Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25097 http://onlinelibrary.wiley.com/doi/10.1002/jia2.25097/full | https://doi.org/10.1002/jia2.25097 REVIEW ARTICLE How should HIV resources be allocated? Lessons learnt from applying Optima HIV in 23 countries Robyn M Stuart1,2 , Laura Grobicki3, Hassan Haghparast-Bidgoli3, Jasmina Panovska-Griffiths4,5,6, Jolene Skordis3, Olivia Keiser7,8, Janne Estill7,8,9, Zofia Baranczuk7,8,10, Sherrie L Kelly2,11 , Iyanoosh Reporter2, David J Kedziora2,11,12, Andrew J Shattock13, Janka Petravic2, S Azfar Hussain2, Kelsey L Grantham11, Richard T Gray13, Xiao F Yap2, Rowan Martin-Hughes2, Clemens J Benedikt14, Nicole Fraser-Hurt14 , Emiko Masaki14, David J Wilson14, Marelize Gorgens14, Elizabeth Mziray14, Nejma Cheikh14, Zara Shubber14, Cliff C Kerr2,12 and David P Wilson2,11 Corresponding author: Robyn M Stuart, Universitetsparken 5, København Ø 2300, Denmark. (robyn@math.ku.dk) Abstract Introduction: With limited funds available, meeting global health targets requires countries to both mobilize and prioritize their health spending. Within this context, countries have recognized the importance of allocating funds for HIV as efficiently as possible to maximize impact. Over the past six years, the governments of 23 countries in Africa, Asia, Eastern Europe and Latin America have used the Optima HIV tool to estimate the optimal allocation of HIV resources. Methods: Each study commenced with a request by the national government for technical assistance in conducting an HIV allocative efficiency study using Optima HIV. Each study team validated the required data, calibrated the Optima HIV epidemic model to produce HIV epidemic projections, agreed on cost functions for interventions, and used the model to calculate the optimal allocation of available funds to best address national strategic plan targets. From a review and analysis of these 23 country studies, we extract common themes around the optimal allocation of HIV funding in different epidemiological contexts. Results and discussion: The optimal distribution of HIV resources depends on the amount of funding available and the char- acteristics of each country’s epidemic, response and targets. Universally, the modelling results indicated that scaling up treat- ment coverage is an efficient use of resources. There is scope for efficiency gains by targeting the HIV response towards the populations and geographical regions where HIV incidence is highest. Across a range of countries, the model results indicate that a more efficient allocation of HIV resources could reduce cumulative new HIV infections by an average of 18% over the years to 2020 and 25% over the years to 2030, along with an approximately 25% reduction in deaths for both timelines. However, in most countries this would still not be sufficient to meet the targets of the national strategic plan, with modelling results indicating that budget increases of up to 185% would be required. Conclusions: Greater epidemiological impact would be possible through better targeting of existing resources, but additional resources would still be required to meet targets. Allocative efficiency models have proven valuable in improving the HIV plan- ning and budgeting process. Keywords: HIV modeling; allocative efficiency; cost-effectiveness; optimal HIV investment; resource allocation; resource needs Additional Supporting Information may be found online in the Supporting information tab for this article. Received 29 June 2017; Accepted 5 March 2018 Copyright © 2018 The Authors. Journal of the International AIDS Society published by John Wiley & sons Ltd on behalf of the International AIDS Society. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. 1 | INTRODUCTION different funding bodies, the influence of other sectors, histor- ical precedent, the desire to promote equity among vulnerable If decisions on the allocation of health resources were guided or prioritized population groups, attempts to reduce financial by the principals of health economics alone, funds would be risk and the desire to maintain health security. In addition, allocated in ways intended to lead to the greatest reductions governments and other health funders are often challenged in disease burden overall. However, economics is not – and by a lack of information about the cost-effectiveness and has never been – the sole factor influencing decisions on the impact of health interventions at a population level. The com- allocation of health funds. Such decisions are also influenced bination of these factors means that the allocation of health by numerous other important factors, including the desires of funds is often vastly different to how it would be under a 1 Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25097 http://onlinelibrary.wiley.com/doi/10.1002/jia2.25097/full | https://doi.org/10.1002/jia2.25097 purely population-level, evidence-based and health-outcome- Although there was variation in the populations and pro- focused framework [1]. grammes that were considered across the 23 studies, our As with any component of health budgeting, planning an focus is on the similarities that emerge from applying the HIV response can be an extremely time-consuming process. same methodological framework for the analysis of the effi- To aid with this process, several different tools have been ciency of the HIV response. Thus, the criteria for inclusion in developed and employed in different contexts, including the this review were: (1) that the study was requested by the widely used GOALS resource estimation tool, the Asian Epi- government of the country in question; (2) that the Optima demic model and the Optima HIV tool [2]. All three tools are HIV model was used to estimate the mathematically optimal equipped with a resource requirements estimation feature distribution of national HIV resources given government-speci- intended to help with budgeting. However, a unique feature of fied epidemiological targets; (3) that the analysis represented the Optima HIV tool is its allocative efficiency optimization the entire country’s epidemic; (4) that the government agreed algorithm. Within health economics, a response is described as for results to be released; and (5) that the study had not been allocatively efficient if funds are allocated across different HIV replaced by a more recent study that was already being interventions and delivery modalities in the way that leads to included. The