Maximising the effect of combination HIV prevention through prioritisation of the people and places in greatest need: a modelling study

Transcription

1 Maximising the effect of combination HIV prevention through prioritisation of the people and places in greatest need: a modelling study Sarah-Jane Anderson, Peter Cherutich, Nduku Kilonzo, Ide Cremin, Daniela Fecht, Davies Kimanga, Malayah Harper, Ruth Laibon Masha, Prince Bahati Ngongo, William Maina, Mark Dybul, Timothy B Hallett Summary Background Epidemiological data show substantial variation in the risk of HIV infection between communities within African countries. We hypothesised that focusing appropriate interventions on geographies and key populations at high risk of HIV infection could improve the effect of investments in the HIV response. Methods With use of Kenya as a case study, we developed a mathematical model that described the spatiotemporal evolution of the HIV epidemic and that incorporated the demographic, behavioural, and programmatic differences across subnational units. Modelled interventions (male circumcision, behaviour change communication, early antiretoviral therapy, and pre-exposure prophylaxis) could be provided to different population groups according to their risk behaviours or their location. For a given national budget, we compared the effect of a uniform intervention strategy, in which the same complement of interventions is provided across the country, with a focused strategy that tailors the set of interventions and amount of resources allocated to the local epidemiological conditions. Findings A uniformly distributed combination of HIV prevention interventions could reduce the total number of new HIV infections by 4% during a 15-year period. With no additional spending, this effect could be increased by 14% during the 15 years almost 1 extra infections, and result in 33% fewer new HIV infections occurring every year by the end of the period if the focused approach is used to tailor resource allocation to reflect patterns in local epidemiology. The cumulative difference in new infections during the 15-year projection period depends on total budget and costs of interventions, and could be as great as 15 (a cumulative difference as great as 22%) under different assumptions about the unit costs of intervention. Interpretation The focused approach achieves greater effect than the uniform approach despite exactly the same investment. Through prioritisation of the people and locations at greatest risk of infection, and adaption of the interventions to reflect the local epidemiological context, the focused approach could substantially increase the efficiency and effectiveness of investments in HIV prevention. Funding The Bill & Melinda Gates Foundation and UNAIDS. Introduction HIV/AIDS remains one of the leading causes of death and disability in much of eastern and southern Africa. 1,2 Fortunately, several important scientific advances in recent years have provided a set of interventions that are highly effective in reducing the risk of acquisition (male circumcision, pre-exposure prophylaxis [PrEP]) and trans mission (early initiation of antiretroviral therapy [ART]). 3,4 The UNAIDS Investment framework 5 and the US President s Emergency Plan s Blueprint to achieve an AIDS-free generation 6 highlight the need to prioritise effective interventions for populations that are at the greatest risk of acquiring and transmitting HIV. The Global Fund to Fight AIDS, Tuberculosis and Malaria has a new strategy that focuses on effect and key affected populations. 7 However, evidence is needed to identify appropriate priority populations. Data from many large household surveys in Africa have recently shown crucial new insights about the highly uneven spread of HIV. For example, across eastern Africa (figure 1), there is remarkable heterogeneity in HIV prevalence, not only between countries but also within countries. HIV prevalence in the southwestern region of Kenya and the southern tip of Mozambique is as high as the national prevalence in South Africa, despite the country-wide prevalence estimate for Kenya being about two-thirds lower than that in South Africa and that for Mozambique being one-third lower. 8 1 Within South Africa, the estimated provincial HIV prevalence ranges from 9 2% in the Western Cape to 27 6% in KwaZulu-Natal. 11 At the same time, the prevalence among key population groups in extremely high. 12,13 Despite a national prevalence of 17 9% in South Africa, HIV prevalence in sex workers can be as high as 6%. 1,13,14 Across sub-saharan Africa, HIV prevalence in men who have sex with men (MSM) is estimated at 18%, substantially higher than the 5% regional adult prevalence reported by UNAIDS. 12 These observations Lancet 214; 384: See Editorial page 27 See Comment page 216 Department of Infectious Disease Epidemiology, Imperial College London, London, UK (S-J Anderson MSc, I Cremin PhD, Prof T B Hallett PhD); Small Area Health Statistics Unit (SAHSU), MRC-PHE Centre for Environment and Health, Imperial College London, London, UK (D Fecht PhD); National AIDS & STI Control Programme (NASCOP), Nairobi, Kenya (P Cherutich MBChB, D Kimanga MSc, W Maina MPH); Liverpool VCT, Care and Treatment, Nairobi, Kenya (N Kilonzo PhD); UNAIDS, Nairobi, Kenya (M Harper MSc (Eco), R Laibon Masha MPH); International AIDS Vaccine Initiative (IAVI), Nairobi, Kenya (P Bahati Ngongo MBA); and The Global Fund to Fight AIDS, Tuberculosis and Malaria, Geneva, Switzerland (M Dybul MD) Correspondence to: Sarah-Jane Anderson, Department of Infectious Disease Epidemiology, Imperial College London, London W2 1PG, UK sarah-jane.anderson@imperial. ac.uk Vol 384 July 19,

