Using a Machine Learning Approach to Predict Snakebite Envenoming Outcomes Among Patients Attending the Snakebite Treatment and Research Hospital in Kaltungo, Northeastern Nigeria.

Abstract

The Snakebite Treatment and Research Hospital (SBTRH) is a leading centre for snakebite envenoming care and research in sub-Saharan Africa, treating over 2500 snakebite patients annually. Despite routine data collection, routine analyses are seldom conducted to identify trends or guide clinical practices. This study retrospectively analyzes 1022 snakebite cases at SBTRH from January to June 2024. Most patients were adults (62%) and were predominantly male (72%). Key factors such as age, sex, and time between bite and hospital presentation were associated with outcomes, including recovery, amputation, debridement, and death. Adult males who took more than four hours to arrive to hospital were identified as a high-risk group for poor outcomes. Using patient characteristics, an XGBoost model was developed and was compared to Random Forest and logistic regression models. In general, all models had high positive predictive value and low sensitivity, meaning that if they predicted a patient to experience amputation, debridement, or death, that patient almost always actually experienced amputation, debridement, or death; however, most models rarely made this prediction. The XGBoost model with all features was optimal, given that it had both a high positive predictive value and relatively high sensitivity. This may be of significance to resource-limited settings like SBTRH, where antivenoms can be scarce; however, more research is needed to build better predictive models. These findings underscore the need for targeted interventions for high-risk groups, and further research and integration of machine-learning-driven decision support tools in low-resource-limited clinical settings.