first three criteria were applied in order to the best possible epidemic outcomes given any relevant con- ensure comparability of results, the fourth was added because straints. This is particularly important in the current epidemio- some of the studies were confidential and intended solely for logical [3] and funding [4,5] context. Since 2002, an estimated internal ministry purposes, and the fifth was added to avoid US$80.3 billion in development assistance for HIV pro- redundancy. The 23 studies mentioned previously comprise grammes has been disbursed in over 100 lower-income coun- the full quota of studies that met these criteria. The Optima tries [6]. However, the trend in funding over the past seven HIV model was also used in other country studies, and in anal- years has been almost flat. Development assistance for HIV in yses at sub-national regions, but these additional studies did 2015 totalled US$7.5 billion, which represented a 13% not meet the above criteria and thus were not included in this decrease from 2014 levels (the first funding decrease in five review (see Table S1 for a complete listing). years) and brought the total amount of funding back to 2008 The Optima HIV tool was designed and developed by the levels [6]. Thus, the question of how to get the most out of Optima Consortium for Decision Science (the Optima Consor- the available HIV funding is more essential now than ever tium) with technical inputs and guidance from the World Bank. before. It is now generally accepted that resource allocation The tool itself is based on a compartmental model of HIV decisions should be informed by, or grounded in, explicit crite- transmission and disease progression, and is capable of pro- ria based on cost-effectiveness to maximize health benefits ducing estimates of epidemic trends, resource needs, and the with the resources available [7-20]. impact and cost-effectiveness of HIV responses. Furthermore, In this paper, we discuss and compare studies conducted it can estimate the allocation of resources across programmes over six years and across 23 countries in Africa, Asia, East- that best addresses national HIV targets while considering ern Europe and Latin America, each of which used Optima various logistic, political and ethical constraints [44]. HIV to estimate the potential gains that could be achieved Each study commenced with a request, made to a devel- by reallocating resources in a more efficient way. We do not opment agency by the national government, for technical intend to provide a formal meta-analysis, but rather a broad assistance in conducting an HIV allocative efficiency study. qualitative comparison of the results that the modelling anal- In most cases, this request was made to the World Bank; yses found in each context. By synthesizing the results, we in two cases (Tajikistan and Uzbekistan) it was made to the aim to identify common principles for the optimal allocation UNDP; and in six cases (Armenia, Belarus, Kazakhstan, Kyr- of HIV resources. gyzstan, Moldova and Ukraine), it was made to a group of funding agencies. The agreement to conduct a study using Optima HIV was then formalized on the basis of a scope of 2 | METHODS work document that outlined the key policy questions for the modelling analysis. An analytic team was then formed to The 23 studies included in this review were conducted in carry out the work agreed upon, with team members typi- Indonesia [21] and Vietnam [22] from the East Asia and Pacific cally including representatives from the government (e.g. (EAP) region; Argentina [23], Colombia [24], Mexico [25], and from the ministry of health, the national team responsible Peru [26] from the Latin America and Caribbean (LAC) region; for monitoring and evaluation, or the national AIDS commis- Armenia [27], Belarus [28], Bulgaria [29], Georgia [30], Kaza- sion), from partnering organizations (e.g. the World Bank, khstan [31], Kyrgyzstan [32], Macedonia [33], Moldova [34], the Global Fund, UNDP, PEPFAR, UNAIDS) and from the Tajikistan [35], Ukraine [36], and Uzbekistan [37] from the Optima Consortium. The analytic team then proceeded to Eastern Europe and Central Asia (EECA) region; Zambia [38] follow the steps outlined in Table 1. Reports with full from the Sub-Saharan Africa (SSA) region; and Cote d’Ivoire details of data, context, methods, results, interpretation and [39], Niger [40], Senegal [41], Sudan [42], and Togo [43] from discussion for all 23 countries are available either on the the West and Central Africa (WCA) region. Studies were con- Optima Consortium website (www.ocds.co) or the World ducted in partnership with institutions including the World Bank’s Open Knowledge Repository, or can be made avail- Bank, the Global Fund to fight AIDS, Tuberculosis and Malaria able upon request. (the Global Fund), the United States President’s Emergency In this review, we analyze the findings across the 23 studies Plan for AIDS Relief (PEPFAR), the HIV Modelling Consortium, using a thematic analysis, stipulating that these themes must the Joint United Nations Programme on HIV/AIDS (UNAIDS) have been a conclusion of at least three studies before they and the United Nations Development Programme (UNDP). could be included in this review. 2 Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25097 http://onlinelibrary.wiley.com/doi/10.1002/jia2.25097/full | https://doi.org/10.1002/jia2.25097 Table 1. Steps in an allocative efficiency study, as followed for each of the countries considered in this review Difficulties encountered and steps Step Rationale Processes followed taken to overcome them 1. Identify the The burden of HIV varies In all 23 studies, the entire national The desire to capture the particulars of population groups considerably within countries population was stratified according the epidemic dynamics must be and HIV according to factors such as to age, sex and risk behaviour. In weighed up against the practical programmes geography, behavioural tendencies, addition, the population was further constraints around data availability. suitable for age and sex. The population groups stratified according to geographical Criteria were defined to guide the inclusion in the included in an allocative efficiency region in Moldova and Cote decision on whether to include a analysis. study should be selected to capture d’Ivoire. population: the population should (a) this heterogeneity. be clearly defined, (b) play a substantial role in the country’s epidemic, (c) currently or could be targeted with HIV programmes, and (d) have a minimum amount of data or reliable estimates on population size and HIV prevalence. 