Multivariate multilevel modeling of quality of life dynamics of HIV infected patients

Abstract

Abstract Background Longitudinal quality of life (QoL) is an important outcome in many chronic illness studies aiming to evaluate the efficiency of care both at the patient and health system level. Although many QoL studies involve multiple correlated hierarchical outcome measures, very few of them use multivariate modeling. In this work, we modeled the long-term dynamics of QoL scores accounting for the correlation between the QoL scores in a multilevel multivariate framework and to compare the effects of covariates across the outcomes. Methods The data is from an ongoing prospective cohort study conducted amongst adult women who were HIV-infected and on the treatment in Kwazulu-Natal, South Africa. Independent and related QoL outcome multivariate multilevel models were presented and compared. Results The analysis showed that related outcome multivariate multilevel models fit better for our data used. Our analyses also revealed that higher educational levels, middle age, stable sex partners and higher weights had a significant effect on better improvements in the rate of change of QoL scores of HIV infected patients. Similarly, patients without TB co-infection, without thrombocytopenia, with lower viral load, with higher CD4 cell count levels, with higher electrolytes component score, with higher red blood cell (RBC) component score and with lower liver abnormality component score, were associated with significantly improved the rate of change of QoL, amongst HIV infected patients. Conclusion It is hoped that the article will help applied researchers to familiarize themselves with the models and including interpretation of results. Furthermore, three issues are highlighted: model building of multivariate multilevel outcomes, how this model can be used to assess multivariate assumptions, involving fixed effects (for example, to examine the size of the covariate effect varying across QoL domain scores) and random effects (for example, to examine the rate of change in one response variable associated to changes in the other).