2 Articles For the Demographic and Health Surveys see See Online for appendix belie earlier assumptions that all individuals would be at high risk of acquisition of infection in generalised epidemics,15,16 and suggest that it is imperative that we move beyond the notion of national epidemics and identify the populations and geographies in greatest need of intervention. Kenya has a wealth of subnational data that shows the diversity of the epidemic across the country. We use those data to investigate the potential gains in the efficiency and effectiveness of investments focused on the people and places at highest risk, rather than a uniform, national approach. Sudan Somalia Central African Republic South Sudan Uganda Kenya DR Congo Rwanda Burundi Tanzania 5 1 km No data Malawi Zambia Mozambique Madagascar Zimbabwe Botswana Swaziland South Africa Lesotho % HIV prevalence >3% Figure 1: Spatial distribution of the HIV epidemic An interpolated map of HIV prevalence taken from the DHS Program Demographic and Health Surveys (conducted between 26 and 212) across eastern Africa. A geostatistical approach to mapping, ordinary kriging, was used to interpolate between observed HIV prevalence at sample cluster locations. 25 Study design We aimed to examine how a fixed amount of resources for HIV prevention can be used to generate reductions in the rate of new HIV infections using two forms of resource allocation. In the first, the rollout of particular interventions is uniform across the country. In the second, interventions can be focused on geographic or key affected populations that contribute to HIV strongholds. This analysis requires three steps: (1) specification of the spatiotemporal course of the HIV epidemic in Kenya; (2) assessment of the effect and cost of uniform, national implementation of interventions; and (3) assessment of the optimum form of intervention when constructing programmes specific to each locality (county or city). The methods for each part are discussed in turn and further details provided in the appendix. Specification of the spatiotemporal course of the HIV epidemic in Kenya Ethiopia Methods A dynamic mathematical model was used to represent the epidemic in every county or large city (in the case of Nairobi, Mombasa, and Kisumu) in Kenya. Building upon our previous work,17,18 the model represents the spread of HIV through sexual contact and tracks disease progression in HIV-positive individuals. Information about the scale-up of male circumcision and of ART in every county was extracted from official reports.19 Information about the proportion of men and women that have casual partners and that buy or sell sex are assessed from national survey data.2 The proportion of MSM was informed from review of the literature and an extensive national mapping of key populations.21 The model was calibrated to these data and also to the available prevalence data, with the timetrend based upon the data from antenatal clinics and levels of prevalence informed by the most recent nationally representative household survey.2,22 For both sources of prevalence data, spatial interpolation techniques were used to generate local-level estimates.23,24 The model includes four forms of intervention (table) and assumed that these can be provided to four population groups independently; female sex workers (FSW), other women, MSM, and other men. The table shows the key assumptions regarding the efficacy, cost, and coverage of these interventions. In the absence of compelling evidence to the contrary, we assume homogeneity in the individual-level efficacy of PrEP and early ART across the risk groups. The model is used to estimate the effect (defined as the number of infections averted during a 15-year period from the start of the intervention in 215) and cost (approximated here by constant unit costs multiplied by the corresponding modelled estimate of consumption) for every possible permutation of interventions and population group in a county or city. Vol 384 July 19, 214

3 Efficacy assumption Coverage assumption Range of unit cost considered* Default unit cost value* Voluntary medical male circumcision Behaviour change communication Early ART (ART initiated at CD4 cell count >35 cells per ml) PrEP Risk of acquisition for circumcised men 6% 8% of eligible men $3 12 per one man less than for uncircumcised men 25,26 circumcised 2% reduction in risk of acquisition for every low-risk person reached, 5% reduction in risk of acquisition for high-risk groups (MSM and FSW) 5 85% reduction in risk of onward transmission for a person on ART relative to others 27 75% reduction in risk of acquisition for a person on PrEP 28 1% of the relevant population group 33% in low-risk women and heterosexual men, 66% in FSW and MSM 25% in low-risk women and heterosexual men, 5% in FSW and MSM $5 8 per person per year reach $25 1 per person on ART per year $1 1 per person on PrEP per year $6 5 $1 in low-risk women and heterosexual men, $2 in FSW and MSM 5 $515 $25 28 All interventions can be applied to all relevant groups; with behaviour change interventions, early ART, and PrEP for FSW, other women, MSM and other men, and voluntary medical male circumcision available to all men. In every case, these estimates were chosen to be consistent with the assumptions used in the UNAIDS Investment Framework analysis. 