Methodological Advances to Address Measurement Error and Model Misspecification in HIV Research

Advances in antiretroviral therapy have led to drastic improvements in survival among people living with HIV. However, as life expectancies have increased, so too has the prevalence of chronic co-morbidities. Liver-related disease is now among the leading causes of death for those living with HIV, particularly for those co-infected with the hepatitis C virus (HCV). The incidence of cardiovascular disease has also increased, largely due to metabolic dysfunction. The impact of HCV suppression following HCV treatment on liver and cardiovascular outcomes remains an open question. Further, unbiased estimation of treatment effects is complex due to the potential for measurement error in laboratory assays used to inform treatment decisions and the difficulty in capturing non-linear relationships.In this work, I consider the effect of successful treatment for HCV on two clinical outcomes using marginal structural models (MSMs). Despite their broad application in this area, measurement error and non-linear relationships between treatment and covariates pose important challenges that have not been systematically addressed in the literature. To investigate the clinical questions of interest, I first addressed two methodological objectives: (i) to explore approaches to correct for measurement error in continuous covariates used to estimate inverse probability weights for MSMs; and (ii) to compare and assess flexible models to capture non-linear associations between treatment and covariates when estimating inverse probability weights. I discuss three analyses, each performed on data from the Canadian Co-infection Cohort Study (CCC). First, I examine the effect of successful treatment for HCV on liver fibrosis progression as measured by the aspartate aminotransferase:platelet ratio index. Subsequently, I continue to explore this relationship while addressing potential non-linearity in the relationship between HCV treatment and gamma-glutamyltransferase, a liver enzyme that serves as a surrogate marker of liver function. In the final analysis, I consider the relationship between successful treatment for HCV and change in patient BMI as a potential indicator of overall patient health and recovery.The first manuscript accomplishes my primary methodological objective. I propose a novel application of the simulation-extrapolation procedure to correct covariate measurement error in the exposure model used to estimate weights for MSMs. The results of my simulation studies show that errors in time-varying covariates may induce substantially biased exposure effect estimators in MSMs and that both the direct and indirect approaches are effective at removing bias given low-to-moderate degrees of measurement error.In addition to modelling error-prone covariates, failure to capture important non-linear relationships in the exposure model represents an additional source of residual confounding and bias. In the second manuscript, I explore the effect of model misspecification when the assumption of linearity between exposure and time-varying covariates is not satisfied in the model used for inverse probability weighting. In simulation studies, I demonstrate the bias and poor covariate balance that results from insufficient flexibility in the treatment model, and demonstrate that more flexible models for the inverse probability of treatment weights improved balance and reduced bias.Finally, in my third manuscript, I consider the effect of sustained virological response (SVR) to HCV therapy on change in BMI among patients co-infected with HIV. Using data collected from the CCC, I employ methods developed in the first two manuscripts. Results suggest that patients co-infected with HIV and HCV may experience an increase in BMI immediately following successful therapy for HCV. This finding is clinically relevant to continued post-treatment monitoring within this patient population, and efforts to improve positive effects on overall patient health

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.