Bunch, Kaden, Kadakia, Nidhi, Shaima, Shamsun Nahar, et al.,
"Leveraging Machine Learning to Predict Pediatric Sepsis Mortality in Bangladesh"
(2024).
Warren Alpert Medical School Academic Symposium.
Brown Digital Repository. Brown University Library.
https://doi.org/10.26300/nfdv-5q98
Relations
Collection:
-
Warren Alpert Medical School Academic Symposium
The Warren Alpert Medical School Academic Symposium is an annual event at Warren Alpert Medical School of Brown University that provides Year II medical students a venue to present their summer research in a poster format. Participation in the Symposium …...