The Intersection Of Chronic Hepatitis C Infection And Cardiovascular Disease

Hepatitis C virus (HCV) infection is highly prevalent in the US. Though its primary sequelae are liver-related, extrahepatic manifestations contribute to the overall morbidity and mortality of infection. Disorders of lipid metabolism, chronic inflammation and immune dysregulation resulting from chronic infection provide a milieu for extrahepatic manifestations. We sought to examine the role of modification of lipid metabolism on HCV viral load, and to determine the relative contribution of chronic HCV infection to cardiovascular disease. We first examined the effect of 3-hydroxy-3-methylglutaryl-CoA reductase inhibitors (statins) on HCV viral load. We found that, on average, treatment with at least 30 days of a statin was associated with a lower HCV viral load than that observed in those unexposed to statins. Additionally, while long-term follow up was not available, statin therapy was not associated with an increased incidence of liver injury. In the second study, we analyzed data from The Health Improvement Network (THIN) to determine if chronic HCV infection was independently associated with incident myocardial infarction (MI). We found no association between chronic HCV infection and incident MI after adjustment for demographics, comorbidities, medication exposures, body mass index (BMI), tobacco use and family history of MI. Additionally, use of a composite cardiovascular endpoint, characterization of medication exposures as time-varying, and accounting for receipt of HCV therapy did not change our findings. In the conduct of the second study, missing data for BMI were imputed using multiple imputation models. For the third study, we examined whether different variable selection approaches for specification of multiple imputation models result in more or less accurate prediction of investigator-simulated missingness for BMI in THIN. Variable selection procedures utilized to predict missingness included insertion of investigator-chosen variables, use of a high-dimensional approach including all administrative data if a statistical threshold was met, and feature selection driven by machine learning algorithms. We found that the high-dimensional and machine learning approaches, while able to incorporate all data elements, resulted in small improvements in bias but were computationally onerous. The small gains in accuracy achieved with the new methods need to be weighed against the costs of implementation