2. Collect and A determination of how to optimally All available data were collected and In several contexts, there were data gaps validate the data target an HIV response must be validated by the analytic teams. in the epidemiological, behavioural and required for the data-driven. Demographic, programmatic data. Often, this step analysis. behavioural and epidemiological and the first step were conducted data for each population group iteratively, with populations being first must be collected, as well as considered for inclusion and then later programmatic data including unit removed if insufficient data were costs, expenditure and historical available. levels of coverage (particularly important for antiretroviral therapy programmes) for each programme and service delivery modality. 3. Calibrate the The calibration process involves Typically, the model was calibrated to Attaining a realistic calibration relies on model to available adjusting a subset of the model’s historical data on HIV prevalence, having good data to input to the data. parameters in order to minimize the number of HIV diagnoses, and model. When difficulties were the mean absolute percentage the number of people receiving experienced with calibrations, this error between the model’s antiretroviral therapy, as well as would often indicate issues with the estimates and the observed data, (where requested) the outputs of underlying data. In this sense, the and then subjecting the projections other models that the country had process of model calibration is produced by the model to a previously used. conducted synchronously with the process of scrutiny and validation process of data validation. by the district, province and national health departments. 4. Establish cost Cost functions define a relationship Each analytic team agreed on realistic Data to inform cost functions is difficult functions. between spending on an HIV assumptions on both the maximal to obtain. In most cases, the cost service and the expected coverage attainable coverage for each functions were partially informed by and outcome of that service programme/modality and the data and partially be expert opinion. amongst the target population. behavioural outcome expected to prevail under that maximal coverage level. 5. Calculate the The allocation of funds that would National strategic targets were In some cases, the initial optimization optimal allocation deliver the outcome closest to identified by the analytic teams, produced a recommendation that the of available funds. national strategic targets can be usually in consultation with country deemed politically or calculated using Optima’s ministries of health or other programmatically infeasible. In such mathematical optimization responsible bodies. cases, there was an option to rerun algorithm. the optimization with additional constraints. 3 Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25097 http://onlinelibrary.wiley.com/doi/10.1002/jia2.25097/full | https://doi.org/10.1002/jia2.25097 Table 1. (Continued) Difficulties encountered and steps Step Rationale Processes followed taken to overcome them 6. Produce epidemic The future evolution of the HIV We projected the future evolution of The future of HIV funding is uncertain. trajectories. epidemic depends on the future of the epidemic assuming that the To account for this uncertainty, the HIV response. The previous future HIV budget was allocated (i) epidemic projections were produced analytic steps defined the nature of as per the last reported HIV under a range of different assumptions this dependency, and determined spending pattern and (ii) as per the about future budget availability. the response that would lead to optimal allocation of funds the best epidemic outcomes. The calculated in the previous step. final step translates these responses into epidemic outcomes. Where possible, we supplement qualitative findings with would be attainable by optimally allocating resources (on top quantitative metrics. In particular: (a) we calculate the average of the reductions due to continuing current HIV responses). In reduction in new infections and HIV-related deaths that was the eight studies where the time horizon for minimizing new estimated to be possible via a reallocation of funds; (b) we cal- infections was 2030, an average reduction in new infections culate the average increase in treatment coverage that was of 25% (IQR 4% to 30%) was estimated to be possible recommended based on the model findings; and (c) we calcu- through better allocation of resources. (The remaining three late correlation coefficients between the proportion of new studies considered timelines to 2025 or 2010; see Table 2.) HIV infections acquired by each population (as defined by Better targeting of resources was also able to further reduce age/risk/geographical location) and the share of the HIV pre- estimated deaths: by 22% (IQR 9% to 28%) on average vention budget that the model recommended should optimally (across nine studies) by 2020, and by 29% (IQR 7% to 36%) be targeted at these populations. on average (across eight studies) by 2030. Details are summa- We note some limitations of the methods used in this rized in Table 2. review to synthesize the results of the studies. Given the dif- ferences in the inputs that were used for each study and the 3.2 | Increased allocations to treatment outputs that were generated, it is challenging to make rigor- ously quantitative comparative statements; thus, we have kept If no increases in the overall HIV budget are expected, the our comparative analysis general and it should not be consid- analyses recommended increasing the share of HIV budgets ered as a formal meta-analysis. The 23 studies included here allocated to ART from 49% to 64% on average, which would were conducted over a period of several years, and there in turn increase estimated average national ART coverage were various changes to the underlying model, the types of from 30% to 42% as a percentage of all PLHIV (Table 2). data that were available, and the types of results that were The model’s recommendations to expand treatment had generated during this time. The predictions provided in these important but varied consequences for the role of HIV testing studies are limited by the quality of the data and assumptions and counselling (HTC) programmes. In 14 countries, there used to inform them. In numerous settings, there were large was known to be a large pool of people who had already been uncertainties and/or missing data for key input variables (such diagnosed but were not on treatment. In these cases, the as key population sizes, prevalence levels, and/or time trends). modelling results indicated that it would be better to increase Furthermore details of the particular limitations of each study funding to ART programmes first (including programmes link- are included in the relevant reports. ing and retaining people to care), and that testing programmes should not be scaled up until those already diagnosed had been initiated on ART. This was consistently the case, even 3 | RESULTS AND DISCUSSION when testing programmes were delivered via high-yield or low-cost service modalities. 