Abstract

As part of routine care and treatment activities for people living with HIV and are on antiretroviral therapy (ART) treatment, health care providers conduct reviews on patients for possible ART treatment failure. However, due to the limited number of available experts and increasing number of treatment failure cases, it is not feasible to manually analyze patients for treatment failure. Most health facilities in Malawi use EMR systems to monitor patients’ performance. This research aimed at finding an efficient and effective model to predict ART treatment failure by utilizing the data available in the EMR systems. An Artificial neural network binary classifier model was built to predict ART treatment failure for first- and second-line regimens. We used ethnographic methods to respond to qualitative objectives which were to establish ART treatment predictors and current algorithms that are followed in treatment failure determination. Participation and observations were employed and a total of 17 experts were interviewed. The methodology used, followed the CRISP-DM framework by first understanding the HIV treatment failure domain, the causes and factors associated with treatment failure, and how treatment failure is currently determined. Only correlated variables to the outcomes were considered in building the ANN prediction model. A random sample dataset of 10,000 patients was generated from the EMR system database. Out of these only 1,722 records had sufficient data and were used in the ANN modelling. SMOTE technique was used to balance the distribution of data on the target variable. A backpropagation ANN model was built using Python3, Sci-Kit Learn library, Keras and TensorFlow backend. The ANN model evaluation scored accuracy of 99.71% which shows that ANN model can be used to predict ART treatment failure outcome. The presented results have demonstrated in this research that it can be a viable technique to model treatment failure prediction using soft computing. The research recommends the model for treatment failure review process.

More details

School	: School of Natural and Applied Sciences
Issued Date	: 2024

Download full document