Date of Award


Degree Type


Degree Name

Doctor of Philosophy (PhD)




Hepatitis C (HCV) is a virus transmitted via contact with contaminated products such as razor-blades or by engaging in high-risk activities (i.e., such as sexual or injecting drug-use activities). Today, there are an estimated 115 million people world-wide living with HCV. Despite recent advancements in antiviral treatments that can ameliorate (or even cure) HCV, treatment remains laborious and costly and is often unavailable in resource-poor areas such as Sub-Saharan Africa (SSA). The primary aim of this study was to estimate the prevalence of HCV in SSA.

A meta-analysis was conducted on the HCV epidemic in SSA. A literature search for evidence of HCV in SSA was conducted and was limited to articles, abstracts, and conference proceedings published in English, Spanish, or French from January 2000 through December 2013. Linear and generalized-linear mixed effects models were used to estimate the pooled prevalence of HCV in SSA as a function of the population at-risk, region of SSA, year of publication, and the assay used to detect viremia. In these models, the estimates were weighted by their inverse variance and pooled separately using no transformation, a canonical logit transformation, and double-arcsine transformation. These three transformation approaches were compared on precision, model fit, and publication bias.

The overall pooled prevalence estimate of HCV in SSA ranged from 3.80% to 5.83% depending on the transformation used. For all three methods, however, prevalence of HCV varied among those at-risk for infection (p < .001) and by region of SSA (p < .001). In fact, the prevalence of HCV among those at-risk for infection depended on region of SSA (p < .001). Conversely, this study was unable to show that prevalence depends on publication year (p > .05) or diagnostic assay (p > .05) under all three transformation methods. Regarding the optimal transformation, prevalence of HCV in SSA tended to be lowest when estimated under a canonical logit transformation and highest when estimated using no transformation of the raw effect sizes. Regarding precision and model fit, confidence intervals for all prevalence estimates were severely overlapping under the three transformation methods, yet normality, linearity, and residual plots consistently revealed that the canonical logit approach was superior when compared to the double-arcsine transformation and raw estimation method.

When estimating the pooled prevalence of HCV in SSA, this study did not identify meaningful differences between the logit and double-arcsine transformations. That is, they were generally comparable in precision and had severely overlapping confidence intervals for all moderator analyses. However, model fit statistics suggest that the canonical logit approach provided a better fit to the data than the double-arcsine transformation or raw estimation method. I caution future researchers considering no transformation of the raw prevalence estimates. When no transformation was used, the pooled prevalence estimate of HCV in SSA was inflated as measured by standardized residuals, individual study variances were severely attenuated, publication bias estimates were quite severe and, in some instances, the study predicted prevalence estimates well below zero. For this reason, I recommend using the canonical logit transformation for meta-analyses of HCV in Sub-Saharan Africa.

Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License.

Included in

Epidemiology Commons