3.1 | Estimated epidemiological impact Across all 23 studies, the modelling results produced by the 3.3 | Increased allocations to the populations with Optima HIV tool indicated that by reallocating existing funds the highest incidence for HIV, it would be possible to reduce both new HIV infec- tions and HIV-related deaths. The magnitude of the epidemio- There was a marked correlation between the share of new logical reductions attainable would depend on multiple factors, HIV infections acquired by each population and the share of including the timeframe of consideration, epidemic type and the HIV prevention budget that the model recommended scale, the response profile and level of resourcing available. In should be targeted to them. The correlation coefficient was 12 studies, the primary objective of analyses (aligned with particularly high (0.77) for programmes targeted at PWID national strategic plans) was to minimize new infections by (Figure 1a). The overall correlation coefficient was lower 2020; in these studies, the modelling results indicated that an (0.43) in the case of programmes for FSW, but this was additional 18% (IQR 6% to 29%) reduction in new infections strongly influenced by the three countries from Western and 4 Table 2. Summary of results from 23 allocative efficiency studies Funding Key data Optimization results under the current budget required for NSP targetsa ART Optimal % Funds coverage ART % reduction required (% of Budget US$/ coverage (% reduction in in as a % of Country Yearb Epidemic PLHIVb PLHIV) (US$m) PLHIV Programme priority areas of PLHIV) infections deaths current budget Eastern Europe and Central Asia Armenia 2013 Concentrated 3600 65% 4.5 1259 ↑ Scale-up ART, OST, programmes 94% 17%c 29%c 265% for PWID & FSW – Maintain PMTCT, programmes for prisoners & PWID ↓ Scale-down GP programmes (SBCC, HTC) Belarus 2013 Concentrated 35,000 32% 20.5 586 ↑ Scale-up ART, OST, programmes 46% 7%c 25%c 125% Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25097 for PWID – Maintain PMTCT, programmes for FSW & MSM ↓ Scale-down GP programmes (SBCC, HTC) http://onlinelibrary.wiley.com/doi/10.1002/jia2.25097/full | https://doi.org/10.1002/jia2.25097 Bulgaria 2014 Concentrated 6000 21% 8.6 1437 ↑ Scale-up OST, programmes for 21% 21%d 7%d 264% PWID, MSM & prisoners – Maintain ART, programmes for FSW ↓ Scale-down GP programmes (SCCC, HTS) Georgia 2014 Concentrated 8900 32% 14.7 1657 ↑ Scale-up ART, HTC for KPs, 59% 16%d 36%d 140% programmes for MSM – Maintain programmes for PWID & FSW, OST (60%) ↓ Scale-down GP programmes (HTC) Kazakhstan 2013 Concentrated 23,000 22% 34.0 1478 ↑ Scale-up ART, HTC, programmes 30% 6%c 22%c 137% for PWID & MSM – Maintain PMTCT, programmes for FSW ↓ Scale-down GP programmes (SBCC, HTC) 5 Table 2. (Continued) Funding Key data Optimization results under the current budget required for NSP targetsa ART Optimal % Funds coverage ART % reduction required (% of Budget US$/ coverage (% reduction in in as a % of Country Yearb Epidemic PLHIVb PLHIV) (US$m) PLHIV Programme priority areas of PLHIV) infections deaths current budget Kyrgyz 2013 Concentrated 7500 13% 16.0 2130 ↑ Scale-up ART, HTC, programmes 41% 28%c 53%c 190% Republic for PWID & MSM – Maintain PMTCT, OST, programmes for FSW Macedonia 2013 Concentrated 900 22% 6.5 7209 ↑ Scale-up ART, HTS for KPs, 63% 85%d 87%d 100% programmes for MSM – Maintain programmes for PWID (NSP, OST) & FSW ↓ Scale-down GP programmes Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25097 (SBCC) Moldova 2013 Concentrated 15,000 24% 0.8 51 ↑ Scale-up ART, programmes for 38% 20%c 16%c 233% FSW, PWID & MSM – Maintain PMTCT ↓ Scale-down GP programmes http://onlinelibrary.wiley.com/doi/10.1002/jia2.25097/full | https://doi.org/10.1002/jia2.25097 (condoms, HTC) Tajikistan 2013 Concentrated 15,000 10% 14.1 940 ↑ Scale-up ART, all KP 15% 5%c Not incl. Not incl. programmes – Maintain HTC, PMTCT ↓ Scale-down Youth, community mobilization, SBCC Ukraine 2013 Concentrated 210,000 30% 85.2 406 ↑ Scale-up ART, lab monitoring 41% 3%c 9%c Not incl. – Maintain all KP programmes, PMTCT ↓ Scale-down GP programmes (HTC) Uzbekistan 2011 to Concentrated 42,000 16% 21.1 502 ↑ Scale-up ART, HTC 17% 44%c Not incl. Not incl. 2012 – Maintain all other prevention ↓ Scale-down youth programmes Latin America and the Caribbean Argentina 2012 Concentrated 100,000 41% 501.9 5020 – Maintain response 41% 0%d 0%d Not incl. 6 Table 2. (Continued) Funding Key data Optimization results under the current budget required for NSP targetsa ART Optimal % Funds coverage ART % reduction required (% of Budget US$/ coverage (% reduction in in as a % of Country Yearb Epidemic PLHIVb PLHIV) (US$m) PLHIV Programme priority areas of PLHIV) infections deaths current budget Colombia 2012 Concentrated 130,000 45% 60.0 545 ↑ Scale-up ART, programmes for 53% 28%d 24%d Not incl. MSM & homeless ↓ Scale-down GP programmes (HTC) Mexico 2011 Concentrated 170,000 52% 432.4 2298.5 ↑ Scale-up ART 56% 4%d 7%d 125% – Maintain PMTCT ↓ Scale-down GP programmes Peru 2014 Concentrated 88,000 57% 91.8 1044 ↑ Scale-up ART 57% 38%d 33%d Not incl. – Maintain PMTCT Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25097 ↓ Scale-down GP programmes (condoms, SBCC, HTC) Sub-Saharan Africa Zambia 2012 Mixed 1,100,000 55% 284.2 258 ↑ Scale-up ART, programmes for 60% 5%d 36%d 133% FSW http://onlinelibrary.wiley.com/doi/10.1002/jia2.25097/full | https://doi.org/10.1002/jia2.25097 – Maintain PMTCT ↓ Scale-down HTC, GP programmes East Asia and the Pacific Indonesia 2012 Mixed 590,000 9% 87.0 147 ↑ Scale-up OST, programmes for Not incl.e 5%c 2%c Not incl. PWID, MSM, FSW ↓ Scale-down GP programmes (condoms, SBCC, HTC) Vietnam 2012 Concentrated 250,000 Not incl. 136.1 544 ↑ Scale-up HTC, programmes for Not incl.e 16%f 1%f Not incl. FSW, MSM ↓ Scale-down GP programmes, NSP, OST, STI programmes West and Central Africa Cote d’Ivoire 2013 Mixed 470,000 29% 106.0 226 ↑ Scale-up ART, HTC, FSW 32% 5%c 6%c 283% programmes ↓ Scale-down GP programmes (condoms, HTC) 7 Table 2. (Continued) Funding Key data Optimization results under the current budget required for NSP targetsa ART Optimal % Funds coverage ART % reduction required (% of Budget US$/ coverage (% reduction in in as a % of Country Yearb Epidemic PLHIVb PLHIV) (US$m) PLHIV Programme priority areas of PLHIV) infections deaths current budget Niger 2012 Concentrated 54,000 24% 16.1 298 ↑ Scale-up ART, PMTCT, FSW 43% 30%g 19%g Not incl. programmes – Maintain programmes for prisoners, migrants, MSM, mine workers, truckers, OVC, PEP ↓ Scale-down GP programmes Senegal 2013 Concentrated 48,000 33% 24.3 505 ↑ Scale-up ART, PMTCT, 50% 31%c 28%c Not incl. programmes for FSW & MSM ↓ Scale-down GP programmes Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25097 (HTC, SBCC) Sudan 2013 Concentrated 56,000 6% 12.3 220 ↑ Scale-up ART, programmes for 12% 36%c Not incl. 134% FSW & clients & MSM ↓ Scale-down GP programmes Togo 2014 Mixed 110,000 31% 20.1 183 – Maintain response 31% 0%g 0%g 155% http://onlinelibrary.wiley.com/doi/10.1002/jia2.25097/full | https://doi.org/10.1002/jia2.25097 Averages 30% 1285 42% 18% to 2020 22% to 176% 25% to 2030 2020 29% to 2030 ART, antiretroviral therapy; OST, opiate substitution therapy; PWID, people who inject drugs; FSW, female sex workers; PMTCT, prevention of mother-to-child transmission; MSM, men who have sex with men; GP, general population; SBCC, social and behaviour change communication; HTC, HIV testing and counselling; OVC, orphans and vulnerable children; KP, key population; VL, viral load; PEP, post-exposure prophylaxis; Not incl., indicator not requested for this study. a Percentage increase over the total expenditure at the last NASA that would be required to meet the National Strategic Plan (NSP) targets, assuming that funds were optimally allocated. b Year for which latest National AIDS Spending Accounts were available at the time study was conducted, and estimate of the number of PLHIV in that year as published in the country reports. c Percentage reduction in cumulative infections/deaths over the years until 2020 that could be obtained via optimally allocating resources. d Percentage reduction in cumulative infections/deaths over the years until 2030 that could be obtained via optimally allocating resources. e In Vietnam, and Indonesia, ART was not considered as part of the pool of funding available for reallocation but rather as required resources earmarked as an essential expense. Therefore, we did not estimate optimal coverage levels for these two countries. f Percentage reduction in cumulative infections/deaths over 2006 to 2010 that could be obtained via optimally allocating resources. g Percentage reduction in cumulative infections/deaths over the years until 2025 that could be obtained via optimally allocating resources. 8 Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25097 http://onlinelibrary.wiley.com/doi/10.1002/jia2.25097/full | https://doi.org/10.1002/jia2.25097 (a) (b) (c) Figure 1. The relationship between the share of infections in a particular population/district, and the share of the HIV budget for preven- tion programmes. Results pertain to the year for which latest National AIDS Spending Accounts were available at the time study was con- ducted – these years are presented in Table 2. The share of infections by sub-population was not available for Peru, Mexico, Colombia, Argentina, Tajikistan or Ukraine. (a) PWID across 17 countries, (b) FSW across 17 countries, (c) MSM across 17 countries. Central Africa in which rates of indirect sex work amongst the among the general population must also be taken into general female population are high, so programmes targeted account. to the general female population are likely to be effective proxies. After removing these countries, we found a correla- 3.4 | Targeting the right delivery approaches to tion coefficient of 0.88 (Figure 1b). However, we found that maximize coverage the correlation coefficient was lower for programmes targeted at MSM (0.20; Figure 1c). The question of how best to target the HIV response to the Almost all countries experience geographical variation in appropriate populations and geographic areas is crucial, but it both the severity of the HIV epidemic and in the costs of must be considered alongside the equally important question delivering the HIV response. Often – especially in concen- of how to deliver these HIV services at the highest possible trated epidemics – there is significant overlap between the quality and the lowest feasible cost in ways that will reach a geographical distribution of the HIV epidemic and the geo- wide variety of the intended populations. In general, informa- graphical distribution of the key affected populations, and as tion on the heterogeneity of the costs and the impact of deliv- a result it may be possible to ensure that the HIV response ering HIV services both within and across countries is scarce, is targeted at the right places simply by targeting the key and information on the determinants of this variation is even populations, where such populations can be safely found and scarcer. Across all 23 studies, significant attention was given supported with programmes. However, in areas where there to the question of how the overall HIV response could be is significant stigma and discrimination or where some key improved by lowering costs while maintaining or improving population behaviour is illegal, geographic targeting as a service delivery quality and modality to ensure the highest proxy for key population targeting might be needed in order coverage. In general, the recommendation was for countries to ensure the safety of programme staff and the key popula- to rigorously review the costs of delivering their HIV tions themselves. When considering mixed and generalized responses with a focus on both the unit costs of delivering epidemics, the geographical distribution of the HIV epidemic core HIV services and on the costs associated with 9 Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25097 