5 MSM=men who have sex with men. FSW=female sex workers. ART=antiretroval therapy. PrEP=pre-exposure prophylaxis. *Neither effect nor costs were discounted in the analysis. Table: Interventions and assumed unit costs A B Number of new infections 213 >4 and 213 >213 and 59 >59 and 139 >139 and 28 >28 and 4552 Figure 2: Modelled new infections by counties in Kenya for 213 (A) An untransformed map of the counties of Kenya; the colours correspond to the number of modelled new infections for 213. (B) A cartogram of modelled new infections by county in 213. The cartogram is a distorted map; with the area of the counties transformed to be proportional to the number of new infections modelled to occur in that county whilst preserving the topology. Those counties with a high number of new infections (shown as red and pink on the map) have expanded to dominate the figure. The ScapeToad application was used to create this plot, which used the Gastner and Newman 24 diffusion-based algorithm for the transformation. Assessment of the effect and cost of a uniform approach Under the best circumstances for decision making at country level, the relative costs and effects of different forms of intervention when applied uniformly to the whole country are compared. We identified the optimum intervention strategy for a given total cost through construction of health production functions. 29 The health production function is created through plotting the cost and effect of all possible candidate strategies. We then remove those strategies that are subject to dominance (ie, provide less health return at a greater cost than an alternative strategy). Successive points that fall on the frontier (ie, are nondominated) are then joined together by straight-line segments to form a piece-wise linear function. We call this the uniform health production function because decisions are made for the whole country. For the ScapeToad application see index.php Vol 384 July 19,

4 A Infections averted (thousands) Assessment of effect and cost of a focused approach An alternative approach to programming would be to determine the allocation between localities by judging where those resources could be used to the greatest effect. Under this focused approach, counties and major cities are not constrained to receive the same set of interventions; instead interventions can be tailored to the Group 3 local epidemiological setting. To simulate this approach, we first constructed local-level health production functions (within the same method as above) to identify those sets of interventions that maximise health returns for a given total cost with a location. We used these locallevel health production functions to inform the allocation of resources between locations as follows: we first define a small increment in the budget for the whole country and for each locality and quantify the health returns that could be gained through the use of that increment using the local-level health production function. The potential health gains across localities are then compared and the increment allocated to the locality that can produce the greatest health gains (ie, infections averted). These steps are repeated until a maximum value for the budget in the whole country is reached. In this way, the national budget is allocated between locations and populations on the basis of where it will have most effect. We term this approach the focused health production function because it represents the health returns that can be achieved across the whole country when programmes can be implemented that are tailored to specific populations at the local level (counties and major cities). B Cost (US$1 million) Medical male circumcision Behaviour change intervention for sex workers Behaviour change intervention for women Behaviour change intervention for men Behaviour change intervention for MSM Early ART for sex workers Early ART for women Early ART for men Early ART for MSM PrEP for sex workers PrEP for women PrEP for men PrEP for MSM Medical male circumcision Behaviour change intervention for sex workers Behaviour change intervention for women Behaviour change intervention for men Behaviour change intervention for MSM Early ART for sex workers Early ART for women Early ART for men Early ART for MSM PrEP for sex workers PrEP for women PrEP for men PrEP for MSM Group 2 Group 1 Uniform approach Implemented Partially implemented Not implemented Location (county or city) in order of ascending HIV prevalence Focused approach Location (county or city) in order of ascending HIV prevalence Figure 3: Intervention expansion pathway using the uniform and focused approaches (A) Local level health production functions. Each curve displays the maximum possible effect (number of infections averted, ) at a given cost, for each county or city. The health production functions are plotted on the same axes to show the difference in scale and shape seen across the country. The health production functions were grouped according to the maximum effect possible in every location for ease of reading. The locations in the group are as follows: group 1 (maximum effect <1 infections averted): Lamu, Garissa, Isiolo, Wajir, Tana River, West Pokot, Turkana, Samburu, Marsabit, Laikipia, Tharaka, Kwale, Baringo, Machakos, Embu, Kitui, Nyeri, Busia, Kajiado, Kirinyaga, Mombasa, Uasin Gishu, Taita Taveta, Kilifi, Nyandarua, Elgeyo-Marakwet, Mandera, Nyamira, Meru, Nandi; group 2 (maximum effect >1 and <2 infections averted): Murang a, Makueni, Nakuru, Bomet, Bungoma, Trans Nzoia, Kiambu, Narok; group 3 (maximum effect >2 infections averted): Kisumu City, Vihiga, Kisumu county, Siaya, Kakamega, Nairobi, Homa