SummaryBackground

High-resolution estimates of HIV burden across space and time provide an important tool for tracking and monitoring the progress of prevention and control efforts and assist with improving the precision and efficiency of targeting efforts. We aimed to assess HIV incidence and HIV mortality for all second-level administrative units across sub-Saharan Africa.

Methods

In this modelling study, we developed a framework that used the geographically specific HIV prevalence data collected in seroprevalence surveys and antenatal care clinics to train a model that estimates HIV incidence and mortality among individuals aged 15–49 years. We used a model-based geostatistical framework to estimate HIV prevalence at the second administrative level in 44 countries in sub-Saharan Africa for 2000–18 and sought data on the number of individuals on antiretroviral therapy (ART) by second-level administrative unit. We then modified the Estimation and Projection Package (EPP) to use these HIV prevalence and treatment estimates to estimate HIV incidence and mortality by second-level administrative unit.

Findings

The estimates suggest substantial variation in HIV incidence and mortality rates both between and within countries in sub-Saharan Africa, with 15 countries having a ten-times or greater difference in estimated HIV incidence between the second-level administrative units with the lowest and highest estimated incidence levels. Across all 44 countries in 2018, HIV incidence ranged from 2·8 (95% uncertainty interval 2·1–3·8) in Mauritania to 1585·9 (1369·4–1824·8) cases per 100 000 people in Lesotho and HIV mortality ranged from 0·8 (0·7–0·9) in Mauritania to 676·5 (513·6–888·0) deaths per 100 000 people in Lesotho. Variation in both incidence and mortality was substantially greater at the subnational level than at the national level and the highest estimated rates were accordingly higher. Among second-level administrative units, Guijá District, Gaza Province, Mozambique, had the highest estimated HIV incidence (4661·7 [2544·8–8120·3]) cases per 100 000 people in 2018 and Inhassunge District, Zambezia Province, Mozambique, had the highest estimated HIV mortality rate (1163·0 [679·0–1866·8]) deaths per 100 000 people. Further, the rate of reduction in HIV incidence and mortality from 2000 to 2018, as well as the ratio of new infections to the number of people living with HIV was highly variable. Although most second-level administrative units had declines in the number of new cases (3316 [81·1%] of 4087 units) and number of deaths (3325 [81·4%]), nearly all appeared well short of the targeted 75% reduction in new cases and deaths between 2010 and 2020.

Interpretation

Our estimates suggest that most second-level administrative units in sub-Saharan Africa are falling short of the targeted 75% reduction in new cases and deaths by 2020, which is further compounded by substantial within-country variability. These estimates will help decision makers and programme implementers expand access to ART and better target health resources to higher burden subnational areas.

Funding

Bill & Melinda Gates Foundation.

IntroductionAs the HIV pandemic enters its fifth decade, several indicators have been proposed to help describe the burden of HIV, measure the effectiveness of public health efforts, and guide decision making. Among the most useful and commonly cited indicators are the HIV incidence rate, the HIV mortality rate, the percentage reduction in the number of incident HIV cases and HIV deaths, and the ratio of incident HIV cases to people living with HIV.1Ghys PD Williams BG Over M Hallett TB Godfrey-Faussett P Epidemiological metrics and benchmarks for a transition in the HIV epidemic. The UN Political Declaration on HIV and AIDS calls for a 75% reduction in new HIV infections and HIV deaths from 2010 to 2020.2UN General Assembly
Political declaration on HIV and AIDS: on the fast track to accelerating the fight against HIV and to ending the AIDS epidemic by 2030. Studies have shown that geographical targeting of resources can improve the efficiency and effectiveness of interventions and strategies intended to address HIV.3Anderson S-J Cherutich P Kilonzo N et al.Maximising the effect of combination HIV prevention through prioritisation of the people and places in greatest need: a modelling study., 4McGillen JB Anderson S-J Dybul MR Hallett TB Optimum resource allocation to reduce HIV incidence across sub-Saharan Africa: a mathematical modelling study. To best aid resource targeting, HIV indicators need to be produced at a refined spatial scale. Yet, for countries in sub-Saharan Africa—those hardest hit by the HIV pandemic—collecting data on indicators of HIV incidence and mortality remains a challenge, particularly at spatially granular levels, where weaknesses in civil registration and vital statistics systems are typically beyond the capacity or funding for HIV programmes alone to address. Historically, most data collection systems for tracking HIV have focused on measuring HIV prevalence, partly because HIV prevalence is inherently more straightforward to measure than HIV incidence or mortality, and partly driven by programmatic needs including daily patient management, ensuring sufficient drug supply, and managing loss to follow-up. The Population-based HIV Impact Assessment survey series and other household surveys have attempted to include direct measures of HIV incidence via recency assays5Justman JE Mugurungi O El-Sadr WM HIV population surveys—bringing precision to the global response. and many countries are introducing routine recency testing for newly diagnosed individuals;6Kim AA Behel S Northbrook S Parekh BS Tracking with recency assays to control the epidemic: real-time HIV surveillance and public health response. however, these data are not yet as widespread as data related to HIV prevalence and concerns remain regarding the validity and reliability of recency assays for accurately estimating HIV incidence.7Kassanjee R Pilcher CD Keating SM et al.Independent assessment of candidate HIV incidence assays on specimens in the CEPHIA repository.Research in context

Evidence before this study

We searched PubMed with no language restrictions for articles published since database inception until Dec 31, 2020, using the following search terms: “hiv[MeSH] AND (“mortality” OR “incidence” OR “prevalence”) AND “subnational” AND (trend*)”. Previous research has shown that substantial local (spatial) variation exists in HIV incidence, and modelling studies comparing geographically targeted with non-geographically targeted prevention strategies have suggested that geographically targeted strategies are more efficient in preventing new HIV infections under the same budgetary constraints. Trends in HIV mortality and incidence have varied at both regional and country levels, resulting in differing trends in HIV prevalence, and this dynamic is further complicated by the paucity of directly observed empirical data on HIV incidence and mortality in sub-Saharan Africa and other high-burden low-income and middle-income countries. Renewed commitment is required to assess progress towards global targets at a subnational scale, to ensure no sub-populations are left behind, and to support sub-Saharan Africa in getting on track to bring HIV infection under control by 2030.

Added value of this study

Although many initiatives provide national estimates for HIV metrics (and at the administrative level in some countries), there are few HIV incidence and mortality estimates and necessary methodological innovation at more detailed subnational scales. This study suggests substantial variation exists in HIV incidence and mortality rates both between and within countries in sub-Saharan Africa, with highly variable rates of reduction in HIV incidence and mortality and the ratio of new infections to the number of people living with HIV from 2000 to 2018. Although most second-level administrative units had declines in the number of new cases and attributable deaths, nearly all appeared well short of the targeted 75% reduction in new cases and deaths between 2010 and 2020.

Implications of all the available evidence

By improving and extending existing HIV incidence and mortality estimates in sub-Saharan Africa at a subnational scale, this study provides valuable estimates to help gauge progress towards ending the HIV epidemic by 2030 (Sustainable Development Goal 3) and provides an important tool to improve the precision and efficiency of targeting interventions within countries.

Because trends in HIV incidence and mortality are largely not directly observed at the national level in sub-Saharan Africa, estimates are developed by fitting mathematical models to data on trends in HIV prevalence. The Estimation and Projection Package (EPP),8Ghys PD Brown T Grassly NC et al.The UNAIDS Estimation and Projection Package: a software package to estimate and project national HIV epidemics. developed by UNAIDS and also used by the Global Burden of Disease (GBD) study,9Frank TD Carter A Jahagirdar D et al.Global, regional, and national incidence, prevalence, and mortality of HIV, 1980–2017, and forecasts to 2030, for 195 countries and territories: a systematic analysis for the Global Burden of Diseases, Injuries, and Risk Factors Study 2017., 10Vos T Lim SS Abbafati C et al.Global burden of 369 diseases and injuries in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019. provides a well-tested structure for leveraging the HIV prevalence data available from population surveys and antenatal care sentinel surveillance sites to estimate HIV incidence and mortality. UNAIDS, the President's Emergency Plan for AIDS Relief, and others have called for incorporating local data and estimates into country HIV response strategies, given subnational heterogeneity in the HIV epidemic. Although subnational estimates of HIV prevalence and antiretroviral therapy (ART) coverage are increasingly common, to our knowledge, estimates of HIV incidence and mortality are not yet routinely available below the first administrative level.Here, we present a modified version of the EPP model, which combines developments in spatial demography,11WorldPop
WorldPop dataset. fine-scale HIV prevalence mapping,12Dwyer-Lindgren L Cork MA Sligar A et al.Mapping HIV prevalence in sub-Saharan Africa between 2000 and 2017. and HIV pandemic modelling9Frank TD Carter A Jahagirdar D et al.Global, regional, and national incidence, prevalence, and mortality of HIV, 1980–2017, and forecasts to 2030, for 195 countries and territories: a systematic analysis for the Global Burden of Diseases, Injuries, and Risk Factors Study 2017., 10Vos T Lim SS Abbafati C et al.Global burden of 369 diseases and injuries in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019., 13Eaton JW Brown T Puckett R et al.The Estimation and Projection Package Age-Sex Model and the r-hybrid model: new tools for estimating HIV incidence trends in sub-Saharan Africa. to produce estimates of HIV incidence and mortality for first-level (eg, provinces) and second-level administrative units (eg, districts) across 44 sub-Saharan African countries. Estimates of these indicators at this fine spatial scale can assist in tracking and accelerating progress towards meeting the Sustainable Development Goal of “ending the AIDS pandemic as a public health threat by 2030”.14UN
Goal 3: ensure healthy lives and promote well-being for all at all ages. United Nations Sustainable Development Goals.Methods Study designOur study follows the Guidelines for Accurate and Transparent Health Estimates Reporting (appendix pp 18, 19).15Stevens GA Alkema L Black RE et al.Guidelines for Accurate and Transparent Health Estimates Reporting: the GATHER statement. We used a modified version of the EPP mathematical compartmental model, tailored specifically to estimate HIV incidence and mortality, among individuals aged 15–49 years for first-level and second-level administrative units in 44 countries in sub-Saharan Africa. EPP fits the transmission rate to the prevalence using a series of assumptions about how the different epidemic indicators relate to each other within a given population. This fitting is achieved using the Bayesian incremental mixture importance sampling solver, which aligns the HIV prevalence output of each EPP simulation to the HIV prevalence from our model-based geostatistical prevalence model. Analyses were done using the R (version 3.6.1) and C (version GNU99) programming languages. Further details of the methods, input data types and sources, and assumptions are provided in the appendix (pp 20–30). Modelling strategyRather than develop a methodology de novo, we sought a tested modelling framework that (1) could leverage HIV prevalence data to estimate HIV incidence and mortality; (2) could use available demographic data without requiring calibration of many behavioural parameters that are rarely available at subnational resolutions; and (3) had a history of producing reliable and widely used results. EPP8Ghys PD Brown T Grassly NC et al.The UNAIDS Estimation and Projection Package: a software package to estimate and project national HIV epidemics. meets these needs and this system has been used by both UNAIDS13Eaton JW Brown T Puckett R et al.The Estimation and Projection Package Age-Sex Model and the r-hybrid model: new tools for estimating HIV incidence trends in sub-Saharan Africa., 16Niu X Zhang A Brown T Puckett R Mahy M Bao L Incorporation of hierarchical structure into estimation and projection package fitting with examples of estimating subnational HIV/AIDS dynamics. and GBD9Frank TD Carter A Jahagirdar D et al.Global, regional, and national incidence, prevalence, and mortality of HIV, 1980–2017, and forecasts to 2030, for 195 countries and territories: a systematic analysis for the Global Burden of Diseases, Injuries, and Risk Factors Study 2017., 10Vos T Lim SS Abbafati C et al.Global burden of 369 diseases and injuries in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019. to produce national and first-administrative level estimates of HIV incidence and mortality across sub-Saharan Africa.EPP is a compartmental epidemiological model, which is designed to estimate the HIV incidence and mortality rates needed to produce a time trend specific to HIV prevalence. The model functions by varying model parameters to identify HIV incidence trends that are most consistent with observed HIV prevalence, given the number of patients on ART (appendix pp 20–30). The version of EPP that we present here is a modified version of that used to create the GBD 2017 HIV estimates9Frank TD Carter A Jahagirdar D et al.Global, regional, and national incidence, prevalence, and mortality of HIV, 1980–2017, and forecasts to 2030, for 195 countries and territories: a systematic analysis for the Global Burden of Diseases, Injuries, and Risk Factors Study 2017. and operates on individuals aged 15–49 years of both sexes as one intermixing population. Developments in EPP13Eaton JW Brown T Puckett R et al.The Estimation and Projection Package Age-Sex Model and the r-hybrid model: new tools for estimating HIV incidence trends in sub-Saharan Africa. have allowed UNAIDS and GBD to use an age-specific and sex-specific version of EPP (EPP-ASM) for national models and, in some cases, at the first administrative level. However, use of that model was not feasible for our spatially granular implementation because of the additional computational burden of fitting this more complex model and because EPP-ASM benefits from age-specific and sex-specific HIV prevalence estimates, which are not yet widely available for second-level administrative units. We modified the GBD 2017 version of EPP to use the HIV prevalence time series produced using a model-based geostatistical framework,12Dwyer-Lindgren L Cork MA Sligar A et al.Mapping HIV prevalence in sub-Saharan Africa between 2000 and 2017. as opposed to direct survey and antenatal care estimates of HIV prevalence, which are typically used in EPP. Because internal migration between second-level administrative units is potentially important for our model but difficult to measure, we adopted the approach now being used by UNAIDS and the GBD in the EPP-ASM model to adjust the population at the end of each timepoint to the expected population in the next timepoint using a simple scalar. This method removes the need to explicitly model migration and instead relies on the population count estimates in each administrative unit. Our last major modification to EPP was to implement the r-hybrid model employed by UNAIDS in 2018 and GBD in 2019, which has been shown to work best for most geographical areas.13Eaton JW Brown T Puckett R et al.The Estimation and Projection Package Age-Sex Model and the r-hybrid model: new tools for estimating HIV incidence trends in sub-Saharan Africa. In brief, r-hybrid estimates the HIV transmission rate differently in the early versus later phases of the HIV epidemic to better match observed data. Model inputs

The model has five key inputs: (1) the boundaries (or shapes) used to define the second-level administrative units we are modelling; (2) the size of the population aged 15–49 years over time that we used as the demographic bases for the hypothetical epidemics; (3) the modelled HIV prevalence in each of the second-level administrative units; (4) the number of people on ART in each second-level administrative unit; and (5) the assumptions used about how likely a person living with HIV is to die from their infection.

To delineate the boundaries of the second-level administrative units we began with the second-level administrative shapefiles that are publicly available from the Database of Global Administrative Areas. These boundaries were modified to correct for known errors and to accommodate recent boundary changes. A full list of changes and the naming convention for first-level and second-level administrative units across the 44 countries in sub-Saharan Africa can be found in the appendix (pp 49–51).To estimate populations in second-level administrative units, we used high-resolution gridded population estimates that were age specific and sex specific from WorldPop.11WorldPop
WorldPop dataset. To create a full time series from 1970 to 2020, we interpolated additional years of data using the population growth rate at the pixel level observed in the WorldPop dataset, assuming exponential growth, and scaled the total population in each country to match GBD national population estimates. Finally, these gridded population estimates were aggregated fractionally to the shapefiles as described, to create second-level administrative unit population estimates.We used an updated version of the model-based geostatistical methodology from our previous work12Dwyer-Lindgren L Cork MA Sligar A et al.Mapping HIV prevalence in sub-Saharan Africa between 2000 and 2017. to produce HIV prevalence estimates for all second-level administrative units across sub-Saharan Africa from 1995 to 2018. This year range was chosen because we were able to extract geolocated sentinel surveillance data for antenatal care and household survey estimates of HIV prevalence over this period. A full list of HIV prevalence data incorporated into this analysis can be found in the appendix (pp 52–101). In summary, 145 surveys (80 surveys with microdata, 28 survey reports, and 37 surveys extracted from published literature) and 134 sources of antenatal care sentinel surveillance data, which in combination resulted in a geopositioned data set of 29 072 survey observations and 11 710 site-years of antenatal care sentinel surveillance, formed the input for the HIV prevalence component.We used several data sources to estimate the number of individuals on ART in each second-level administrative unit. The UNAIDS annual estimate files17UNAIDS
National HIV estimates file. UNAIDS Spectrum EPP. provide the number of adults receiving ART in each country but this information is not sufficiently granular to use at the second administrative unit level. Therefore, we did a systematic data-seeking exercise to extract all available subnational ART data in the 44 countries included in this analysis and successfully identified subnational-level ART information in 29 countries (appendix pp 201–03). In countries where subnational ART information was available, we modelled these data to create a full time series (appendix pp 27–29). In the 15 countries where we were unable to locate subnational ART information, we assumed that the national ART coverage rate was consistent in all second-level administrative units and redistributed ART patients accordingly.HIV mortality rates were calculated for individuals with HIV of varying disease severity separately for those who were and were not receiving ART.10Vos T Lim SS Abbafati C et al.Global burden of 369 diseases and injuries in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019. EPP divides the population with HIV into seven CD4 cell count categories as a proxy for disease severity and into two treatment categories (on or off ART). We tracked progression through the CD4 categories and onto treatment so that at every time step of the EPP model, we had an estimate of the size of the population for each disease severity and treatment category (appendix p 30). We then applied the estimated HIV mortality rate from GBD that is specific to each CD4 and treatment category to the population in these groups to estimate HIV mortality. Effect of ARTART is a key input to this model because the treatment substantially decreases viraemia and thus the probability that an individual with HIV will die from or pass on their infection;18Cohen MS Chen YQ McCauley M et al.Antiretroviral therapy for the prevention of HIV-1 transmission. thus, ART fundamentally changes the relationship between HIV incidence, prevalence, and mortality. We were not able to identify and extract subnational ART information in all countries, and so, to assess the effect of using these data, in the 29 countries where we were able to extract subnational ART data we ran EPP using both the extracted subnational ART data and assuming the national ART coverage rate. We then compared the two scenarios to ascertain the effect of ART on our models. As expected, ART coverage in each second-level administrative unit had a substantial effect on the HIV incidence and mortality estimate in that location (appendix pp 31, 32). Subnational variation in HIV incidence and mortality, however, was still present in countries where we assumed the national ART coverage for all second-level administrative units, driven by variation in the level and trend of HIV prevalence. Figures in the appendix (pp 33–36) show the relationship between estimated HIV incidence, mortality, prevalence, and ART coverage. Uncertainty interval estimationTo account for uncertainty in our model inputs, including the disease progression and mortality parameters, we ran 100 simulations of EPP varying these parameters. Within each simulation, we generated 1000 draws from the approximated posterior distribution of HIV prevalence, HIV incidence, and HIV mortality. To create a single, combined posterior distribution we sampled ten draws from each of the 100 simulations and then combined these draws. For consistency with national-level estimates using much of the same underlying data, we calibrated our results to match the estimates from the GBD 2019.10Vos T Lim SS Abbafati C et al.Global burden of 369 diseases and injuries in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019. Further details are provided in the appendix (pp 24, 25).

To account for uncertainty in our estimates of HIV incidence and mortality when assessing progress towards achievement of the UNAIDS target of a 75% reduction in new HIV infections and HIV deaths, we calculated the posterior probability of achieving these targets as the percentage of draws from the estimated posterior distribution where these targets were achieved.

 Role of the funding source

The funders of this study had no role in the study design, data collection, data analysis, data interpretation, or writing of the report.

ResultsWe found marked regional differences in HIV incidence and mortality among individuals aged 15–49 years from Jan 1, 2000, to Dec 31, 2018. Across the entire modelled region in 2018, the HIV incidence rate was 218·1 (95% uncertainty interval [UI] 196·4–239·1) cases per 100 000 people and the HIV mortality rate was 87·2 (76·6–101·1) deaths per 100 000 people. At the national level in 2018, HIV incidence ranged from 2·8 (2·1–3·8) cases per 100 000 people in Mauritania to 1585·9 (1369·4–1824·8) cases per 100 000 people in Lesotho (figure 1A; appendix p 37), and HIV mortality ranged from 0·8 (0·7–0·9) deaths per 100 000 people in Mauritania and 676·5 (513·6–888·0) deaths per 100 000 people in Lesotho (figure 2A; appendix p 39). The variation in both incidence and mortality was substantially greater at the subnational compared with the national level and the highest estimated rates were accordingly higher. The first-level administrative unit with the highest estimated HIV incidence rate in 2018 was Gaza Province in Mozambique, with an incidence rate of 2805·9 (2118·0–3611·2) cases per 100 000 people (figure 1B). Among second-level administrative units, Guijá District in Gaza Province, Mozambique, had the highest estimated HIV incidence, with 4661·7 (2544·8–8120·3) cases per 100 000 people in 2018 (figure 1C). Among second-level administrative units, Inhassunge District in Zambezia Province, Mozambique, had the highest HIV mortality rate estimate at 1163·0 (679·0–1866·8) deaths per 100 000 people (figure 2C).

Figure 1HIV incidence among individuals aged 15–49 years in sub-Saharan Africa in 2018

Show full caption

Incidence among individuals aged 15–49 years by (A) country, (B) first-level administrative unit, and (C) second-level administrative unit. Lakes and areas with fewer than ten people per 1 × 1 km and classified as barren or sparsely vegetated are coloured light grey. Areas in dark grey were not included in the analysis. Estimates in areas that are crossed are based on national, rather than subnational, estimates of antiretroviral therapy coverage only.

Figure 2HIV mortality among individuals aged 15–49 years in sub-Saharan Africa in 2018

Show full caption

(A) HIV mortality among individuals aged 15–49 years by country, (B) first-level administrative unit, and (C) second-level administrative unit. Lakes and areas with fewer than ten people per 1 × 1 km and classified as barren or sparsely vegetated are coloured light grey. Areas in dark grey were not included in the analysis. Estimates in areas that are crossed are based on national, rather than subnational, estimates of antiretroviral therapy coverage only.

In addition to large-scale variation across the region, we also found substantial within-country variation in HIV incidence and mortality. In 2018, 15 countries (Angola, Benin, Burkina Faso, Burundi, Cameroon, Democratic Republic of the Congo, Ethiopia, Kenya, Mozambique, Nigeria, Senegal, Somalia, Tanzania, Uganda, and Zambia) had a greater than ten-times difference in HIV incidence between the second-level administrative units with the lowest and highest estimated incidence levels. Of those 15 countries, 11 also had a greater than ten-times difference in HIV mortality rates between their lowest and highest second-level administrative units. Kenya was a particularly extreme example of this variability, with incidence rate estimates ranging from 14·2 (95% UI 4·1–41·3) cases per 100 000 people in Eldas Constituency, Wajir County, to 1767·0 (939·7–2957·9) cases per 100 000 people in Rarieda Constituency, Siaya County, and HIV mortality rate estimates ranging from 5·7 (2·4–15·2) deaths per 100 000 people in Eldas Constituency, Wajir County, to 789·5 (524·9–1165·1) in Suba Constituency, Homa Bay County, in 2018.

In absolute terms, incident HIV cases and HIV deaths were highly concentrated in high-population locations. In 2018, we estimated 1 138 827 (95% UI 1 025 447–1 248 270) incident HIV cases across the 44 modelled countries. 50% of these incident HIV cases in 2018 were located in just 148 (3·6%) of 4087 second-level administrative units that collectively represented 13·7% of the total population in this region (figure 3A). Most of these high-burden administrative units were located in southern sub-Saharan Africa; in particular, both Lesotho and South Africa had more than 50% of their second-level administrative units in this category. Conversely, 2630 (64·4%) of 4087 second-level administrative units, representing 38·2% of the total population, accounted for less than 10% of the total estimated incident HIV cases.

Figure 3Incident HIV cases and deaths among individuals aged 15–49 years in sub-Saharan Africa in 2018

Show full caption

(A) Number of incident HIV cases and (B) HIV deaths among individuals aged 15–49 years in 2018 by second-level administrative unit. Lakes and areas with fewer than ten people per 1 × 1 km and classified as barren or sparsely vegetated are coloured light grey. Areas in dark grey were not included in the analysis. Estimates in areas that are crossed are based on national, rather than subnational, estimates of antiretroviral therapy coverage only.

In 2018, we estimated that 455 244 (95% UI 399 851–527 712) HIV deaths took place in the 44 modelled countries. Only 224 (5·5%) of 4087 second-level administrative units, representing 22·3% of the total population, accounted for 50% of the estimated deaths (figure 3B). 2364 (57·8%) of 4087 second-level administrative units, representing 30·0% of the total population, contributed less than 10% of the total estimated HIV deaths in 2018.The UNAIDS fast-track goals,2UN General Assembly
Political declaration on HIV and AIDS: on the fast track to accelerating the fight against HIV and to ending the AIDS epidemic by 2030. which are designed to set measurable targets for public health action, call for a 90% reduction in both incident HIV cases and HIV deaths by 2030, as compared with 2010 levels. An additional intermediate target was set to a 75% reduction in both indicators by 2020. Across our modelled region, incident HIV cases reduced by 16·9% (95% UI 6·8–25·1) and the HIV incidence rate reduced by 33·8% (25·8–40·3) between 2010 and 2018, both falling well short of the UNAIDS intermediate goal to reduce HIV incidence by 75% by 2020. We estimated that no country has yet achieved the goal of a 75% reduction in new infections on a national scale (figure 4D). Our estimates show wide variability among subnational areas in progress towards achieving this goal. We estimated that only six (0·9%) of 686 first-level administrative units (figure 4B) and 64 (1·6%) of 4087 second-level administrative units (figure 4C) had already achieved a 75% reduction in incident HIV cases by 2018.

Figure 4Percentage reduction in incident HIV cases in sub-Saharan Africa from 2010 to 2018

Show full caption

(A) Reduction in the number of incident HIV cases (%) between 2010 and 2018 among individuals aged 15–49 years by country, (B) first-level administrative unit, and (C) second-level administrative unit. Lakes and areas with fewer than ten people per 1 × 1 km and classified as barren or sparsely vegetated are coloured light grey. Areas in dark grey were not included in the analysis. Estimates in areas that are crossed are based on national, rather than subnational, estimates of antiretroviral therapy coverage only. A 75% reduction in HIV incidence by 2020 is a UNAIDS fast-track goal. Progress towards this target by country highlighting the best and worst performing subnational units is shown in panel D.

Our estimates suggest that increases in the number of incident HIV cases are far too common. At the national level, Angola (61·2% [95% UI 49·2–73·9] increase), Equatorial Guinea (77·3% [52·1–103·8]), Guinea (14·3% [0·72–32·2])