http://onlinelibrary.wiley.com/doi/10.1002/jia2.25097/full | https://doi.org/10.1002/jia2.25097 management, human resources, administration, enablers, sup- (a) port and synergies. With regard to unit costs of service delivery, we found large heterogeneities both between countries and within countries. Across the 11 studies completed in the Eastern Europe & Central Asia region, there was significant variation in the unit costs of ART, FSW programmes, MSM programmes, opiate substitution therapy programmes, and needle-syringe pro- grammes [45]. A separate study found large variation in the unit costs of delivering needle-syringe programmes, opiate substitution therapy, and ART across 52 sites in three oblasts in Ukraine [46]. It can be difficult to distinguish intrinsic heterogeneity from inefficiency, but benchmarking exercises are the first step in identifying potential mechanisms for streamlined service delivery. It is also difficult to make a cross-country comparison of expenditure on management and other supporting pro- (b) grammes, as methods for accounting for these costs vary sig- nificantly across countries and not all costs were included in all studies. It was not within the scope of these studies to identify ways for strategic purchasing (or commissioning) for reducing cost components of HIV responses, but many of the stakeholders involved in the studies indicated that cost reduc- tions were desired and potentially feasible. 3.5 | The optimal programme mix varies by resource availability In 16 of the 23 studies, the optimal distribution of the total HIV budget under different funding constraints was estimated. Across these studies, it was found that when very little money is available, the optimal strategy is to focus on funding fewer Figure 2. Allocations of HIV budgets prior to Optima HIV study programmes in order to take advantage of economies of scale, (left bars), the mathematically optimal allocation recommended by rather than continuing to fund the full mix of programmes at the Optima HIV analysis (middle bars) and the allocation that was lower levels. As more resources become available, the next adopted by the country after the budgeting process was complete most cost-effective programmes should be introduced and (right bars). (a) Sudan (b) Belarus. Note that in Sudan, the total bud- then scaled up. This was consistently the case across all 16 get envelope was decreased from US$12.3 m to US$9.9 m. studies that contained this analysis (Table 2), and has been explored in greater detail in a separate publication [47]. The majority of countries around the world have set ambitious national targets for HIV reduction, yet have not invested or acquired close to the level of resources for direct HIV pro- 3.6 | More resources are required to achieve grammes necessary to realistically achieve these targets, even national targets if their resources were invested in the best possible mix of In 13 of the 23 studies, the minimal level of investment programmes. It may be possible to free up more funds for required to achieve the epidemiological targets described in core HIV services by improving the overall technical efficiency the country’s national strategic plan was estimated. In all but of the HIV response – for example, via the integration of HIV one case (Macedonia), the amount being invested in the HIV services into primary care, or by leveraging regional-level response at the time that the study was conducted was esti- negotiating power to bring down drug procurement costs – mated to be insufficient. The modelling results indicated that but even after exploiting all possible gains from technical and budget increases up to 185% would be required to attain the allocative efficiency improvements, it is almost certainly still targets within the strategic plan timeframes (Table 2). This is the case that additional funds will be required [48]. consistent with estimates published elsewhere of the resources required to achieve global HIV targets [48]. Note 3.7 | Adoption of model recommendations that these resource estimates pertain to the targets contained in the national strategic plan that was in place at the time that Allocative efficiency studies are most useful when conducted the study was conducted, which may since have changed. A prior to the budget- or target-setting process, so that they can detailed description of the particular targets is contained in help inform health-related targets and determine the funding each of the reports. envelopes and allocations commensurate with these targets. Knowledge of the funding environment and the likely The studies conducted in Sudan and Belarus represent two amount of resources that will be made available for HIV is an examples from the suite of studies considered in this review essential component of planning an effective HIV response. where the timing meant that the studies’ recommendations 10 Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25097 http://onlinelibrary.wiley.com/doi/10.1002/jia2.25097/full | https://doi.org/10.1002/jia2.25097 could be taken into account in the budget-setting process, and developed by CCK, RMS, DJK, AJS, RTG and DPW, with substantial technical both countries ultimately shifted their HIV budget allocations inputs from CJB, NF-H, DJW, MG, EM, EM, NC and ZS. closer towards what the optimization analyses recommended (Figure 2). Most other countries have used the results of the ACKNOWLEDGEMENTS Optima studies to inform their planning processes, resource The authors are grateful for the collaborative efforts of the teams who worked allocations or programmatic priorities. We hope to generate evi- on the allocative efficiency studies conducted in all 23 countries discussed in dence of these examples where modelling has been useful to this review. Those activities were conducted separately to this research project improve disease control strategies. but discussions in the country planning activities were valuable in highlighting distinct principles of relevance to other settings globally which created some Analyses such as these are just one step in the process of motivation for this research study. We value the leadership of our country part- bringing about optimal resource allocation and maximum health ners in ongoing relationships and Optima modelling activities to guide HIV outcomes; the real challenge lies in mobilizing funding and response planning and decision making in each country. potentially changing