The HIV epidemic is highly heterogeneous across countries and settings in terms of prevalence, incidence and transmission patterns. Currently, 95% of infections occur in developing countries,² which face additional hurdles in epidemic management due to lack of resources, poorly developed infrastructure, and sometimes unstable governance. Sub-Saharan Africa accounts for 67% of people living with HIV and 75% of AIDS-related deaths.¹ The highest HIV prevalence in the world is registered in countries such as Swaziland (26% of adults), Botswana (24.1% of adults), and South Africa (18.8% of adults).^{1, 6} However, HIV incidence is highest in Eastern Europe and Central Asia (Russian Federation, Ukraine), and some Asian countries (Indonesia, Vietnam).¹ Additionally, transmission patterns vary widely: in sub-Saharan Africa, HIV transmission occurs mainly among heterosexual couples and from mothers to their newborns; elsewhere the epidemic is primarily driven by injection drug use (IDU), commercial sex workers (CSWs), and men who have sex with men (MSM).¹

Despite years of concerted efforts and investment, an estimated 2.7 million people became newly infected with HIV in 2007.^{1, 7} Although significant decreases in HIV incidence have been achieved in some countries (e.g., Zimbabwe, Botswana, Thailand), incidence has increased in other areas of the world (e.g., Eastern Europe, Central Asia).^{1, 7}

Recently, considerable progress has been achieved in expanding access to treatment and increasing funding for HIV/AIDS programs.⁸ A six-fold increase in financing for HIV treatment in low- and middle-income countries in the past six years has led to a ten-fold increase in the number of people receiving antiretroviral drugs, with almost 3 million people in treatment by the end of 2007.¹ However, in 2007 in low- and middle-income countries an estimated 70% of eligible individuals still did not receive treatment.^{1, 9}

Prevention efforts have lagged behind treatment efforts in many parts of the world, even though prevention is considered essential for reversing the epidemic.^{1, 10} A recent estimate suggests that for every two patients who initiated antiretroviral therapy in 2007, five more acquired HIV infection.¹¹ In addition, it has been estimated that fewer than 10% of individuals at risk in 2007 were reached by key prevention interventions.^{2, 11} Many potentially effective prevention programs (e.g., condom promotion, harm reduction, education) have not been targeted or scaled up sufficiently to reach the desired number of individuals.^{12, 13} In some cases, funds have been allocated to largely ineffective approaches such as ABC programs (Abstinence, Be faithful, Condom use).^{12, 14, 15}

Additionally, prevention and treatment efforts need to be sustained by appropriate investments in health care infrastructure, development of human resources, multisectoral programs addressing education and stigma, and management structures that will support the HIV programs in the long term.⁵

Funds for HIV prevention and treatment fall far short of estimated need,¹⁶ even in developed countries.^{17, 18} In developing countries, the bulk of funding for HIV prevention and treatment has come from international donations and sponsor contributions — funding sources that may be jeopardized by the current economic downturn.^{5, 7} It is estimated that treatment programs will be impacted by changes in available funding in 33% of countries, affecting 61% of people on treatment.¹⁹ With scarce funding, and much progress remaining to be achieved in both treatment and prevention, efficient resource use is critical. No country has found an ideal solution yet, but relative successes in countries such as Thailand and Uganda have demonstrated that a consistent effort that includes a variety of interventions and sustained funding can achieve significant results.²⁰

In addition to the goal of improving health, many political and social considerations affect decisions about investment in HIV prevention and treatment.^{13, 21} Without adequate scientific data to support their adoption, some effective interventions, particularly those targeting most-at-risk populations (IDUs, CSWs, MSM), may be difficult to include in the HIV agenda, even if they are highly cost-effective or even cost saving. For example, longstanding debates have occurred over the acceptability of opiate substitution therapy and needle exchange for IDUs,^{22, 23} as well as condom promotion programs,^24–26 despite their proven effectiveness in reducing risky behavior. Ethical, cultural, and equity concerns are also important. For example, while the most effective use of funds may direct resources toward a single population group, in practice it may not be acceptable to neglect the rest of the population, due to ethical and equity concerns.

Significant scale up of global HIV prevention and treatment efforts is needed if the sixth Millennium Development Goal is to be achieved. However, determining the optimal investment in HIV prevention and treatment programs for any particular region is challenging. The heterogeneity of the epidemic across different regions means that different intervention packages with varying levels of investment will be most effective in different settings.^{13, 14, 27, 28} Moreover, different regions have different sets of accepted and available interventions and different levels of HIV funding; the effectiveness of interventions in any given setting may be uncertain; and the HIV epidemic is dynamic and nonlinear. How, then, can decision makers determine the best allocation of scarce HIV prevention and treatment resources for their setting?

