Search

Binary logistic regression as an instrument for the prediction of immunological deterioration from nutritional indicators in persons infected with HIV-AIDS. In the evolution of infection with HIV diverse markers of progression have been used, of immunological, virological and genetic type. However, the significance of nutritional indicators for prognosis has been insufficiently studied. The objective of this investigation was to use binary logistic regression as a tool to find a mathematical model which would allow to predict the counting of CD4+ cells in persons infected with HIV-AIDS after a year's follow-up. 118 persons with HIV-AIDS were studied (initial time and after a year of evolution) which were classified in four subgroups according to the use or not use of antirretrovirals and the counting of CD4 at the initial time. Nutritional, biochemical and anthropometric variables were used, as well as the gradients between both observations. Binary logistic regression was applied, having as dependent variable the counting of CD4 as an immunological criterion of antirretroviral therapy. The logistical models found were adjusted according to the subgroup they belonged to, and they were compared to one another according to the goodness of fit and accuracy. The step-by-step method chose only the gradient of body mass index (IMC) as independent variable. The adjustment of the model allowed to propose a mathematical equation by subgroup in order to predict the need of antirretroviral therapy after a year. The groups which had initial values of CD4 < 350, even with an increase in IMC ran the risk of keeping immunological deterioration after a year. All models showed specificity values higher than those of sensitiveness, though models II and III were the best, as proved by the probabilities shown in ROC curves. The conclusion is that the IMC gradient in a year of evolution might be used in a logistical mathematical model to predict the need of antirretroviral therapy in individuals with HIV-AIDS. [ABSTRACT FROM AUTHOR]

In the paper the results of the combined strategy formed by Robust Principal Component Analysis and CUR matrix decomposition are applied to a practical problem for adjusting a regression model by using Partial Least Squares (PLS). The conclusions concerning their advantages of such an approach as well information about the adjusted models are given. [ABSTRACT FROM AUTHOR]