the nature or type of programmes that pol- icymakers implement. This can be challenging due to the multi- FUNDING tude of funding sources and the large proportion of HIV funding Australian National Health and Medical Research Council; the World Bank. that is dictated by external funding agendas and allocation crite- O Keiser was supported by a professorship grant from the Swiss National ria. Policy recommendations will be most useful when they are Science Foundation (grant no 163878). J Panovska-Griffiths was supported by accompanied with an operational plan, supplemented with tech- the National Institute for Health Research (NIHR) Collaboration for Leadership in Applied Health Research and Care North Thames at Bart’s Health NHS Trust nical support, which sets out a clear pathway and implementa- (NIHR CLAHRC North Thames). The views expressed in this article are those of tion details. Whilst priority setting is a difficult process, it will the authors. become increasingly important as countries move toward uni- versal health coverage and will need to make decisions based on objectives, health system users and cost-effectiveness. REFERENCES 1. World Health Organization. Public financing for health in Africa: from Abuja to the SDGs. Lancet Glob Health. 2016;4(10):e663. 4 | CONCLUSIONS 2. The HIV Modelling Consortium, Allocative efficiency tools & methods to support country HIV programme budget allocation: methods meeting summary report. 2015. 3. Wang H, Wolock TM, Carter A, Nguyen G, Kyu HH, Gakidou E, et al. Esti- Our study collates the findings of 23 country analyses that mates of global, regional, and national incidence, prevalence, and mortality of modelled how to optimally allocate the distribution of funds HIV, 1980-2015: the Global Burden of Disease Study 2015. The Lancet HIV. across HIV programme areas in real-world budget allocation 2016;3(8):e361–87. decision-making. According to the results produced in these 4. Dieleman JL, Campbell M, Chapin A, Eldrenkamp E, Fan VY, Haakenstad A, et al. Future and potential spending on health 2015-40: development assistance country studies, it would be possible to achieve substantial for health, and government, prepaid private, and out-of-pocket health spending (5% to 30%) reductions in infections and deaths through bet- in 184 countries. The Lancet. 2017;389(10083):2005–30. ter targeting of existing resources. The findings also under- 5. Dieleman J, Campbell M, Chapin A, Eldrenkamp E, Fan VY, Haakenstad A, score the benefits of additional resources targeted towards et al. Evolution and patterns of global health financing 1995-2014: development assistance for health, and government, prepaid private, and out-of-pocket health HIV prevention and care. Having tools such as Optima HIV to spending in 184 countries. The Lancet. 2017;389(10083):1981–2004. assist country decision-makers and their partners has value in 6. The Henry J. Kaiser Family Foundation, Financing the Response to HIV in improving the HIV planning and budgeting process. Low- and Middle-Income Countries: International assistance from donor Governments in 2015. 2016. 7. Caunhye AM, Nie X, Pokharel S. Optimization models in emergency logistics: AUTHORS’ AFFILIATIONS a literature review. Socio-Econ Plann Sci. 2012;46:4–13. 1 Department of Mathematical Sciences, University of Copenhagen, Copenhagen, 8. Earnshaw SR, Richter A, Sorensen SW, Hoerger TJ, Hicks KA, Engelgau M, Denmark; 2Burnet Institute, Melbourne, VIC, Australia; 3Institute for Global et al. Optimal allocation of resources across four interventions for type 2 dia- Health, University College London, London, UK; 4Clinical Operational Research betes. Med Decis Making. 2002;22(5 Suppl):S80–91. Unit, Department of Mathematics, University College London, London, UK; 9. Chen MM, Bush JW. Maximizing health system output with political and 5 Department of Applied Health Research, University College London, London, administrative constraints using mathematical programming. Inquiry. 1976;13 UK; 6Department of Global Health and Development, Faculty of Public Health (3):215–27. and Policy, London School of Hygiene & Tropical Medicine, London, UK; 7Insti- 10. Tao G, Zhao K, Gift T, Qiu F, Chen G. Using a resource allocation model to tute of Global Health, University of Geneva, Geneva, Switzerland; 8Institute of guide better local sexually transmitted disease control and prevention programs. Social and Preventive Medicine, University of Bern, Bern, Switzerland; 9Institute Oper Res Health Care 2012;1:23–9. of Mathematical Statistics and Actuarial Science, University of Bern, Bern, 11. Parker BR. A program selection/resource allocation model for control of Switzerland; 10Institute of Mathematics, University of Zurich, Zurich, Switzer- Malaria and related parasitic Diseases. Comput Oper Res. 1983;10(4):375–89. land; 11School of Public Health and Preventive Medicine, Monash University, 12. Revelle C, Feldmann F, Lynn W. An optimization model of tuberculosis epi- Melbourne, VIC, Australia; 12School of Physics, University of Sydney, Sydney, demiology. Manage Sci. 1969;16(4):B190–211. NSW, Australia; 13The Kirby Institute, UNSW Sydney, Sydney, NSW, Australia; 13. Feldstein MS, Piot MA, Sundaresan TK. Resource allocation model for pub- 14 The World Bank Group, Washington, DC, USA lic health planning. A case study of tuberculosis control. Bull World Health Organ. 1973;48(Suppl):5–110. 14. Heidenberger K. Strategic investment in preventive health care: quantita- COMPETING INTERESTS tive modelling for programme selection and resource allocation. OR Spektrum. The authors declare no competing interests. 1996;18:1–14. 15. Mbah Ndeffo ML, Gilligan CA. Resource allocation for epidemic control in metapopulations. PLoS ONE. 2011;6(9):e24577. AUTHORS’ CONTRIBUTIONS 16. Stinnett AA, Paltiel AD. Mathematical programming for the efficient alloca- RMS, LG and DPW wrote the manuscript. Analyses were carried out by HH-B, tion of health care resources. J Health Econ. 1996;15(5):641–53. JS, JP-G, LG, OK, JE, ZB, SLK, IR, DJK, AJS, JP, SAH, KLG, RTG, XFY, RM-H, 17. Flessa S. Where efficiency saves lives: a linear programme for the optimal CJB, NF-H, EM, ZS, NC, RMS and CCK. Guidance was provided by MG, EM, allocation of health care resources in developing countries. Health Care Manag NC and ZS. The Optima HIV software (used to conduct analyses) was primarily Sci. 2000;3(3):249–67. 