Despite the broad body of research estimating the cost-effectiveness of various single HIV/AIDS interventions, decision makers often have little guidance as to which interventions or sets of interventions will yield optimal results in their particular settings. Since resources are limited and cannot be expanded indefinitely, it is essential to use them in a way that maximizes health outcomes (in measures of epidemic control such as HIV infections averted, quality-adjusted life years (QALYs) saved, etc.). Research allocation theory – defined as the methodology for allocating resources to optimize a goal (in this case, health benefits), subject to budgetary and other types of constraints²⁹ – effectively addresses this question. Much work has been done on the theoretical problem of resource allocation for controlling HIV, but little work has been done to translate the results of such work into practical tools for decision makers. Lasry and others^{30, 31} describe a number of difficulties encountered in implementing academic models of HIV resource allocation for practical purposes, including model complexity and data requirements; the presence of multiple stakeholders; political, social, and ethical considerations; and funding limitations and historical funding patterns.

In this section we describe key modeling approaches that have been used to address the HIV resource allocation problem, ranging from simple, deterministic linear projections to highly complex stochastic simulation models. Our goal is not to provide an exhaustive review of the literature; rather, we describe three broad categories of models (linear, dynamic, and simulation models) and present illustrative examples of each (Table 1). Comprehensive reviews of resource allocation models for epidemic control can be found elsewhere.^{46, 47}

Existing resource allocation models vary in their approach to incorporating the evolution of the epidemic, with some models assuming linear growth and others including nonlinear epidemic effects. Some models allow for a nonlinear relationship between funds invested in a program and its direct effects (e.g., the relationship between intensity of an educational program and the resulting level of change in risky behavior). For many programs, this relationship may not be linear.^{34, 48–52} Some models incorporate stochastic effects (e.g., variability in individual response to treatment) whereas others use only deterministic values (e.g., average rates of disease progression). Advantages and disadvantages of different types of models of infectious disease evolution and control are described elsewhere.^{46, 47, 53, 54}

Existing resource allocation studies often compare two or more interventions in one specific population or risk group in a small setting. It is often unclear how transferable these results are between different settings (e.g., low-income versus high-income countries), and how the results would change at the country level.^64–66 Moreover, few studies look at how the effects (e.g., HIV infections averted) of joint interventions differ from simply superimposing two or more single interventions. Depending on which interventions are chosen, the overlap can have significant effects, since it may decrease or increase total costs or total benefits, due to synergies or non-additive results.⁶⁴ For example, the total health effect of a program that reduces risky behavior combined with a program that expands HIV treatment will be less than the sum of the two programs' separate health effects: this is because the same HIV infection cannot be prevented twice.

Only a limited number of studies have analyzed the impact of decreasing/ increasing returns to scale for specific interventions or sets of interventions, though it has been shown to be an important phenomenon.^{34, 48–52} Most times, a linear relationship between funds or efforts invested and outcomes is assumed, which may distort projected results when scaling up programs.^{65, 66} Brandeau and others⁶⁷ analyzed prevention interventions for MSM and IDU populations representative of California, and showed that suboptimal decisions may be made without sufficient information about prevention program production functions: without any knowledge of production functions, 83% of the budget would be allocated to MSMs and the rest to IDUs, whereas with complete knowledge, 100% of the budget would be allocated to IDUs, averting seven times as many infections. The authors also demonstrated that linear approximations are particularly inappropriate in limited-resource situations.

Most existing models consider the allocation of resources between HIV prevention programs but do not consider the simultaneous allocation of resources to HIV treatment programs. This is a major gap in the resource allocation literature. Treatment has a variety of effects on the epidemic, including extended life span and reduced infectivity of treated individuals (and thus reduced HIV transmission).^68–72 Consideration of these effects along with prevention program effects provides a more accurate estimate of the effects of HIV program scale up and allows the decision maker to gain insight into the appropriate balance of prevention and treatment in a particular setting. Recent work by Cleary and others⁷³ included budget limitations in an analysis of HIV treatment scale up in South Africa, but focused only on treatment and did not consider prevention interventions.

In most HIV resource allocation models, the impact of interventions on the overall health care system, in terms of resources needed (human, capital, infrastructure), is rarely considered.^{74, 75} However, continuing progress in HIV control requires the expansion of structural interventions that strengthen health care systems (e.g., an increase in the capacity of rural health clinics to dispense medications), as well as combined approaches that include both behavioral and structural interventions.^{13, 76}

Finally, academic attempts to create accurate models have yielded theoretical insights but have often yielded complex systems that are not user-friendly and cannot be easily adopted by decision makers, who typically are not trained in epidemic modeling, simulation, and optimization. Often, in the interest of simulating the epidemic as closely as possible, the models require a significant number of parameters as inputs, many of which are unknown or difficult to estimate in practice.

