Our model estimates reductions in new HIV infections associated with current and optimized HIV prevention expenditures and results in three main recommendations. First, the allocation to testing interventions should increase and further target MSM and IDUs. Second, counseling and education interventions ought to provide a greater focus on HIV-positive persons. And lastly, more funds should be allocated to those at high risk rather than the general population. As expected, annual incidence is highest given no allocation of funds and minimized under the model’s optimized allocation of the budget.
Relative to no investment of funds, the baseline and optimal allocation scenarios avert 29,035 and 60,271 HIV infections, respectively. Given the programmatic annual budget of $327 million, this implies a cost of $56,311 per infection averted for the baseline as compared to no allocation and $27,128 per infection averted for the optimal allocation as compared to no allocation. Therefore, both the current baseline and the optimal allocation of funds can be considered cost-saving when compared to the HIV lifetime treatment costs. The baseline and optimal scenarios allocate the same budget amount so a cost per infection averted relating both scenarios cannot be inferred. The undiscounted expenditures total $1.6 billion while the total medical cost of treating the infections should they not be averted would be $11 billion under the current allocation strategy and $22 billion under the optimized allocation. In the incremental budget scenario, the marginal infections averted decreases and the cost per infection averted increases as the annual budget increases reflecting the decreased rate at which new infections are averted given additional funding. At the lowest budget level, funds are spent on the most cost-effective interventions and target groups, as the budget increases and the maximum capacity constraints are reached for those most cost-effective interventions and target groups, the additional funds are allocated to less cost-effective targets thereby increasing the cost per infection averted. Nonetheless, even at $500 million per year, the lifetime HIV treatment cost of $367,000 
exceeds the estimated cost per infection averted by $332,000 indicating a cost-saving level of investment.
Given the many inputs and outputs to our model, sensitivity analysis is not straightforward. Our objective in sensitivity analysis was to evaluate whether the main model recommendations are upheld throughout reasonable variations in the input data. Of over 100 sensitivity analysis scenarios conducted, only 9 scenarios altered the key results presented in by more than ten percentage points. However, those scenarios tended to reinforce the model’s recommended focus on MSM. The budget constraint scenario highlights the critical importance of testing, targeting high-risk groups and diagnosed positives. The classic epidemic control theory of focusing on high-transmission core groups 
endorses our results.
Our model is a simplified representation of the actual allocation process of DHAP’s HIV prevention resources and thus translating the model’s output into practice can be difficult. The model assumed that all members of high-risk population subgroups are reachable and can be perfectly targeted with interventions. This simplifying assumption may lead to an overestimation of the model’s impact because some target population members cannot be easily reached (and may not even know that they or their partner are at risk for HIV) and programs for these populations will have to contact some persons who are not target population members in order to reach those who are most at-risk. Thus, the programmatic efficiency assumed in the model cannot be achieved in real-world programs. Nonetheless, our model provides insights on causal relationships and helps us to identify areas where prevention programs can have the most impact.
Results of the HPTN052 randomized control trial among serodiscordant couples indicated that antiretroviral therapy reduces incidence of HIV transmission to the negative partner by 89% 
. Our model does not explicitly consider antiretroviral treatment costs or the variation in HIV transmission rates associated with viral load suppression and stage of disease. However, early diagnosis and treatment reduces HIV transmission so the benefits of HIV diagnoses may be underestimated in this analysis and an even greater focus on testing than suggested by our model results may be warranted.
We made linear assumptions on scalability in the intervention cost and outcome functions of the model; and the shape of such functions may make a difference in the optimal allocation of funds 
. Economic evaluations (such as cost-effectiveness analysis) of HIV prevention programs typically consider one funding level and their results do not inform on the additional benefit to be gained (or lost) through increasing or decreasing the investment 
. While our model structure could support more complex functions, the true shape of such functions is not known and no reliable data exist to support them.
The per-person intervention costs in our model were largely derived from microcosting methods where all cost items relating to an intervention are detailed and cumulated. In contrast with program budget methods where the overall allocated budget is simply divided by the number of persons served, microcosting leads to an underestimate of per-person cost 
. In actuality, federally allocated funds are not typically spent completely on the direct costs of implementing programs; some portion may be spent on the indirect costs of organizations that implement the interventions and other entities involved in getting funds to those organizations. For example, DHAP allocates funds to state health departments who in turn fund local health jurisdictions, who may fund community-based organizations to deliver an intervention. Microcosting does not include the additional funds spent channeling the resources through these levels to the intended intervention. Though the method we applied may overestimate the numbers of persons that can be reached given the overall budget constraint, the budget constraint scenario presented mitigates this limitation. If the overage associated with indirect costs is 60%, then the available budget for program spending would be about $200 million, and that scenario echoes the main conclusions of the model.
As with most studies, input data are often being revised and updated. All data inputs were reviewed internally by a panel of 25 persons within the CDC, data were then validated through an external peer review process involving by 18 experts. We plan to publish updates of this analysis, including the effects of newly implemented programs, including the Expanded HIV Testing Initiative for populations disproportionately affected by HIV, primarily African Americans.
Models cannot encompass all the dynamic, complex and often qualitative realities of HIV resource allocation or be expected to provide prescriptive results 
. Resource allocation models for the control of infectious diseases, such as the one presented here, are intended to provide guidance on improving the allocation of funds; they are not designed to provide accurate epidemic projections. Such models are typically comprised of an optimization component that sequentially supplies allocation scenarios to an epidemic component which projects infections until the optimal allocation is found. This process can easily reach millions of iterations and depends on the number of variables and constraints; therefore, the epidemic component must be kept simple and efficient. More comprehensive epidemic models that are neither bound to an optimization engine nor anchored to a time horizon that reflects a budget cycle can be designed to measure epidemic growth more precisely over time. Given the aforementioned considerations, our model results are not intended to make epidemic projections over time, though their interpretation as a measure of the difference in impact between scenarios is robust.
Our results have, in part, provided the impetus for other programs, such as the Expanded HIV Testing Initiative. A recent analysis of the Initiative suggests that the program achieved a return of $1.95 for every dollar invested by CDC 
. Lastly, according to the National HIV/AIDS Strategy (NHAS), intensifying prevention efforts for MSM, IDUs, black and Hispanic men and women, and increasing the percentage of people living with HIV who know their serostatus is required to reduce annual HIV incidence 
. The main conclusions of the HIV resource allocation model serve to substantiate and provide rational economic evidence for the NHAS goals. Decision makers responsible for the allocation of DHAP’s programmatic funds report that the national HIV resource allocation model, along with program and other data, provides valuable guidance to target resources and improve the impact of HIV prevention efforts in the United States.