11 Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25097 http://onlinelibrary.wiley.com/doi/10.1002/jia2.25097/full | https://doi.org/10.1002/jia2.25097 18. Flessa S. Priorities and allocation of health care resources in developing 36. World Bank. Value for Money in Ukraine’s HIV Response: Strategic Invest- countries: A case-study from the Mtwara region Tanzania. Eur J Oper Res. ment and Improved Efficiency. Washington, DC: World Bank; 2016 [cited 2018 2003;150:67–80. Mar 27]; Available from: https://openknowledge.worldbank.org/handle/10986/ 19. Chalabi Z, Epstein D, McKenna C, Claxton K. Uncertainty and value of 24967 inforamtion when allocating resources within and between healthcare pro- 37. UNDP, Modelling an optimised investment approach for Uzbekistan. 2014. grammes. Eur J Oper Res 2008;191:530–9. [cited 2018 Mar 27]; Available from: http://www.eurasia.undp.org/content/dam/ 20. Blumstein JF. The Oregon experiment: the role of cost-benefit analysis in rbec/docs/UNDP%20Modelling%20Uzbekistan_web_V2.pdf. the allocation of Medicaid funds. Soc Sci Med. 1997;45(4):545–54. 38. World Bank. Zambias HIV Response: Prioritised and Strategic Allocation of 21. Gray RT, Reyes J, Nadjib M, Harimurti P, Wilson DP. Assessment of the allocation HIV Resources for Impact and Sustainability. Washington, DC: World Bank; of HIV funding in Indonesia. Washington, DC: World Bank; 2012 [cited 2018 Mar 2015. [cited 2018 Mar 27]; Available from: https://openknowledge.worldbank. 27]; Available from: https://openknowledge.worldbank.org/handle/10986/29500. org/handle/10986/22080 22. Zhang L, Pham QD, Do MH, Kerr C, Wilson DP. Returns on investments of 39. World Bank. HIV Investment in Co ^te d’Ivoire: Optimized Allocation of HIV HIV prevention in Vietnam. Washington, DC: World Bank; 2013. [cited 2018 Mar Resources for a Sustainable and Efficient HIV Response. Washington, DC: World 27]; Available from: https://openknowledge.worldbank.org/handle/10986/29501. Bank; 2016 [cited 2018 Mar 27]; Available from: https://openknowledge. 23. World Bank, Optimizacio n en VIH/SIDA en Argentina. 2017 n de la inversio worldbank.org/handle/10986/29349 License: CC BY 3.0 IGO. [cited 2018 Mar 27]; Available from: http://documents.worldbank.org/curated/ 40. Fraser N, Cheikh N, Haacker M, Masaki E, Karamoko D, Gorgens M, et al. en/615041521181023433/Optimizaci%C3%B3n-de-lainversi%C3%B3n-en-VIH- Niger’s HIV response: targeted investments for a healthy future. Washington, SIDA-en-Argentina DC: World Bank; 2014 [cited 2018 Mar 27]; Available from: https://openknowl 24. The World Bank, Optimizando la Inversio n del Plan de Respuesta de edge.worldbank.org/handle/10986/20600 Colombia al VIH. 2018. 41. World Bank. Optimizing Investments for a Sustainable and Efficient HIV 25. World Bank, Optimizing investments in the national HIV response of Mex- Response in Senegal: Findings from an HIV Allocative Efficiency Study. Washing- ico, 2017. ton, DC: World Bank; 2016 [cited 2018 Mar 27]; Available from: https://open 26. World Bank, Optimizacio n de las inversiones para la respuesta al HIV en knowledge.worldbank.org/handle/10986/29383 Peru; 2017 [cited 2018 Mar 27]; Available from: http://documents.worldbank. 42. Fraser N, Benedikt C, Obst M, Masaki E, Go €rgens M, Stuart R, et al. org/curated/en/200381521184384616/Optimizaci%C3%B3n-de-las-inversione Sudan’s HIV Response: Value for Money in a Low-Level HIV Epidemic. Washing- spara-la-respuesta-al-VIH-en-Per%C3%BA. ton, DC: World Bank; 2014 [cited 2018 Mar 27]; Available from: https://open 27. World Bank. Optimizing HIV investments in Armenia. Washington, DC: knowledge.worldbank.org/handle/10986/21104 World Bank; 2015 Available from: https://openknowledge.worldbank.org/handle/ 43. World Bank, Optimizing investments in Togo’s HIV response. 2018. 10986/25397 44. Kerr CC, Stuart RM, Gray RT, Shattock AJ, Fraser-Hurt N, Benedikt C, 28. World Bank. Optimizing investments in Belarus for the national HIV et al. Optima: a model for HIV epidemic analysis, program prioritization, and response. Washington, DC: World Bank; 2016 [cited 2018 Mar 27]; Available resource optimization. J Acquir Immune Defic Syndr. 2015;69(3):365–76. from: https://openknowledge.worldbank.org/handle/10986/25395 45. Benedikt C, Kelly SL, Wilson D, Wilson DP. Allocative and implementation 29. World Bank. Optimizing investments in Bulgaria’s HIV response. Washing- efficiency in HIV prevention and treatment for people who inject drugs. Int J ton, DC: World Bank; 2018. [cited 2018 Mar 27]; Available from: https://open Drug Policy. 2016;38:73–80. knowledge.worldbank.org/handle/10986/29502 46. World Bank, Ukraine HIV program efficiency study: can Ukraine improve 30. World Bank. Optimizing investments in Georgia’s HIV response. Washing- value for money in HIV service delivery? 2014. ton, DC: World Bank; 2016. [cited 2018 Mar 27]; Available from: https://open 47. Stuart RM, Kerr CC, Haghparast-Bidgoli H, Estill J, Grobicki L, Baranczuk knowledge.worldbank.org/handle/10986/24966 Z, et al. Getting it right when budgets are tight: using optimal expansion path- 31. World Bank. Optimizing Investments in Kazakhstan’s HIV Response. Wash- ways to prioritize responses to concentrated and mixed HIV epidemics. PLoS ington, DC: World Bank; 2016 [cited 2018 Mar 27]; Available from: https://ope ONE. 2017;12(10):e0185077. nknowledge.worldbank.org/handle/10986/24965 48. Stover J, Bollinger L, Izazola JA, Loures L, DeLay P, Ghys PD. What is 32. World Bank. Optimizing Investments in the Kyrgyz Republic’s HIV required to end the AIDS epidemic as a public health threat by 2030? The cost Response. Washington, DC: World Bank; 2016 [cited 2018 Mar 27]; Available and impact of the fast-track approach. PLoS ONE. 2016;11(5):e0154893. from: https://openknowledge.worldbank.org/handle/10986/25377 33. World Bank. Optimizing Investments in Former Yugoslav Republic of Mace- donia’s HIV Response. Washington, DC: World Bank; 2015 [cited 2018 Mar 27]; Available from: https://openknowledge.worldbank.org/handle/10986/25378 34. World Bank. Optimizing Investments in Moldova’s HIV Response. Washing- SUPPORTING INFORMATION ton, DC: World Bank. 2015 [cited 2018 Mar 27]; Available from: https://open knowledge.worldbank.org/handle/10986/25396 Additional Supporting Information may be found in the online 35. UNDP, Modelling an optimised investment approach for Tajikistan. 2014. [cited 2018 Mar 27]; Available from: http://www.eurasia.undp.org/content/dam/ version of this article: rbec/docs/UNDP%20Modelling%20Tajikistan_English.pdf Table S1: Full list of Optima HIV studies.