Support vector machines to forecast changes in CD4 count of HIV-1 positive patients

There are currently 5.5 million confirmed cases of HIV/AIDS in South Africa. HIV infection can be effectively managed with antiretroviral (ARV) drugs, but close monitoring of the progression of the disease is vital. One of the better surrogate markers for disease progression is the use of CD4 cell counts. Forecasting CD4 cell count will help clinicians with treatment management and resource allocation. The aim of this paper was to investigate the application of machine learning to predict future CD4 count change. A support vector machine classification model that predicted the degree of CD4 count change was built. The model took as input the genome, current viral load and number of weeks from baseline CD4 count and predicted the range of CD4 count change. The model produced an accuracy of 83%. This pilot project shows that a change in CD4 count may be accurately predicted using machine learning on genotype, viral load and time. Clinical studies to validate this are required. The aim of this paper was to investigate the application of machine learning to predict future CD4 count change. A Support Vector Machine classification model that predicted the degree of CD4 count change was built. The model took as input the genome, current viral load, and number of weeks from baseline CD4 count and predicted the range of CD4 count change.