Bay, Kisii, Migori, Kericho. (B) Plots to show the pattern of implementation of the constituent interventions by location under both the uniform strategy and focused strategy at a budget US$6 million over the 15-year intervention period. We construct the focused and uniform approaches from a total of 13 possible constituent intervention strategies; each strategy referring to a different intervention method (male circumcision, behaviour change, early ART, and PrEP), which can be applied to the different population groups as appropriate (female sex workers, other women, men who have sex with men, other men). In each panel, each row represents one of these 13 constituent interventions and each column represents a county or city. The colour code indicates whether an intervention strategy is used in a given location. Yellow indicates an intervention is implemented, green indicates that it is not implemented, and light green indicates it is implemented but not at full coverage. In the uniform approach, the same set of interventions is used in all locations, as the allocative decisions are made for Kenya as a whole. In the focused approach, different interventions can be used in different locations, and allocation of interventions between locations is driven by the achievable impact an intervention can have and where. Although patterns of interventions are broadly conserved between locations, the choice by location is no longer restricted to be identical as in the uniform approach. Under the focused approach, generally those locations which have a higher HIV prevalence (to the right hand side) will receive a fuller complement of interventions. All of the interventions are specified such that the characteristics of the intervention strategy are the same (coverage, efficacy and unit cost) throughout this analysis (table). The difference between the approaches is whether or not they are implemented by location. MSM=men who have sex with men. ART=antiretroviral therapy. PrEP=pre-exposure prophylaxis Vol 384 July 19, 214

5 New infections per year (thousands) A No interventions Uniform approach Focused approach Year Infections averted (thousands) B Cost (US$ million) Proportion of simulations C to <28 28 to <42 42 to <56 56 to <7 7 to <84 84 to <98 98 to < to < to <14 14 to <155 Additional number of infections averted under the focused approach (thousands) Figure 4: Difference in health gains between the uniform and focused approaches (A) Annual number of new infections with time (in thousands) using the uniform approach (green line), the focused approach (red line), and a baseline scenario of no additional interventions (blue line). (B) Health production functions for the uniform (green line) and focused approach (red line), across a range of total budgets for the interventions. The vertical difference between the lines is the additional HIV infections averted with the focused approach compared with the uniform approach. Vertical axis gives the HIV infections averted (in thousands) for and the horizontal axis give the corresponding total cost in that period. Note that horizontal axis is on a log scale. (C) Histogram to display the results of a sensitivity analysis assessing how the difference in impact between the focused and uniform approaches changes under different assumptions of the unit costs of the component interventions. The simulation and analysis were repeated using different unit cost assumptions for the component interventions drawn at random from the cost ranges shown in the table. The additional number of infections averted through using the focused approach as opposed to the uniform approach was calculated for each simulation. The results are displayed in the histogram, with the horixontal axis corresponding to the number of additional infections averted under the focused approach as opposed to the uniform approach, and the vertical axis the proportion of simulations which resulted in additional infections averted in that category. In this way the histogram displays the distribution of the additional impact achievable using the focused approach compared to the uniform approach under the range of plausible unit cost assumptions. This approach will be compared with the uniform health production function. The appendix provides further details of the methods used and intervention-related assumptions. Role of the funding source The funder had no role in the data collection, data analysis, or data interpretation; writing of the report, or in the decision to submit the paper for publication. S-JA and TBH had full access to all the data and made the decision to submit the paper for publication. Results The model was calibrated to epidemiological data from Kenya. Across the 47 counties in Kenya, estimated HIV prevalence varies substantially, from less than 1% to 22% in the year 213. In the model, the estimated number of new infections is concentrated in five counties (Migori, Homa Bay, Kisii, Nairobi, and Kericho) that account for almost 4% of all new HIV infections in the country (figure 2A). The south-western counties dominate the epidemic (figure 2B). The first step in our analysis was to calculate the uniform health production function by allocating funds to particular combinations of interventions for the whole country. The uniform health production function is available in the appendix. The uniform health production function displays a familiar form, with substantial gains in health being achieved with small increments in cost at low overall total costs, as behaviour change communication