A meta-analysis by Van Howe of 109 populations confirms the well-known association of male circumcision (MC) with reduced HIV prevalence. He then performed meta-regression adjusting for location, risk and MC prevalence. When one or two of these adjustments in combination were applied MC appeared protective, but when all three were introduced the association remained significant in high-risk populations, but not in general populations within Africa with a hypothetical MC prevalence of <25% or elsewhere with hypothetical MC prevalence of <75%. However, many MC prevalence values given differed from those reported in references cited (including all US studies). This and other problems invalidate his adjustments for MC prevalence, undermining most of his meta-regression results. Meta-regression is a highly sophisticated statistical tool and is prone to error if not applied correctly. The study contained a high risk of bias arising from confounding. We also question his use of crude, rather than adjusted, odds ratios and his inclusion of unpublished data, so precluding replication by others. Flawed statistics, opaque presentation of results and inclusion of previously repudiated arguments downplaying a role for MC in HIV prevention programmes should lead readers to be sceptical of the findings and conclusions of Van Howe's study.
Keywords: Circumcision; HIV; male; meta-analysis; policy; prevalence; prevention.