2026 Data Sources
The data sources for facts and figures featured in the 2026 Goalkeepers Report are listed here by section. Brief methodological notes are included for unpublished analyses.
- Introduction
- Explore the Data
- Call to Action
- Impact
- Guest Essays
- Methodology for 2026 Goalkeepers Custom Analysis
- SDG Indicators
This report was developed by Gates Foundation staff and partners. AI tools were used to support some of the editing process. All content was led, reviewed, and approved by human writers, designers, editors, data experts, and program staff.
Introduction
NOTHING HAS FELT LIKE THIS
AI Adoption Speed
AI tools have reached global scale faster than any technology in history. ChatGPT reached 100 million monthly active users within approximately two months of its launch in late 2022, compared to approximately nine months for TikTok and two and a half years for Instagram. Earlier technologies spread more slowly still: it took decades for the personal computer to reach comparable levels of household penetration in the United States.
Hu, K. “ChatGPT Sets Record for Fastest-Growing User Base—analyst note,” Reuters, February 2, 2023. www.reuters.com/technology/chatgpt-sets-record-fastest-growing-user-base-analyst-note-2023-02-01.
Meta Platforms, Inc., SEC filings and historical company reporting, (2004–2008), Facebook user growth records. Facebook reached 100 million users in 2008, approximately four and a half years after its February 2004 launch.
Petska-Juliussen, K., Juliussen, E., The 8th Annual Computer Industry Almanac. Austin, TX: The Reference Press, 1996.
U.S. Department of Commerce. “The Emerging Digital Economy.” Washington, DC. 1998. www.cga.ct.gov/PS99/rpt%5Colr%5Chtm/99-R-0047.htm
GBD 2023 Causes of Death Collaborators. “Global Burden of 292 Causes of Death in 204 Countries and Territories and 660 Subnational Locations, 1990–2023: A Systematic Analysis for the Global Burden of Disease Study 2023.” The Lancet 406, no. 10513 (2025): 1811-1872. www.thelancet.com/journals/lancet/article/PIIS0140-6736(25)01637-X/fulltext.
United Nations Inter-agency Group for Child Mortality Estimation (UN IGME). Levels and Trends in Child Mortality 2025. March 2026. data.unicef.org/resources/levels-and-trends-in-child-mortality-2025/.
World Bank. Poverty and Inequality Platform. Version 20260324_2021. Data set. World Bank Group, 2025. pip.worldbank.org/.
Institute for Health Metrics and Evaluation (IHME). Financing Global Health 2025: Cuts in Aid and Future Outlook. Seattle, Washington. IHME, 2025. www.healthdata.org/research-analysis/library/financing-global-health-2025-cuts-aid-and-future-outlook
Apeagyei, A. E., Bisignano, C., Elliott, H., Hay, S.I., Lidral-Porter, B., Nam, S., Shyong, C., Tsakalos, G., Zlavog, B., Barış, E., Murray, C.J.L., and Dieleman, J.L. “Tracking Development Assistance for Health, 1990–2030: Historical Trends, Recent Cuts, and Outlook.” The Lancet 406, no. 10501 (2025): 337–348. doi.org/10.1016/S0140-6736(25)01240-1
Explore the Data
WE’VE MADE PROGRESS. BUT IF CURRENT TRENDS PERSIST, SO DOES INEQUALITY.
Institute for Health Metrics and Evaluation, (July 2026). [Custom analysis. Full methodology is detailed below].
Friedman, J., York H, Graetz N., et al. “Measuring and Forecasting Progress towards the Education-Related SDG Targets.” Nature 580, 636-639 (2020). doi.org/10.1038/s41586-020-2198-8
Institute for Health Metrics and Evaluation, (July 2026). [Custom analysis. Full methodology is detailed below].
Institute for Health Metrics and Evaluation, (July 2026). [Custom analysis. Full methodology is detailed below].
The data visualizations in this report are representations of the data. Full datasets are available upon request.
Impact
WHERE THE DIFFERENCE GETS MADE: A NURSE, A FARMER, A TEACHER
Institute for Health Metrics and Evaluation, (July 2026). [Custom analysis. Full methodology is detailed below].
In addition to the custom analysis, there is published research on the health workforce challenges in sub-Saharan Africa, including:
Ahmat, A., et al. “The Health Workforce Status in the WHO African Region: Findings of a Cross-Sectional Study.” BMJ Global Health 7, suppl. 1 (2022). pmc.ncbi.nlm.nih.gov/articles/PMC9109011/
GBD 2019 Human Resources for Health Collaborators. “Measuring the Availability of Human Resources for Health and Its Relationship to Universal Health Coverage for 204 Countries and Territories from 1990 to 2019: A Systematic Analysis for the Global Burden of Disease Study 2019.” The Lancet 399, no. 10341 (2022): 2129–2154. doi.org/10.1016/S0140-6736(22)00532-3
Call to Action
FIRST, AI TOOLS HAVE TO WORK IN EVERY LANGUAGE PEOPLE SPEAK.
Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, et al. “Language Models are Few-Shot Learners.” Advances in Neural Information Processing Systems, 33 (2020): 1877–1901. arxiv.org/abs/2005.14165
Wang, W., Tu, Z., Chen, C., Yuan, Y., Huang, J., Jiao, W., and Lyu, M. “All Languages Matter: On the Multilingual Safety of LLMs.” In Findings of the Association for Computational Linguistics: ACL 2024, 5865–5877. Bangkok, Thailand: Association for Computational Linguistics, 2024. aclanthology.org/2024.findings-acl.349/
Martínez, G., Conde, J., Merino-Gómez, E., Bermúdez-Margaretto, B., Hernández, J. A., Reviriego, P., and Brysbaert, M. “Establishing Vocabulary Tests as a Benchmark for Evaluating Large Language Models. PLOS ONE, 19, no. 12(2024): e0308259. doi.org/10.1371/journal.pone.0308259
Liu, Y., Huang, J., and Wang, H. Who on Earth is Using Generative AI? Global Trends and Shifts in 2025. Policy Research Working Paper 11231. Washington, DC: World Bank, 2025. openknowledge.worldbank.org/bitstreams/c5a7eb99-17c5-4d0e-a5ce-3d9fcfe184a7/download
Microsoft AI Economy Institute. Global AI Diffusion Report: 2025 H2. Microsoft, 2026. www.microsoft.com/en-us/research/wp-content/uploads/2026/01/Microsoft-AI-Diffusion-Report-2025-H2.pdf
The “one billion” and “7 billion” figures are illustrative, not precise measurements. The cited sources (World Bank, 2025; Microsoft AI Economy Institute, 2026) document a large and widening AI access gap by income and region.
Olatunji, T., Afonja, T., Yadavalli, A., Emezue, C.C., Singh, S., Dossou, B.F.P., Osuchukwu, J., Osei, S., Tonja, A.L., Etori, N., Mabataku, C. “AfriSpeech-200: Pan-African accented speech dataset for clinical and general domain ASR.” Transactions of the Association for Computational Linguistics, 11 (2023): 1669–1685. doi.org/10.1162/tacl_a_00627
Guest Essays
EARLY EVIDENCE SHOWS WHAT’S POSSIBLE.
The four examples from Kenya, Sierra Leone, the United States, and India were selected because they represent early evidence of what becomes possible when AI tools are built for frontline workers. They are not the only tools showing promise in these sectors. Many others exist.
The Gates Foundation has provided support to Penda Health, Kiddom, MahaVistaar, and to Fab AI the implementing partner for the Gemini Guided Learning trial in Sierra Leone.
Korom, R., Kiptinness, S., Adan, N., Said, K., Ithuli, C., Rotich, O., Kimani, B., King’ori, I., Kamau, S., Atemba, E., Aden, M., Bowman, P., Sharman, M., Soskin Hicks, R., Distler, R., Heidecke, J., Arora, R. K., and Singhal, K. “AI-Based Clinical Decision Support for Primary Care: A Real-World Study.” arXiv preprint, 2025. arxiv.org/abs/2507.16947
OpenAI. “Pioneering an AI Clinical Copilot with Penda Health.” July 22, 2025. www.openai.com/index/ai-clinical-copilot-penda-health.
Ghahramani, Z. “Measuring the Impact of Learning with AI in Sierra Leone and Beyond.” Google DeepMind / Fab AI, June 9, 2026. deepmind.google/blog/measuring-the-impact-of-learning-with-ai-in-sierra-leone-and-beyond/.
Manjee, A. “In NYC District, Technology Works With Pencil and Paper to Help Kids Learn Math,” The 74, July 9, 2026. www.the74million.org/article/in-nyc-district-technology-works-with-pencil-and-paper-to-help-kids-learn-math. Note: Author is co-founder of Kiddom.
Kiddom. “Kiddom Launches Atlas, the First AI-Powered Instructional Technology Layered on High-Quality Instructional Materials,” BusinessWire, February 23, 2026. www.businesswire.com/news/home/20260220676382/en/Kiddom-Launches-Atlas-the-First-AI-Powered-Instructional-Technology-Layered-on-High-Quality-Instructional-Materials.
Government of Maharashtra / OpenAgriNet (2025), MahaVISTAAR Program Monitoring Data [program monitoring data; available at: maharashtra.gov.in/ or upon request from Maharashtra Agriculture Department, India].
MahaVISTAAR launched through the Maharashtra state government’s agricultural digital infrastructure. Cost per farmer figure reflects government operational costs per active user as of late 2025.
Methodology for 2026 Goalkeepers Custom Analysis
WE’VE MADE PROGRESS. BUT IF CURRENT TRENDS PERSIST, SO DOES INEQUALITY.
Educational attainment
The schooling gap estimate draws on educational attainment by age cohort and region, modeled using the framework described in Friedman et al., 2020, which models within-country distributions of years of schooling, from which mean years of educational attainment is calculated.
To produce forecasts, for each country, we forecast the mean years of schooling (MYS) within ages 25 to 29 by modeling the year-to-year change in MYS as a function of the MYS level. When a location’s rate of change in 2023 was substantially faster or slower than the model prediction, we allowed the rate of change to gradually approach the predicted value over time before following the model’s assumed dynamics. Forecasts of educational attainment in other age groups were then estimated based on forecasts for ages 25 to 29.
Reference:
Friedman, J., York, H, Graetz, N., et al. “Measuring and Forecasting Progress towards the Education-Related SDG Targets.” Nature. 580, 636-639 (2020). doi.org/10.1038/s41586-020-2198-8
Extreme poverty
The Institute for Health Metrics and Evaluation (IHME) estimates the extreme poverty rate as the proportion of the population living below a consumption level of $US3.00 per person per day in 2021 purchasing power parity-adjusted dollars, following the World Bank’s current extreme poverty line definition.
Drawing on the methods described in Moses et al 2021, we estimated national mean consumption and the national consumption Lorenz curve. We first extracted household consumption and income data from various survey sources, cleaned and standardized the data, and cross-walked income data to consumption using a linear mixed effects model. We then modeled a complete series of estimates across 204 countries for the years 1980 to 2023, generating 500 draws to incorporate uncertainty. We combined the mean consumption and Lorenz curve estimates to generate the consumption distribution for each country, year, and population percentile. From these consumption distributions, we extracted the percentage of the population living on less than $US3.00 per day to obtain estimates of the national extreme poverty rate.
National extreme poverty rates were forecasted from 2024 to 2045. We first fit a meta-stochastic frontier model to historical consumption and poverty estimates, extracted the modeled inefficiency from the most recent observed year, and held it constant by country over 2024 to 2045. Consumption was forecast from 2023 to 2045 using a linear mixed effects model with IHME GDP per capita forecasts as a predictor and country-level random effects. We then applied the frontier model to the forecast levels of inefficiency and consumption for each country and year to produce the national extreme poverty rate forecasts.
Reference:
Moses, M. W., Kharas, H., Miller-Petrie, M. K., Tsakalos, G., Marczak, L., Hay, S. I., Murray, C. J. L., and Dieleman, J. L. Global Poverty and Inequality from 1980 to the COVID-19 Pandemic. SocArXiv, March 18, 2021. https://osf.io/preprints/socarxiv/x47np_v1
WHERE THE DIFFERENCE GETS MADE: A NURSE, A FARMER, A TEACHER
Physician density
The health workforce estimates are based on forecasted physician density per capita, using data from Knight et al, 2026. Estimates were produced for 204 countries and territories from 1990 to 2023, and forecasts to 2045 were produced from an ensemble model driven by historical rates of change with varying weights across child models for weighting recent years more heavily, informed by out-of-sample predictive validity.
Reference:
Knight, M., Haakenstad, A., & GBD 2023 Collaborators. (2026, July). “Measuring the composition and availability of human resources for health by sex in 204 countries and territories, 1990–2023, and workforce gaps for universal health coverage: A systematic analysis for the Global Burden of Disease Study 2023.” The Lancet. www.thelancet.com/journals/lanpub/article/PIIS2468-2667(26)00145-3/fulltext
SDG Indicators
IHME methodology
Measuring the impact of foreign aid cuts on SDG indicators
To estimate the impact of recent aid cuts, there are two primary areas of work that IHME undertook: 1) modeling and estimating the impact of funding cuts across all relevant sources on projections of development assistance for health (DAH) and total health spending (THE); and 2) estimating the associated impact that reductions in health spending will have on the Goalkeepers Sustainable Development Goal (SDG) indicators.
To comprehensively measure funding cuts both domestically and internationally to foreign aid, IHME collected and standardized estimates of global health spending from diverse data sources, including commitments and disbursements from development project records, annual budgets, financial statements, and revenue reports. Long-term forecasts through 2050 for DAH were generated using donor targets, historical trends, and GDP-based forecasting, while future THE was predicted through ensemble models assessing key indicators like government expenditure and out-of-pocket costs. IHME’s estimates and forecasts of foreign aid are based on publicly available data as of July 15, 2026. Detailed methodologies are available in the online Methods Annex of the 2025 Financing Global Health report www.healthdata.org/research-analysis/library/financing-global-health-2025-cuts-aid-and-future-outlook.
To quantify the impact of reductions in development aid on the SDG indicators, IHME analyzed the relationship between each indicator and historical health spending. We used a nonparametric stochastic frontier analysis approach, where the “frontier” represents the best possible outcome (e.g., the lowest number of new tuberculosis cases) for a given level of health spending. The model assumes that this efficiency persists into the future, allowing us to calculate the marginal cost, in decreased health coverage, lives lost, or additional cases, of reduced funding.
For each SDG indicator reported, apart from smoking prevalence and HIV incidence, forecasts were produced using spending trajectories derived from the best available country-level expenditure estimates. For 2025, these estimates capture observed funding cuts; from 2026 onward, they reflect country-specific projections rather than a fixed percentage reduction. We leveraged donor countries’ official development assistance (ODA) per gross national income (GNI) targets where available and otherwise applied projected growth rates of ODA per GNI based on a linear regression model and expected changes in GDP per capita. Details for how we captured the impacts of funding on HIV are described in the indicator section below.
Reference:
Apeagyei, A., Barış, E., Dieleman, J., Elliott, H., Leach-Kemon, K., Lidral-Porter, B., Murray, C. J. L., Nam, S., Shyong, C., Tsakalos, G., and Zlavog, B. Financing Global Health 2025: Cuts in Aid and Future Outlook—Methods Annex. Seattle, Washington, Institute for Health Metrics and Evaluation, 2025. www.healthdata.org/research-analysis/library/financing-global-health-2025-cuts-aid-and-future-outlook
Indicators estimated by IHME
IHME produced estimates and forecasts for 13 of the SDG indicators included in the 2026 Goalkeepers Report. The section below provides details on how each indicator is estimated.
Stunting
IHME measures child stunting prevalence as height-for-age more than two standard deviations below the reference median on the height-age growth curve based on WHO 2006 growth standards for children 0 to 59 months.
The modeling uses a combination of mean height-for-age z-scores (HAZ) and severity-specific stunting prevalences (mild, moderate, severe, and extreme) to estimate the entire distribution of HAZ for each location, sex, and year for each of six distinct age groups of children <5. Each continuous distribution was integrated to produce the final stunting prevalence estimate.
Our forecasts of stunting are driven by several factors: historic temporal patterns, household consumption, the sociodemographic index and climate change. To generate forecasts of stunting that incorporate the effect of climate change, we analyzed individual-level geolocated data from the Demographic and Health Surveys (DHS) to determine the relationship between the prevalence of stunting and climate factors. We utilized data totaling over 1.1 million geo-located child observations from 145 surveys, covering 57 countries. We implemented a statistical model that estimated the relationship between stunting prevalence and climate factors (mean annual temperature and days over 30°C) while controlling for altitude, household consumption and country-level random effects. We also include an interaction term between household consumption and days over 30°C to capture the mitigating factor of increasing household consumption on heat stress. In addition to the climate-specific portion of the projection, we also model any remaining or residual variation between the predictions and the historical GBD estimates using the sociodemographic index (SDI) as a predictor of the residuals. The better and worse scenarios were produced by taking the 85th and 15th percentile rates of change observed across location-years in the past and applying those rates of change to all locations in the future.
To estimate the impact of reduced development assistance for health, we used a nonparametric stochastic frontier approach (as detailed in the “Measuring funding cuts to foreign aid and their impact on SDGs” section above) to model how decreased spending—accounting for total health spending including World Food Programme assistance and varying by country and year—would impact stunting prevalence and adjusted our forecasts per the reference scenario for each country through 2045 accordingly.
Under-5 mortality rate
IHME defines the under-5 mortality rate as the probability of death between birth and age 5. It is expressed as the number of deaths per 1,000 live births.
Estimates used all available data from vital registration, sample registration, surveys, and censuses, which were modeled via a statistical model for age-specific mortality rates that incorporates both parametric and non-parametric methods. Age-specific mortality rates for ages 0 to 6 days, 7 to 27 days, 1 to 5 months, 6 to 11 months, 1 to 2 years, and 2 to 4 years were jointly estimated in the model, then converted to the under-5 mortality rate.
Forecasts to 2045 use the modeling approach published in GBD 2021 Forecasting Collaborators, based on a combination of key drivers, including Global Burden of Disease (GBD) risk factors linked to individual causes of death in the under-5 age group (e.g. child stunting, wasting, and underweight), selected interventions (e.g., vaccines), and SDI. Forecasts are produced for each cause of death and are then aggregated to all-cause mortality, which are then used to compute the final indicator.
Better and worse scenarios were constructed by applying the 85th and 15th percentile of historically observed rates of change to all drivers (risk factors, sociodemographic index, and interventions) across location-years, respectively.
To estimate the impact of reduced development assistance for health, we used a nonparametric stochastic frontier approach (as detailed in the “Measuring funding cuts to foreign aid and their impact on SDGs” section above) to model how decreased spending—varying by country and year—would increase under-5 mortality, and adjusted our forecasts per the reference scenario for each country through 2045 accordingly.
References:
- GBD 2021 Forecasting Collaborators. (2024). “Burden of disease scenarios for 204 countries and territories, 2022–2050: A forecasting analysis for the Global Burden of Disease Study 2021.” The Lancet, 403 (10440), 2204–2256. doi.org/10.1016/S0140-6736(24)00685-8
Neonatal mortality rate
IHME defines neonatal mortality rate as the probability of death in the first 28 completed days of life. It is expressed as the number of deaths per 1,000 live births. Estimates used all available data from vital registration, sample registration, surveys, and censuses, which were modeled via a statistical model for age-specific mortality rates that incorporates both parametric and non-parametric methods. Age-specific mortality rates for ages 0-6 days and 7-27 days were jointly estimated in the model, then converted to neonatal mortality rate.
Forecasts to 2045 use the modeling approach published in GBD 2021 Forecasting Collaborators, based on a combination of key drivers, including Global Burden of Disease (GBD) risk factors linked to individual causes of death in the under-5 age group (e.g. child stunting, wasting, and underweight), selected interventions (e.g., vaccines), and SDI. Forecasts are produced for each cause of death and are then aggregated to all-cause mortality which are then used to compute the final indicator. Better and worse scenarios were constructed by applying the 85th and 15th percentile of historically observed rates of change to all drivers (risk factors, sociodemographic index and interventions) across location-years, respectively.
To estimate the impact of reduced development assistance for health, we used a nonparametric stochastic frontier approach (as detailed in the “Measuring funding cuts to foreign aid and their impact on SDGs” section above) to model how decreased spending—varying by country and year—would increase neonatal mortality, and adjusted our forecasts per the reference scenario for each country through 2045 accordingly.
References:
- GBD 2021 Forecasting Collaborators. (2024). “Burden of disease scenarios for 204 countries and territories, 2022–2050: A forecasting analysis for the Global Burden of Disease Study 2021.” The Lancet, 403(10440), 2204–2256. doi.org/10.1016/S0140-6736(24)00685-8
HIV
IHME estimates the HIV rate as new HIV infections per 1,000 population. Estimation of HIV incidence employs two primary modeling packages: A GBD-adapted Estimation and Projection Package Age-Sex Model (EPP-ASM) for countries with robust HIV prevalence data from antenatal care clinics or representative population-based seroprevalence surveys; and a GBD-adapted Spectrum framework for the remaining locations. Forecasts are based on forecasted antiretroviral therapy (ART), prevention of mother-to-child transmission (PMTCT) coverage, and transmission rate. ART coverage data leverages nationally reported ART coverage program data, and for Goalkeepers 2025 for the year 2025 for sub-Saharan Africa adult ART coverage data came from the UNAIDS 2026 data release. Updates to this year’s estimates and forecasts reflect newly integrated data for ART, updated household survey and antenatal care prevalence data, as well as refreshed input parameters reflecting the latest evidence on HIV mortality and disease progression.
Tuberculosis
IHME estimates new and relapse tuberculosis (TB) cases diagnosed within a given calendar year (incidence) using data from prevalence surveys, case notifications, and cause-specific mortality estimates as inputs to a statistical model that enforces internal consistency among the estimates.
IHME estimates incidence differently between countries with high-quality cause of death data and those without. For countries with high-quality data, we build incidence directly from age and sex-specific case notifications, combining all new and relapse cases. For countries with weaker systems, where notifications are less reliable, we instead model incidence from the relationship between TB mortality and incidence using the high-quality countries as a training set. We estimate mortality to incidence ratios as a function of age, sex, region and country specific summary measure of Healthcare Access Quality and then combine those ratios with our TB mortality estimates to derive incidence. All inputs are reconciled in a Bayesian disease model that estimates incidence, prevalence, remission, and mortality together and keeps them mutually consistent. An adjustment to align modeled incidence with reported notifications while accounting for the fact that not every notified case is true TB. Drawing on high-quality countries, we estimate the fraction of notified cases likely to be true (bacteriologically confirmed) cases and apply this to lower quality settings, correcting the overestimation that results from treating all notifications as true cases.
Forecasts to 2045 used an ensemble model driven by historical rates of change with varying weights across child models for weighting recent years more heavily, informed by out-of-sample predictive validity. Better and worse scenarios were constructed by applying the 85th and 15th percentile of historically observed rates of change across location-years, respectively.
To estimate the impact of reduced development assistance for health, we used a nonparametric stochastic frontier approach (as detailed in the “Measuring funding cuts to foreign aid and their impact on SDGs” section above) to model how decreased spending—varying by country and year—would increase TB incidence, and adjusted our forecasts per the reference scenario for each country through 2045 accordingly.
References:
- Ledesma et al. “Global, Regional, and National Burden of Tuberculosis and Multidrug-Resistant Tuberculosis by HIV Status, 1990–2023: A Systematic Analysis for the Global Burden of Disease Study 2023.” The Lancet Infectious Diseases. Forthcoming 2026. Supplementary appendix.
- Ledesma et al. “Global, Regional, and National Age-Specific Progress towards the 2020 Milestones of the WHO End TB Strategy: A Systematic Analysis for the Global Burden of Disease Study 2021.” The Lancet Infectious Diseases, 24, no. 7 (2024): 698-725. www.thelancet.com/journals/laninf/article/PIIS1473-3099(24)00007-0/fulltext
- Ledesma et al. “Global-, Regional-, and National-Level Impacts of the COVID-19 Pandemic on Tuberculosis Diagnoses, 2020–2021.” Microorganisms 11, no. 9 (2023): 2191. www.mdpi.com/2076-2607/11/9/2191
Malaria
IHME estimates the malaria rate as the number of new cases per 1,000 population. Malaria incidence estimates utilize data inputs from routine malaria case reports from national routine surveillance systems; cross-sectional, geolocated, and community-representative observations of Plasmodium falciparum parasite rate (PfPR); data on Plasmodium knowlesi specifically for Malaysia; data from the Multiple Indicator Cluster Survey (MICS) and Demographic and Health Survey for the estimation of treatment seeking; estimates of malaria interventions such as insecticide-treated bednets, indoor residual spraying, and effective treatment with an antimalarial drug; and other environmental, sociodemographic, and economic covariates.
Forecasts to 2045 were produced with a five-component model, all fit at the second administrative level: Model 1: modeling Plasmodium falciparum prevalence in 2 to 10 year olds (Pfpr2-10); Model 2: modeling the all-age prevalence-case rate given PfPR2-10; Model 3: modeling the age- and sex-specific case rate given all-age case rate; Model 4: modeling the all-age prevalence-fatality rate given PfPR2-10; and finally Model 5: modeling the age- and sex-specific fatality rate given the all-age fatality rate. Each model used Admin 2-level data from 2000 to 2025 on both the malaria outcomes but also the weighted fraction of the year where temperature is suitable for transmission (weighted by a continuous measure of suitability, ranging from 0 to 1 as a non-linear function of temperature), yearly flood days per capita, GDP per capita, and DAH spending on malaria (which captured any observed and forecasted spending cuts). The better and worse scenarios were produced by taking the 85th and 15th percentile rates of change observed across location-years in the past and applying those rates of change to all locations in the future.
Neglected tropical diseases
IHME measures the sum of the prevalence of 15 NTDs per 100,000 that are currently measured in the annual Global Burden of Disease study: human African trypanosomiasis, Chagas disease, cystic echinococcosis, cysticercosis, dengue, food-borne trematodiases, Guinea worm, soil-transmitted helminths (STH, comprising hookworm, trichuriasis, and ascariasis), leishmaniasis, leprosy, lymphatic filariasis, onchocerciasis, rabies, schistosomiasis, and trachoma.
The methods used to estimate prevalence for each NTD reflect differences in data availability, transmission dynamics, and disease natural history. Across most NTDs, input data are sex- and age-split prior to modelling. Two modelling tools are used most commonly: DisMod Bayesian Meta-Regression (DisMod-MR), a disease modelling framework that estimates internally consistent epidemiological parameters (incidence, prevalence, remission, and excess mortality); and spatiotemporal Gaussian process regression (ST-GPR), a statistical approach that borrows information across geography and time to produce a complete time series of estimates.
Forecasts to 2045 used an ensemble model driven by historical rates of change with varying weights across child models for weighting recent years more heavily, informed by out-of-sample predictive validity. Better and worse scenarios were constructed by applying the 85th and 15th percentile of historically observed rates of change across location-years, respectively.
To estimate the impact of reduced development assistance for health, we used a nonparametric stochastic frontier approach (as detailed in the “Measuring funding cuts to foreign aid and their impact on SDGs” section above) to model how decreased spending—varying by country and year—would increase NTD prevalence and adjusted our forecasts per the reference scenario for each country through 2045 accordingly.
Family planning
IHME estimates the proportion of women of reproductive age (15 to 49 years) who have their need for family planning satisfied with modern contraceptive methods. Modern contraceptive methods include the current use of male or female sterilization, male or female condoms, diaphragms, cervical caps, sponges, spermicidal agents, oral hormonal pills, patches, rings, implants, injections, intrauterine devices (IUDs), and emergency contraceptives.
Estimates are derived from two types of sources: individual-level survey microdata and tabulated survey reports. Major survey series include Demographic and Health Surveys (DHS), Multiple Indicator Cluster Surveys (MICS), Performance Monitoring for Action (PMA) surveys, Generations and Gender Programme (GGP) surveys, and country-specific surveys. From microdata, key variables extracted include contraceptive usage, marital status and sexual activity, fecundity, desire for children, and pregnancy and postpartum amenorrheic status. For tabulated survey reports, estimates of any contraceptive use, modern contraceptive use, unmet need for family planning, and marital status breakdowns are manually extracted. All sources are collapsed to age-specific estimates and combined prior to modeling.
Need for family planning is determined separately for partnered and unpartnered women aged 15 to 49 using the 2012 DHS revised definition of unmet need, applied through a standardized algorithm. Women are classified as having a need through three main pathways: 1) they are currently using any contraceptive method (modern or traditional); 2) they are currently pregnant or postpartum amenorrheic from a birth in the last two years and express desire to have prevented or delayed that pregnancy; or 3) they are fecund, do not want a child in the next two years, and are either married/in-union or have been sexually active in the last 30 days. Unpartnered women must additionally meet a sexual activity criterion that married women are assumed to satisfy automatically. When surveys are missing key components of the algorithm — such as desire for children, fecundity information, or postpartum amenorrheic status — counterfactual re-extractions are performed under alternative assumptions (e.g. assuming all women have a need), and an adjustment factor is estimated and applied to adjust those estimates relative to gold standard surveys that contain full information.
Rather than modeling met need and contraceptive prevalence directly, IHME uses a nested proportions approach that models three underlying components separately: (1) any contraceptive prevalence; (2) the proportion of any contraceptive use that is modern; and (3) the proportion of non-users with unmet need for family planning. Each component is modeled separately for partnered and unpartnered women using spatiotemporal Gaussian process regression (ST-GPR). The Socio-Demographic Index (SDI) is a key covariate in all models. Partnered model results are used as custom covariates in the unpartnered models to account for data sparsity among unmarried women. An additional ST-GPR model of the proportion of women currently partnered is used to aggregate marital-status-specific results to all-women estimates. Modeling proceeds in two stages: preliminary runs use age- and marital-status-specific data only, with results used to split age-aggregated and all-women data; final runs incorporate all available data. Age-aggregated data are split into five-year age bins and all-women data are split into partnered and unpartnered estimates using draws from the preliminary models.
Final estimates for the indicator are calculated by combining draws across partnered and unpartnered estimates using marital status results, then aggregating to all ages. Forecasts to 2045 used an ensemble model driven by historical rates of change with varying weights across child models for weighting recent years more heavily, informed by out-of-sample predictive validity. Better and worse scenarios were constructed by applying the 85th and 15th percentile of historically observed rates of change across location-years, respectively.
To estimate the impact of reduced development assistance for health, we used a nonparametric stochastic frontier approach (as detailed in the “Measuring funding cuts to foreign aid and their impact on SDGs” section above) to model how decreased spending—varying by country and year—would decrease met need for modern contraception and adjusted our forecasts per the reference scenario for each country through 2045 accordingly.
Universal health coverage
The universal health coverage (UHC) effective coverage index is a metric composed of 23 effective coverage indicators spanning population-age groups across the life course: maternal and newborn, children under age 5, youths ages 5 to 19, adults ages 20 to 64, and adults ages 65 and older. These indicators fall within three health service domains: promotion, prevention, and treatment.
The health system promotion indicator is met need for family planning with modern contraception.
Health system prevention indicators include the proportion of children receiving the third dose of the diphtheria-tetanus-pertussis vaccine and the proportion receiving the first dose of measles-containing vaccine. Antenatal care for mothers and antenatal care for newborns are considered indicators of both prevention and treatment of diseases affecting maternal and child health.
Indicators of treatment for communicable diseases include the mortality-to-incidence (MI) ratios for lower respiratory infections, diarrhea, and tuberculosis, as well as coverage of antiretroviral therapy (ART) among people with HIV/AIDS. Indicators of treatment for noncommunicable diseases include MI ratios for acute lymphoid leukemia, appendicitis, paralytic ileus and intestinal obstruction, cervical cancer, breast cancer, uterine cancer, and colorectal cancer, along with mortality-to-prevalence (MP) ratios for stroke, chronic kidney disease, epilepsy, asthma, chronic obstructive pulmonary disease, and diabetes, and the risk-standardized death rate due to ischemic heart disease.
Effective coverage indicators are weighted in the index according to the potential health gain each country could achieve by improving coverage of that indicator.
To forecast the UHC index from 2025 to 2045, we fit a meta-stochastic frontier model using total health spending per capita projections as the independent variable. Country- and year-specific inefficiencies were then extracted from the model and forecasted to 2045 using a linear regression with exponential weights across time for each country. These forecasted inefficiencies, together with forecasted total health spending per capita, were substituted into the fitted frontier to produce forecasted UHC for all countries for 2025–2045. For countries with recent or ongoing conflicts, we also adjusted UHC estimates using ACLED conflict data. From this dataset, we identified seven countries with high severity of conflict and over 5 percent of the population impacted by conflict (Iran, Palestine, Lebanon, Syria, Ukraine, Haiti, Myanmar). For these countries and years impacted by conflict, we reduced UHC estimates by applying the country’s maximum observed inefficiency and scaling the magnitude of the conflict shock by the proportion of the population impacted by conflict. Finally, we linearly attenuated the magnitude of the decline in UHC due to conflict back to baseline levels over 10 years.
References:
- Armed Conflict Location & Event Data Project “(ACLED). ACLED: Real-time data and analysis on political violence and protest.” acleddata.com.
Smoking
IHME measures the age-standardized prevalence of any current use of smoked tobacco among those age 15 and older. IHME collates information from available representative surveys that include questions about self-reported current use of tobacco and information on the type of tobacco product smoked (including cigarettes, cigars, pipes, hookahs, and local products). IHME converts all data to its standard definition of any current smoking within the last 30 days so that meaningful comparisons can be made across locations and over time.
Our models are informed by self-reported use of current smoking (occasional or daily use) from validated nationally representative surveys, such as DHS, MICS, GATS, and others. Additional surveys are included as well, based on quality and representativeness. Individual-level data is collapsed into GBD-standard five-year age-sex bins (male or female, ages 10 to 14 through 95+) for each modelled location and year. Tabulated data is split into GBD standard bins via age and sex patterns modeled on location and region-specific individual-level data. Recall periods are cross walked to match our gold standard definition. The full dataset is used to inform a spatiotemporal gaussian process regression model (ST-GPR) of current smoking prevalence. The ST-GPR yields temporally smoothed estimates that incorporate regional influences in data-sparse locations.
To account for the inherent link in the populations of current and former smokers—current smokers who quit become former smokers—we forecast current and former smoking prevalence jointly. We define ever smoking prevalence as the proportion of people who are either current or former smokers and compute the ratio of current to ever smoking prevalence. We then use a generalized ensemble model to forecast 1) ever smoking prevalence and 2) the ratio of current to ever smoking prevalence annually from 2024 through 2045 and recover current and former smoking prevalence forecasts from these two quantities. This model incorporates historical rates of change in conjunction with the relationship between the sociodemographic index and smoking prevalence. Better and worse scenarios are constructed by applying the 85th and 15th percentile of historically observed rates of change across location-years, respectively.
Vaccines
IHME’s measurement of immunization coverage reports on the coverage of the following vaccines separately: three-dose diphtheria, tetanus and pertussis containing vaccine (DTP3), measles second dose (MCV2), and three-dose pneumococcal conjugate vaccine (PCV3).
IHME estimates annual vaccine coverage from 1990 to the present using two complementary streams of data: country-reported data (both administrative and official coverage data) and survey data. Country-reported administrative and official coverage are submitted each year through the World Health Organization (WHO)–United Nations Children’s Fund (UNICEF) Joint Reporting Form (JRF), which provides near-complete annual time series. Survey data are obtained from nationally-representative household surveys, principally the Demographic and Health Surveys (DHS) and Multiple Indicator Cluster Surveys (MICS), supplemented by other multi-country and country-specific surveys. Because country-reported coverage is subject to systematic bias, IHME compares paired survey and country-reported observations from the same country-years, matched by birth cohort, then estimates patterns of bias by country and over time using meta-regression, with the Healthcare Access and Quality (HAQ) Index as a covariate (assuming that reporting bias varies with the general quality of health services). These models are then used to adjust country-reported data for bias prior to inclusion in vaccine coverage models.
Survey and bias-adjusted country-reported data are then used as inputs into spatio-temporal Gaussian process regression (ST-GPR) models, which are used to produce coverage estimates. ST-GPR consists of a three-stage framework that fits a covariate-based regression, borrows strength across space and time to improve estimates where data are sparse, and follows data closely where available and reliable. ST-GPR model covariates include the HAQ Index, mortality due to war and conflict, and vaccine disruptions arising from stockouts, the COVID-19 pandemic, and other events. As in last year’s report, pandemic-related disruptions were estimated for the years 2020-2023. To maintain internal consistency across doses and antigens, DTP3 is estimated by modelling the ratio of DTP3 to the first dose of that vaccine (DTP1), then multiplying this ratio by DTP1 coverage estimates, while MCV2 and PCV3 are modelled as ratios to MCV1 and to DTP3, respectively, and constrained to be less than one. Coverage estimates for MCV2 and PCV3 additionally account for scale-up in the years following country-specific vaccine introduction using hierarchical spline models. This approach allows global patterns of scale up to inform country-specific scale-up models, strengthening early post-introduction estimates in locations where data are sparse or absent.
Forecasts to 2045 were modeled using a linear mixed effects model to forecast vaccine coverage, using SDI the primary covariate. For vaccines that have not been universally introduced (MCV2 or PCV3), coverage is modeled as the ratio of MCV2 coverage to MCV1 coverage and the ratio of PCV3 to DTP3 coverage. For countries that have not yet introduced either vaccine, a Weibull distribution parameterized by lag-distributed income, GAVI eligibility status and education attainment is used to simulate introduction years, with coverage scaling up to that of MCV1 or DTP3, respectively. Better and worse scenarios were constructed by applying the 85th and 15th percentile of historically observed rates of change across location-years, respectively.
To estimate the impact of reduced development assistance for health, we used a nonparametric stochastic frontier approach (as detailed in the “Measuring funding cuts to foreign aid and their impact on SDGs” section above) to model how decreased spending—varying by country and year—would decrease vaccine coverage and adjusted our forecasts per the reference scenario for each country through 2045 accordingly.
Sanitation
IHME estimates the proportion of the population with access to safely managed sanitation. As defined by the Joint Monitoring Programme (JMP), a safely managed facility must meet three criteria: (1) is not shared with multiple households, (2) is an improved sanitation facility, and (3) its wastewater is disposed of safely. Safe wastewater disposal can consist of being treated and disposed of in situ, stored temporarily and treated off-site, or transported through a sewer and treated. Safely managed treated wastewater must have received at least secondary treatment. IHME estimated households with piped sanitation (with a sewer connection or septic tank); households with improved sanitation but without a sewer connection (pit latrine, ventilated improved latrine, pit latrine with slab, composting toilet); households without improved sanitation (flush toilet that is not piped to sewer or septic tank, pit latrine without a slab or open pit, bucket, hanging toilet or hanging latrine, no facilities); and wastewater that receives at least secondary treatment for sewer-connected households, as defined by the JMP for Water Supply and Sanitation.
We developed models to estimate three components of safely managed sanitation: 1) the proportion of collected wastewater that receives at least secondary treatment, 2) the proportion of sewer-connected facilities that are safely managed, and 3) the proportion of improved, non-sewer facilities that are safely managed.
Data for estimating the proportion of collected wastewater that receives at least secondary treatment were extracted from Eurostat, Aquastat, the Organisation for Economic Co-operation and Development (OECD), and national surveys. Data for estimating the proportion of sewer-connected facilities that are safely managed were extracted from Eurostat, Aquastat, Demographic and Health Surveys (DHS), UNICEF Multiple Indicator Cluster Surveys (MICS), OECD, and national surveys (Republic of Korea, Singapore, Andorra, Austria, and Ireland). Data for estimating the proportion of improved, non-sewer facilities that are safely managed were extracted from MICS, DHS, Eurostat, and national surveys (Canada, Norway, and the United States).
We estimated the proportion of the total population with safely managed sanitation as the sum of the proportion of the population with safely managed sewer-connected facilities and the proportion of the population with safely managed improved non-sewer facilities.
Forecasts to 2045 were generated using an ensemble model that incorporated both historical rates of change and SDI projections. Component model weights were determined through out-of-sample predictive validity testing, with greater weight assigned to models that more heavily emphasize recent trends. Better and worse scenarios were constructed using the 85th and 15th percentile of historically observed rates of change across location-years, respectively.
To estimate the impact of reduced water, sanitation, and hygiene (WASH) overseas development assistance and government spending, we used a nonparametric stochastic frontier approach (as detailed in the “Measuring funding cuts to foreign aid and their impact on SDGs” section above) to model how decreased WASH spending—varying by country and year—would decrease safely managed sanitation coverage, and adjusted our forecasts per the reference scenario for each country through 2045 accordingly.
References:
World Health Organization and United Nations Children’s Fund. SDG Indicator Metadata: Indicator 6.2.1(a), Proportion of Population Using Safely Managed Sanitation Services. Updated December 20, 2021. WHO/UNICEF Joint Monitoring Programme PDF
IHME indicator sources
Data source information for each indicator are below, a detailed reporting of data sourcing for GBD 2023 estimates can be found at https://sources.healthdata.org/collection/sources-2023
| Indicator and Component | Total Sources |
|---|---|
| Child mortality | 25,260 |
| Child stunting | 1,102 |
| Family planning (met need) | 1,148 |
| Malaria | 13,099 |
| Maternal mortality | 8,107 |
| Neonatal mortality | 25,260 |
| HIV | 7,849 |
| NTD chagas | 1,199 |
| NTD visceral leishmaniasis | 5,815 |
| NTD cutaneous and mucocutaneous leishmaniasis | 1,503 |
| NTD African trypanosomiasis | 3,075 |
| NTD schistosomiasis | 3,855 |
| NTD cysticercosis | 3,901 |
| NTD cystic echinococcosis | 3,731 |
| NTD lymphatic filariasis | 496 |
| NTD onchocerciasis | 351 |
| NTD trachoma | 109 |
| NTD dengue | 3,950 |
| NTD rabies | 4,058 |
| NTD ascariasis | 4,599 |
| NTD trichuriasis | 868 |
| NTD hookworm disease | 873 |
| NTD food-borne trematodiases | 57 |
| NTD leprosy | 1,595 |
| NTD guinea worm disease | 458 |
| Sanitation safely managed | 1,356 |
| Smoking prevalence | 3,630 |
| Tuberculosis | 8,422 |
| UHC maternal disorders | 8,107 |
| UHC met need | 1,148 |
| UHC live births | 16,589 |
| UHC neonatal mortality | 25,260 |
| UHC diphtheria | 4,479 |
| UHC pertussis | 9,667 |
| UHC tetanus | 4,679 |
| UHC dtp vaccination | 9,005 |
| UHC measles | 11,333 |
| UHC measles vaccination | 8,893 |
| UHC LRI | 4,860 |
| UHC diarrhea | 6,087 |
| UHC HIV treatment | 7,849 |
| UHC TB | 4,842 |
| UHC lymphoid leukemia | 4,492 |
| UHC asthma | 3,349 |
| UHC diabetes | 4,627 |
| UHC IHD treatment | 4,603 |
| UHC stroke | 4,631 |
| UHC Chronic kidney disease | 4,381 |
| UHC Chronic obstructive pulmonary disease | 3,366 |
| UHC cervical cancer | 4,461 |
| UHC breast cancer | 4,583 |
| UHC uterine cancer | 4,460 |
| UHC colon and rectum cancer | 4,550 |
| UHC epilepsy | 4,370 |
| UHC appendicitis | 4,468 |
| UHC paralytic ileus and intestinal obstruction treatment | 4,342 |
| Vaccine coverage dtp3 | 10,035 |
| Vaccine coverage mcv2 | 3,602 |
| Vaccine coverage pcv3 | 2,440 |
Indicators estimated from other sources
Poverty
Lakner, C., Foster, E.M., Jolliffe, D., Ibarra, G.L., and Baah, S. Reproducibility Package for Global Poverty Revisited Using 2021 PPPs and New Data on Consumption. Data set. RR_WLD_2025_346. World Bank, 2025. doi.org/10.60572/qmda-qt67.
For methodology, see: World Bank, Global Poverty Revisited Using 2021 PPPs and New Data on Consumption, Policy Research Working Paper 11137, Washington, DC: World Bank, 2025. documents1.worldbank.org/curated/en/099503206032533226/pdf/IDU-e2e09dcf-0af2-481a-a60a-64adf28171d0.pdf.
This report uses the international poverty line and purchasing power parity (PPP) data incorporated in the World Bank’s June 2025 update, which raised the international poverty line from $US2.15 a day (2017 PPP) to $US3.00 a day (2021 PPP). This update is reflected in the current edition to ensure consistency and comparability with the World Bank’s current global poverty methodology.
Agriculture
Food and Agriculture Organization of the United Nations. Income of Small-Scale Food Producers, PPP (Constant 2011 International USD) Data set, SDG Indicator 2.3.2, Dataflow DF_SDG_2_3_2, series SI_AGR_SSFP FAO Data Explorer, 2025. de-public-statsuite.fao.org.
Small food producers’ income growth is included for selected countries with at least two data points in the data set. For each country, income growth is calculated between the earliest and most recent years available, except where specific data points were excluded following expert review of data quality. The specific years used for each country are listed below:
| Location | