Purpose: HIV prevention funds are often allocated by decision makers at several different levels. High-level decision makers may allocate funds to regions, while regional decision makers then allocate those funds to specific programs. Thus, high-level decision makers may have no control over how the funds are ultimately spent. We investigate whether top-level decision makers can incorporate knowledge of lower-level decisions into their resource allocation decisions to improve the overall allocation process (in particular, to increase the number of HIV infections averted).

Methods: We develop an optimization model of resource allocation at two levels: an aggregate upper level and multiple local levels. We consider two types of allocation: an optimal allocation that maximizes the number of HIV infections averted, and a proportional allocation that allocates funds proportional to HIV incidence. We illustrate with data from 40 US states and regions.

Results: If resources are allocated optimally at only one level, then the total number of infections averted is always greater than if proportional allocation is used at both levels but always less than if resources are allocated optimally at both levels. In the base case, the use of optimization at the upper level and equity at the lower level results in a 10% gain in the number of HIV infections averted relative to using equity at both levels. Using optimization at both levels results in an additional 17% in the number of HIV infections averted. For relatively small budgets, optimization applied at the higher level yields more gains than at the lower level; for relatively large budgets, optimization at the lower level yields greater gains than optimization at the higher level.

Conclusions: Optimization applied only at the higher level cannot completely overcome poor allocation decisions at the lower level. In general, if optimization can only be applied at one level, then optimization at the lower level yields greater gains than optimization at the higher level. Thus, rules by donor organizations that require very detailed planning by recipients may be justified. Significant benefit could be achieved by investing in resources that allow optimization-based approaches to be used at the lower level. Failure to allocate resources using optimization at either level can have significant health consequences.