Abstract

To account for missingness in the outcome variable and potential measurement error in covariates, a Simulation Extrapolation Inverse Probability Weighted Generalized Estimating Equations (SWGEE) model that incorporates missing and measurement errors was used to model the data. The outcome variable of this study was viral load, measured at scheduled follow-up visits of patients. Baseline CD4 count, baseline weight, age at the start of antiretroviral therapy (ART), and gender were the non-dynamic covariates measured at the ART initiation. In contrast, adherence to ART and weight were the dynamic covariates measured at follow-up visits. The results showed that logarithm of viral load got reduced as time in ART increased, with high baseline weight, high weight at follow up, in WHO stage III, and with CD4 count of at least 200. The log-viral load was high in male patients, those who adhered to ART, older patients, and WHO stages II and IV. Chances of a patient missing next follow-up were not related with previous viral load, their adherence status, weight at follow-up, and baseline age. Hence, the missing values of viral load were missing completely at random (MCAR). It is evident that both missingness and measurement error should consistently be accounted for to avoid any inferential bias.

Original language	en
Pages (from-to)	407-416
Publication status	Published - 2025