Journal de santé et d'informatique médicale

A Predictive Model for Demand for First-Line Antiretroviral (ARV) Drugs Using Data Mining Techniques

Abstract

Asasira Obed

One of the most important indices of defining general welfare and quality-of-life of people in the world is physical
and mental health of individuals.
Health care managers and planners therefore must make future demand for healthcare services and the need for
medicines to achieve fully and reliable supply. With the introduction of the test and treat methodology of managing HIV
patients, first line Antiretroviral drugs (ARVs) must be in adequate availability to enable facilities implement this strategy
of HIV eradication. Discontinuation of antiretroviral therapy Antiretroviral drugs (ART) due to shortages may result into
viral rebound, immune decomposition, and clinical progression of the virus, therefore there is need to plan ahead of
time to avail the most required stock for ARV drugs.
There are no proper forecasting and anticipation mechanisms of future demand for first line Antiretroviral drugs
(ARV) and this is a cross cutting problem for all the public health facilities in Mbarara District and this has led to
overstocking and understocking of these drugs leading to shortages and wastage related to expiry
This study aimed at designing a predictive model for demand of first-line ARV drugs in Mbarara district, using
data mining techniques. Using the Cross Industry Standard Process for Data Mining (CRISP-DM) methodology, the
objectives of this study were to extract and prepare dataset required for data mining, examine different methods used
in demand prediction, Design a model for predicting demand and evaluate this model.
The model was trained under the Waikato Environment for Knowledge Management (WEKA) which is a data mining
environment and predicted the demand for first line ARVs in various health facilities Mbarara district Uganda. The test
results showed that the forecasting in time series approach was more suitable and efficient for drug cycles ahead
demand forecasting. Forecast results demonstrated that the model performed remarkably well with increased number
of actual data and iterations. A regression model gave more accurate forecast results with 7.3% Mean Percentage
Error as compared to alternative methods of demand forecasting whose error was above 30%.