Prediction of Antiretroviral Therapy Treatment Failure in Malawi using a Soft Computing Modelling approach
Dumisani Ndhlovu, Kondwani Munthali · 2024
As part of routine care in HIV programming, healthcare providers routinely assess patients for potential ART treatment failure. But, due to a shortage of experts and an increasing number of cases, manually analyzing patients for treatment failure is not practical. This study aimed to develop an efficient and effective model using Soft Computing Modeling techniques to predict ART treatment failure. An artificial neural network binary classifier model was developed to predict treatment failure for both first- and second-line regimens. The research employed a qualitative and quantitative method design. Ethnographic methods were utilized to address qualitative objectives, which involved identifying predictors of ART treatment failure and existing algorithms for determining treatment failure. The methodology adhered to the CRISP-DM framework and a backpropagation ANN model was constructed using Python, Scikit-learn library, Keras, and Tensorflow backend. The evaluation outcomes of the ANN model demonstrated an accuracy of 99.71%, indicating that the ANN model can effectively predict ART treatment failure outcomes. The findings in this research suggest that modeling treatment failure prediction using soft computing is a viable technique, and the research recommends the model for incorporation into the treatment failure review process.