Development of a predictive statistical system for proactive monitoring of Sepsis among inpatients in Uganda

Abstract

This study aims at developing and evaluating a predictive statistical model for early detection of sepsis among inpatients in Ugandan healthcare facilities. Sepsis is a life-threatening condition characterized by the body's inflammatory response to an infection, and its timely detection and management are crucial for improving patient outcomes. The study utilized a dataset from Kaggle donated by The Johns Hopkins University to train, test, and validate the model. The model was developed using a combination of statistical methods like logistic regression and machine learning algorithms such as Naive Bayes, Random Forest, Support Vector Machines (SVM), and Extreme Gradient Boosting (XGBoost). The model's performance was evaluated using metrics like area under the receiver operating characteristic curve (AUROC), sensitivity, and specificity. The study found that the developed SVM model performed well in predicting sepsis onset in a Ugandan inpatient population. The most significant predictors of sepsis identified were Blood Work Result-1, Body Mass Index, Blood Work Result-4, Age and Plasma Glucose. The model's performance was compared to existing methods for sepsis prediction in Uganda, demonstrating its superiority in detecting sepsis. This study contributes to the knowledge on sepsis management and healthcare delivery in low-resource settings like Uganda. The predictive model developed can be integrated into clinical decision support systems to aid in early detection and treatment of sepsis, ultimately improving patient outcomes. The findings of this study can inform future research and development efforts, guiding the optimization of predictive models for broader application in resource-limited settings.