and male circumcision programmes are implemented. This corresponds to the steep rise at the beginning of the curve, which is followed by a region of diminishing returns, whereby greater effect is possible by scaling up early ART and PrEP, but at a much greater cost. Next, we constructed the local-level health production functions (figure 3). Differences in the demography and epidemiology across the different counties and cities result in local-level health production functions that vary substantially in both shape and scale (figure 3A). The order in which additional interventions are added in every county or city as costs increase displays some broad consistency but also considerable variation. These differences are exploited in a focused approach for HIV prevention shown through a comparison of the expansion pathways in each location in the appendix. Figure 3B contrasts the configuration of interventions across localities under the uniform and focused forms of programming during the 15-year period, with a budget of US$6 million. In the uniform approach, if an intervention is implemented, it is scaled up across all localities. By contrast, under the focused approach, every county or city has its own mix of interventions. In many localities, programmes for male circumcision are not required. In some places but not others, all relevant interventions (early ART, PrEP, and behaviour change) are provided for FSW, whereas in others the focus is more on MSM. In some localities the total cost of programmes is far greater than that in others, because the high incidence of HIV infection could justify the widespread use of expensive interventions, such as early ART. There is a relation between the total costs allocated to each location and the number of persons living with HIV in that location (appendix), but this does not fully Vol 384 July 19,

6 Panel: Research in context Systematic review We searched PubMed for studies published up to Nov 2, 213, with the terms HIV and (priorit* OR target* OR focus*) and (spatial OR geog* OR subnational) and (model OR modelling OR modeling). Very few relevant modelling studies were identified. One model showed how male circumcision interventions could be distributed in South Africa, in light of variation in HIV prevalence. 3 However, this study examined only one intervention and did not consider the dynamic infectious transmission of HIV. Early studies that considered the epidemic at different geographical scales suggested that the effect of the epidemic on population structure is most visible at fine resolution. 31 The need for data at finer geographical scales to allow for better estimates of the geographical burden of infection has been emphasised. 32 Studies have noted the apparent inadequacy of models that do not capture substantial within-country epidemiological variation. 33 The use of transmission models of subnational units has been proposed in the evaluation of interventions, 34 but we have not identified further examples of such models in prospective programming. Other studies only described the spatial distribution of infection, HIV-related mortality, or service availability, or looked for factors associated with prevalence. Considerable variation in prevalence within countries and local-level clustering of the burden of infection has been observed Several studies have examined the spatial distribution of health-care services, in particular antiretroviral therapy and sexual and reproductive health, and assessed implications for access to care for the surrounding populations Additional studies identified in the search provided real world examples of a focused approach being deployed programmatically. In particular, India has used epidemiological indicators to classify districts into categories that partly determine their programme design. 44 South Africa has also set targets for care provision at the local level. 45 In Mozambique, health services are tailored for key populations in specific high-risk locations. 46 Furthermore, the need to identify high-risk populations at key localities, such as those who work in high-risk sectors in mapped geographies, has been emphasised. 47 Interpretation To our knowledge, this is the first study to integrate spatial analyses, transmission dynamic modelling of HIV, and economic evaluation to examine the optimum configuration of the full set of HIV prevention interventions available. Despite all of the elements of a spatially structured modelling analysis present in the literature, we found no examples of integration of these methods for intervention planning. We have shown that a focused approach, which uses data for local epidemiology to allocate prevention interventions optimally across locations, is able to achieve greater effect than a uniform approach for the same amount of investment. This study provides an example of how local epidemiological data can be used to improve prevention programmes for HIV, and uses available information about geographical heterogeneity in spatially structured modelling analyses. explain the variation in allocated costs, because the scope of the potential HIV prevention effect and the cost of the most appropriate intervention also have an effect. Figure 4 summarises the difference in the total health gains between the uniform and focused approaches. The difference in the projected number of new infections in the two strategies, at a budget of $6 million, is considerable and increases with time (figure 4A). Figure 4B shows the difference in effect between the uniform and focused approaches