A model to predict the correct stocking levels of artemisinin-based combination therapies in Kira Health Center III and Kisenyi Health Centre IV

Abstract

Introduction The persistent low stock/stock-outs of malaria essential medicines, such as, Artemisinin-based Combination Therapies (ACTs), in public health facilities is a major hinderance to treatment and management of malaria. Therefore, this study aimed to develop a predictive model capable of forecasting the required stock of weight appropriate ACTs in public health facilities in Uganda. Methodology Data for malaria cases was collected from two health facilities; Kira Health Center III and Kisenyi Health Center IV. We then cleaned and grouped the data according to age categories for which ACTs are packed in. Thereafter, we applied ARIMA, PROPHET, and ARIMAX machine learning technique to this dataset to develop forecasting models. To identify the best models, we evaluated them using Akaike Information Criteria. Results ARIMAX model performed better than the other models that were developed. It was able to accurately predict the expected number of cases to expect for the different age categories which then informed the number of ACTS to stock. Conclusions Forecasting using facility health data while considering external factors can be used to accurately forecast number of ACTs a facility needs to stock