Statistical analysis of outliers and leverage points in CD4 Counts

The study investigated the CD4 outliers for HIV patients of the Tshepang Clinic of Dr George Mukhari Academic Hospital in Ga-Rankuwa, Tshwane. The investigation of outliers was done relative to the guideline cut-off CD4 values for starting antiretroviral treatment, which show the level of severity of the epidemic.The CD4 counts and viral loads were collected from Tshepang clinic files for the period January 2006 to December 2010. Conventional statistical methods and robust methods were used to analyse the data using statistical packages STATA and PASW version 18. Results of robust methods were compared with conventional methods descriptive and inferential statistical methods that are known to be sensitive to outliers.The influence of outliers was confirmed about the data. The values of robust measures obtained were least absolute deviation of 21735.6, M-estimator of 821.8 and least trimmed squares of 2375735.9. The study recommends that the South African Department of Health should revise the starting level of using ARVs to 300 cells/mm3, and that the outlier or cut-off value in CD4 counts be changed to 300 cells/mm3.