ABSTRACT: The US budget for global health funding, which was by far the largest of similar funding in the world, increased from US $1.3 billion in 2001 to more than US $10 billion in recent years. More than 54% of this funding was allocated to the Global Fund to Fight HIV/AIDS through the US President's Emergency Plan for AIDS Relief (PEPFAR) in Africa. However, recent studies indicate contradictory results regarding the effectiveness of PEPFAR. One by Bendavid, Holmes, Bhattacharya, and Miller shows positive effects of PEPFAR in reducing adult mortality in Africa, while another by Duber, Coates, Szekeras, Kaji, and Lewis finds that there are no significant differences in reducing adult mortality in countries that received PEPFAR funding vs countries that did not. Due to their potential impact on policy decisions regarding critical global health funding, we wanted to assess why the results are discrepant. To do this, we replicated the Bendavid study. The replication provides verification that the study replicable and that the analytic choices of the authors are robust to different assumptions or restrictions. This allows us to assess the different choices and data available to the two research groups and draw some conclusions about why the results may be different. Then, focusing on two of the prominently discrepant studies, i.e., the Bendavid study (1998-2008) and the Duber study (2000-2006), we establish why the two studies are in disagreement. We apply appropriate individual-level and country-level analytical methodology as used by Bendavid over the analytical time period used for the Duber study (2000-2006), which originally focused on nationally aggregated data and differed in some key focus countries. For our first objective, we replicated the original Bendavid study findings and our findings support their conclusion that between 1998-2008 all-cause mortality decreased significantly more (OR = 0.84, CI, 0.72-0.99) in countries that implemented PEPFAR. For our second objective (Bendavid's data and methodology applied to Duber's study period), we found reduction in all cause adult mortality to be borderline insignificant (OR = 0.87 CI, 0.75-1.01, p = 0.06), most possibly reflecting the abbreviated fewer number of events and sample size over a shorter period. Therefore, our overall analyses are consistent with the conclusion of positive impact of the PEPFAR program in reducing adult mortality. We believe that the discrepancy observed in the original studies mainly a reflection of shortcomings in the analytical approach necessitated by the Duber study's nationally aggregated dataset or "may reflect a lack of data quality" in the Duber study (Duber, et al. 2010).