The United Nations HIV/AIDS Programme (UNAIDS) provides guidelines for planners who want to intensify HIV prevention efforts.⁷⁹ In these guidelines, epidemics are classified as low-level, concentrated, generalized, or hyper-endemic, depending on total prevalence and prevalence in key risk groups. Decision makers are advised to “know their epidemic and their current response,” “match and prioritize their response,” “set ambitious, realistic and measurable prevention targets,” “tailor prevention plans,” and “utilize and analyze strategic information.”⁷⁹ However, these guidelines are insufficient: although they provide a useful framework for decision makers to analyze their epidemic, they do not fully reflect the complexity present in real-world settings and do not provide specific guidance about which sets of interventions are optimal for each setting.^{27, 80}

When guidance is insufficient, decision makers often rely on political interests, personal values, historical patterns, and their own professional experience to allocate resources.^{27, 34, 55, 81, 82} While such heuristics can be useful for decision making, they can lead to inconsistencies across time and settings, and to resource allocations that do not maximize health benefit. Ruiz and others⁵⁵ noted a lack of consistency in the allocation of federal HIV funds across the United States. Typical prevention funding decisions were made based on “proportionality” to reported AIDS cases. However, this strategy was found to be inadequate for the distribution of prevention resources, since it reflects the past condition of the epidemic and rewards states that are ineffectively managing their resources.⁸²

A review of allocation decisions made by HIV Prevention Community Planning Groups (CPGs) in the United States, which are local community planning groups that identify and prioritize unmet HIV prevention needs and interventions to address them,⁸³ showed that the decision-making process, though a laudable effort, was plagued by insufficient epidemiological information, imperfect correlation between available data and decision outcomes, and lack of consistency across CPGs in the methodology for making decisions.⁸¹_{Many CPGs} relied on “consensus decision making” based on their members' personal expertise and agenda (which can vary based on membership and location), or assigned “numerical weights” across interventions in order to facilitate the prioritization process. (See Kerr and Tindale⁸⁴ for an overview of research on group decision making.) While some CPGs considered the effectiveness of interventions in their decisions, not all of them took into account target population groups. Unfortunately, typical heuristics employed in practice for allocating resources (such as allocating funds in proportion to HIV incidence or reported AIDS cases) are likely to yield sub-optimal results.^{55, 85} This highlights the heavy burden CPGs face when making resource allocation decisions and the tradeoff between local insights and a standardized optimal decision process.

Some of the tools available for decision makers, such as the UNAIDS Estimation and Projection Package (EPP) and the Spectrum set of tools are meant only to provide projections of prevalence, incidence, AIDS deaths, treatment needs and numbers of orphans, based on surveillance data and measured epidemic patterns.⁸⁶ However, these tools do not model epidemic dynamics, do not specifically include a range of prevention interventions, and are not intended to provide insights into economic costs and benefits of interventions.

Some researchers have developed simple models to address investment in a single type of HIV prevention. For example, the AVERT model (http://www.fhi.org/en/hivaids/pub/guide/avert.htm) is a spreadsheet-based model that estimates the number of HIV infections averted by interventions that aim to reduce sexual transmission of HIV.⁸⁷ Another spreadsheet-based model (http://www.futuresinstitute.org/pages/MaleCircumcision.aspx) estimates cost and impact (number of HIV infections averted, change in HIV incidence and prevalence) for different levels of scale up of male circumcision services and for different target populations.⁸⁸

The HIV Prevention Funding Allocation Model (http://www.cdc.gov/hiv/topics/prev_prog/ce/resources.htm), designed for use by CPGs in the US, determines the investment in each priority population group and intervention that maximizes the weighted number of HIV infections averted over a five-year horizon, subject to budget, equity, and other constraints.^{89, 90} The model quantifies the effects of multiple interventions in terms of infections averted, as a function of investment, and accounts for variability of intervention effectiveness as a function of HIV prevalence. The model allows decision makers to change constraints so as to determine the tradeoffs between different allocations of prevention funds.

At the global level, aside from decentralized country-level efforts, the most notable comprehensive model currently available to decision makers is the GOALS Model (http://www.policyproject.com/software.cfm?page=Software&ID=GOALS). The model, designed in Microsoft Excel, has a user-friendly interface and allows decision makers to compare the cost and epidemic impact (HIV prevalence, incidence, and infections averted) of various budget allocations between prevention, care, and mitigation.⁹¹ Users input values for multiple parameters describing their local situation. When accurate parameter values are unavailable, the model defaults to a series of standard values. The model uses a simple epidemiological projection based on the size of the susceptible and infected populations, and the probability of transmission between them (calculated based on risk profiles).⁹²

The GOALS model represents a notable attempt to create a practical tool for decision makers, has gained widespread use in practice, and has had a significant impact on policy making. However, the model has several limitations. Decreasing/increasing returns to scale of investment are not considered in the model, nor is the impact of combinations of interventions versus single interventions. The model does not capture uncertainty about parameters, which can be problematic if the country's epidemic has characteristics that are very different from the case considered in the default parameters (e.g., importance of different risk groups or transmission modes). Additionally, the GOALS model does not optimize health outcomes (e.g., HIV infections averted or QALYs gained) given the available level of resources. Instead, users must try various budget allocations between interventions, and then compare results, a process that may not lead to the best use of scarce resources.

The intended end-user of a model with these features is the decision maker in charge of country-level HIV planning (such as a ministry of health or a national HIV response program), but the customization embedded in the model would make it suitable for other levels of planning (regional, local, or NGO-based plans). In practice, multiple entities may be in charge of various parts of the HIV programs, and interventions may have different streams of funding. However, most countries have a coordinated national strategy¹ to which all these entities aim to adhere, so such a model could be used at upper and lower decision making levels.