over a range of total costs. At very high total cost, there is no difference between the strategies, because almost all interventions are deployed in all populations. However, at lower total costs, there can be very large differences in the overall health gains. For a cost of $6 million between 215 and 229, the total additional health gain for the focused approach is almost 1 extra infections averted during the 15-year period, a 14% increase. Further, 33% fewer new HIV infections would occur in the year 229 under the focused approach than under the uniform approach, equivalent to 8 fewer infections in that year. Repeat analysis with different assumptions for the unit cost of interventions (from within the plausible cost ranges given in the table) at the same total budget (table) shows that the additional infections averted under the focused approach compared with the uniform approach might be as high as 15 during the 15-year period (figure 4C). Discussion We compared two approaches to tackle a large HIV epidemic and noted that an investment strategy that focuses on epidemiological strongholds of HIV could yield substantially greater effect than a strategy that allocates resources in a uniform way across a country. To our knowledge, this is the first study to integrate spatial analyses, transmission dynamic modelling of HIV, and economic evaluation to inform the configuration of available interventions for planning (panel). Prioritising interventions to the people and places with high rates of HIV infections in this way makes intuitive sense. Focusing resources to those at greatest risk has long been a central theme in guidance issued by WHO and UNAIDS, with emphasis on detailed collection of epidemiological data and beginning prevention pro gramming with a know your epidemic assess ment. 48 The strategic direction of resources is a fundamental principle in the UNAIDS investment frame work. 5 Recognition is increasing of the potential to leverage geographical variation in epidemics for programme planning. 49 Although prioritisation is applied in countries it is unclear at what geographical level such decisions are made and there has been a lack of unified methods to guide national level resource allocation decisions utilising subnational epidemiological data. Focusing on hot spots is standard practice for infections such as malaria and schistosomiasis and, to a lesser extent, tuberculosis To our knowledge, this is the first report to synthesise epidemiological data from a country, including data for high-risk groups in specific locations, to estimate the value of a focused approach for HIV. Data from many countries with low and middle income, 55,56 and also high-income countries, 57,58 show substantial variation in HIV rates by geography and by key affected populations, making this approach widely applicable. Administrative regions (counties) are the main unit of analysis in this study because they correspond to both the lowest level of resource allocation decisions and the finest resolution of available input data. We did not consider transmission between modelled locations Vol 384 July 19, 214

7 (representing cross-county migration), which could reduce the value of differentiating interventions between geographies; although for Kenya, every county is large (average size 12 km² with a population of about 8 people). 59 Fine-scale targeting to particular venues within an area can be informed by other approaches, such as the successful PLACE method. 6 Further data will be essential to refine this analysis, in particular the spatial variation in unit cost: for instance, the cost of services in remote areas could be much higher than that in central areas. 61 The analysis could be expanded to assess the benefit of updating the configuration of the interventions through time, which might further enhance efficiency. Several crucial points must be considered in assessment of feasibility of applying such a strategy in practice. The political and practical issues of focusing resources in certain areas of a country, in particular in countries with complex sociocultural variation, must not be underestimated. Extremely high rates of HIV could be masked if data are aggregated to regional level in an area with otherwise very low prevalence. Collection of accurate data for key affected populations both across and within areas of each country is important. The issue of key affected populations, often marginalised and even criminalised in many countries, poses unique challenges. Cultural norms around highly effective interventions, for example voluntary male circumcision and condoms, can also pose obstacles. Although substantial progress has been made towards preventing the transmission of HIV, recent scientific advances afford the opportunity to drive rates to lower levels. A uniform strategy, which does not use available intelligence on the epidemic, will fail to be as effective as a strategy that does. By use of a public health approach that focuses resources based on an epidemiological understanding of subnational geographical areas and key affected populations, and selects the package of interventions most likely to have an effect according to the drivers of each HIV stronghold, the efficiency and effectiveness of programming could be greatly increased. However, increased effectiveness and efficiency does not mean fewer resources are needed, at least in the near term, but rather that so-called smarter spending can generate an even greater effect. Contributors TBH, S-JA, and MD conceived the study. TBH and