A Copula Approach To Sample Selection Modeling Of Treatment Adherence And Viral Suppression Among HIV Patients On Antiretroviral Therapy (ART) In Namibia

ABSTRACT

Namibia has a generalized human immunodeficiency virus (HIV) epidemic, with HIV mainly transmitted through heterosexual transmission. Although the number of people receiving ART has increased, the achievement of the 90-90-90 strategy on testing, treatment and suppression has not been evaluated. Moreover, examining factors associated with treatment adherence and viral suppression will assist in designing appropriate accelerated interventions. However, modelling treatment adherence and viral suppression may result in biased estimates if sample selection is ignored. The study fit a joint distributional model of ART treatment adherence and viral suppression, to adjust for sample selection bias among HIV patients on ART in Namibia, to examine the presence of tail dependence in sample selection bias, and investigate the factors associated with viral suppression, viral load and ART adherence. The study used two datasets; HIV data of patients, aged above 16 years, on antiretroviral therapy in Erongo region and the selected health facilities in Windhoek, Namibia. A Heckman-type selection analysis using copula were used on the two models: (i) ART adherence with viral suppression margins and (ii) ART adherence with viral load. The families of copulas i.e. Normal, Frank, FGM, AMH, Student-t and the 0, 90, 180 and 270 degrees rotated versions of Joe, Gumbel and Clayton, to capture dependence in the outcomes, were modelled and selected based on the lowest AIC and BIC. The results showed a strong negative correlation between adherence to ART treatment and viral load suppression. The results also showed the dependence structure between ART adherence and viral load margins. The results further showed that Frank and the 180 degrees rotated versions of Gumbel, and Clayton copulas were the best models.