S-JA developed the methods and did the analyses. IC assisted in parameter estimation and DF provided additional maps and advice on methodology. PC, NK, DK, MH, RLM, PBN, and WM provided data used to inform the parameter estimates and advised on analysis and interpretation. TBH, MD, and S-JA wrote the first draft of the report. All authors reviewed and approved the final version. Declaration of interests TBH reports grants and personal fees from the Bill & Melinda Gates Foundation during the conduct of the study; grants and personal fees from World Bank, UNAIDS, and The Rush Foundation; personal fees from the University of Washington, New York University, Children s Investment Fund Foundation, and Global Fund outside of the submitted work. Other authors declare no competing interests. Acknowledgments Funding for this study was provided by the Bill & Melinda Gates Foundation and UNAIDS. References 1 Lozano R, Naghavi M, Foreman K, et al. Global and regional mortality from 235 causes of death for 2 age groups in 199 and 21: a systematic analysis for the Global Burden of Disease Study 21. Lancet 212; 38: Ortblad KF, Lozano R, Murray CJ. The burden of HIV: insights from the Global Burden of Disease Study 21. AIDS 213; 27: Jones A, Cremin I, Abdullah F, et al. Transformation of HIV from pandemic to low-endemic levels: a public health approach to combination prevention. Lancet 214; published online April Piot P, Quinn TC. Response to the AIDS pandemic a global health model. N Engl J Med 213; 368: Schwartländer B, Stover J, Hallett T, et al. Towards an improved investment approach for an effective response to HIV/AIDS. Lancet 211; 377: The United States President s Emergency Plan for AIDS Relief (PEPFAR). PEPFAR Blueprint: creating an AIDS-free generation. The Office of the Global AIDS Coordinator, pepfar.gov/documents/organization/21386.pdf (accessed July 2, 214). 7 The Global Fund. New funding model theglobalfund.org/en/about/grantmanagement/fundingmodel/ (accessed July 2, 214). 8 Joint United Nations Programme on HIV/AIDS (UNAIDS). Kenya 213. Available from: countries/kenya/ (accessed July 2, 214). 9 Joint United Nations Programme on HIV/AIDS (UNAIDS). Mozambique 213. Available from: regionscountries/countries/mozambique/ (accessed July 2, 214). 1 Joint United Nations Programme on HIV/AIDS (UNAIDS). South Africa 213. Available from: regionscountries/countries/southafrica/ (accessed July 2, 214). 11 Shisana O. HIV/AIDS in South Africa: at last the glass is half full. 6th SA AIDS Conference; Durban, South Africa; June (accessed July 2, 214). 12 Beyrer C, Baral SD, van Griensven F, et al. Global epidemiology of HIV infection in men who have sex with men. Lancet 212; 38: Baral S, Beyrer C, Muessig K, et al. Burden of HIV among female sex workers in low-income and middle-income countries: a systematic review and meta-analysis. Lancet Infect Dis 212; 12: van Loggerenberg F, Mlisana K, Williamson C et al. Establishing a cohort at high risk of HIV infection in South Africa: challenges and experiences of the CAPRISA 2 acute infection study. PLoS One 28; 3: e Sumartojo E, Carey J, Doll L, Gayle H. Targeted and general population interventions for HIV prevention: towards a comprehensive approach. AIDS 1997; 11: UNAIDS/WHO Working Group on Global HIV/AIDS and STI Surveillance. Guidelines for second generation HIV surveillance, 2. edc_2_5.pdf (accessed July 2, 214). 17 Hallett T, Gregson S, Mugurungi O, Gonese E, Garnett G. Assessing evidence for behaviour change affecting the course of HIV epidemics: a new mathematical modelling approach and application to data from Zimbabwe. Epidemics 29; 1: Hallett TB, Singh K, Smith JA, White RG, Abu-Raddad LJ, Garnett GP. Understanding the impact of male circumcision interventions on the spread of HIV in Southern Africa. PLoS One 28; 3: e Kenya National AIDS and STI Control Program (NASCOP). Kenya, County HIV Service Delivery Profiles, org/wp-content/uploads/213/9/the-final-county-profiles- REPORT-FOR-KENYA-revised_2.pdf (accessed July 2, 214). 2 Kenya National Bureau of Statistics (KNBS) and ICF Macro. Kenya Demographic and Health Survey Calverton, Maryland: KNBS_21_Kenya_DHS_29_final_report.pdf (accessed July 2, 214). Vol 384 July 19,

8 21 Kenya National AIDS and STI Control Programme (NASCOP), National AIDS Control Council (NACC). Geographic Mapping of Most at Risk Populations for HIV (MARPs) in Kenya, Kenya National AIDS and STI Control Programme (NASCOP). Sentinel surveillance for HIV and Syphilis infection among pregnant women from antenatal clinics in Kenya. Nairobi, Kenya: surveillance%2reports%2for%2211.pdf (accessed July 2, 214). 23 Shepard D. A two dimensional interpolation function for irregularly-spaced data. Proceedings of the 1968 ACM National Conference; New York, NY; Aug 27 29, Larmarange J, Vallo R, Yaro S, Msellati P, Méda N. Methods for mapping regional trends of HIV prevalence from Demographic and Health Surveys (DHS). Cybergeo: European Journal of Geography, Systèmes, Modélisation, Géostatistiques, document 558, le 26 octobre 211, consulté le 1 juillet org/2466 (accessed July 2, 214). 25 Auvert B, Taljaard D, Lagarde E, Sobngwi-Tambekou J, Sitta R, Puren A. Randomized, controlled intervention trial of male circumcision for reduction of HIV infection risk: the ANRS 1265 trial. PLoS Med 25; 2: e Bailey RC, Moses S, Parker CB, et al. Male circumcision for HIV prevention in young men in Kisumu, Kenya: a randomised controlled trial. Lancet 27; 369: Cohen MS, Chen YQ, McCauley M, et al. Prevention of HIV-1 infection with early antiretroviral therapy. N Engl J Med 211; 365: Cremin I, Alsallaq R, Dybul M, Piot P, Garnett G, Hallett TB. The new role of antiretrovirals in combination HIV prevention: a mathematical modelling analysis. AIDS 213; 27: Drummond M, Sculpher M, Torrance G, O Brien B, Stoddard G. Methods for Economic Evaluation of Health Care Programmes. Oxford: Oxford University Press, Verguet S. Efficient and equitable HIV prevention: a case study of male circumcision in South Africa. Cost Eff Resour Alloc 213; 11: Low-Beer D, Stoneburner RL, Mukulu A. Empirical evidence for the severe but localized impact of AIDS on population structure. Nat Med 1997; 3: Williams BG, Gouws E, Abdool Karim SS. Where are we now? Where are we going? The demographic impact of HIV/AIDS in South Africa. S Afr J Sci 2; 96: Mishra S, Sgaier SK, Thompson LH, et al. HIV epidemic appraisals for assisting in the design of effective prevention programmes: shifting the paradigm back to basics. PLoS One 212; 7: e Chandrasekaran P, Dallabetta G, Loo V, et al. Evaluation design for large-scale HIV prevention programmes: the case of Avahan, the India AIDS initiative. AIDS 28; 22: Qian HZ, Taylor RD, Fawal HJ, Vermund SH. Increasing AIDS case reports in the South: U.S. trends from AIDS Care 26; 18: Kandala NB, Campbell EK, Rakgoasi SD, Madi-Segwagwe BC, Fako TT. The geography of HIV/AIDS prevalence rates in Botswana. HIV AIDS (Auckl) 212; 4: Musenge E, Vounatsou P, Collinson M, Tollman S, Kahn K. The contribution of spatial analysis to understanding HIV/TB mortality in children: a structural equation modelling approach. Glob Health Action 213; 6: Tanser F, Bärnighausen T, Cooke GS, Newell M-L. Localized spatial clustering of HIV infections in a widely disseminated rural South African epidemic. Int J Epidemiol 29; 38: Moran D, Jordaan JA. HIV/AIDS in Russia: determinants of regional prevalence. Int J Health Geogr 27; 6: Law DC, Serre ML, Christakos G, Leone PA, Miller WC. Spatial analysis and mapping of sexually transmitted diseases to optimise intervention and prevention strategies. Sex Transm Infect 24; 8: Cooke GS, Tanser FC, Barnighausen TW, Newell ML. Population uptake of antiretroviral treatment through primary care in rural South Africa. BMC Public Health 21; 1: Wilson DP, Blower SM. Designing equitable antiretroviral allocation strategies in resource-constrained countries. PLoS Med 25; 2: Yao J, Murray AT, Agadjanian V. A geographical perspective on access to sexual and reproductive health care for women in rural Africa. Soc Sci Med 213; 96: Sgaier SK, Claeson M, Gilks C, et al. Knowing your HIV/AIDS epidemic and tailoring an effective response: how did India do it? Sex Transm Infect 212; 88: Leydon NG, Venter F, Webster PD, Moleko W, Osih R, Barker PM. Achieving the HIV and AIDS National Strategic Plan: a practical calculator for local target setting in district health facilities. S Afr Med J; 1: Lafort Y, Geelhoed D, Cumba L, et al. Reproductive health services for populations at high risk of HIV: performance of a night clinic in Tete province, Mozambique. BMC Health Serv Res 21; 1: Halli SS, Buzdugan R, Ramesh BM, et al. Assessing HIV risk in workplaces for prioritizing HIV preventive interventions in Karnataka State, India. Sex Transm Dis 29; 36: WHO. Guidelines for second generation HIV surveillance: an update: Know your epidemic. Geneva: World Health Organization, Joint United Nations Programme on HIV/AIDS (UNAIDS). Location, location. Connecting people faster to HIV services. Geneva: LocationLocation_EN_EMBARGOED_19%2Nov-Noon%2(3).pdf (accessed July 2, 214). 5 Soares Magalhães RJ, Biritwum NK, et al. Mapping helminth co-infection and co-intensity: geostatistical prediction in Ghana. PLoS Negl Trop Dis 211; 5: e Kabatereine N, Standley C, Sousa-Figueiredo J, et al. Integrated prevalence mapping of schistosomiasis, soil-transmitted helminthiasis and malaria in lakeside and island communities in Lake Victoria, Uganda. Parasit Vectors 211; 4: Bousema T, Griffin JT, Sauerwein RW, et al. Hitting hotspots: spatial targeting of malaria for control and elimination. PLoS Med 212; 9: e Cotter C, Sturrock HJ, Hsiang MS, et al. The changing epidemiology of malaria elimination: new strategies for new challenges. Lancet 213; 382: Dowdy DW, Golub JE, Chaisson RE, Saraceni V. Heterogeneity in tuberculosis transmission and the role of geographic hotspots in propagating epidemics. Proc Natl Acad Sci USA 212; 19: National AIDS Control Organisation (NACO). HIV Sentinel Surveillance A Technical Brief, aidsdatahub.org/sites/default/files/documents/hss_21-11_ Technical_Brief_3_Nov_12.pdf (accessed July 2, 214). 56 Tanzania Commission for HIV and AIDS. United Republic of Tanzania. UNGASS reporting for 21 (Tanzania mainland and Zanzibar). country_progress_report_en.pdf (accessed July 2, 214). 57 Health Protection Agency. HIV in the United Kingdom: 212 Report. London: Health Protection Services, webc/hpawebfile/hpaweb_c/ (accessed July 2, 214). 58 US Centers for Disease Control and Prevention. HIV and AIDS in the United States by geographic distribution gov/hiv/statistics/basics/geographicdistribution.html (accessed July 2, 214). 59 Kenya National Bureau of Statistics. GEOHIVE: Kenya. geohive.com/cntry/kenya.aspx (accessed July 2, 214). 6 Weir SS, Boerma T. From people to places: overview of the PLACE method and lessons learned. MEASURE Evaluation Bulletin 22 Nov 4, 22: /at_download/document (accessed July 2, 214). 61 Meyer-Rath G, Over M. HIV treatment as prevention: modelling the cost of antiretroviral treatment state of the art and future directions. PLoS Med 212; 9: e Vol 384 July 19, 214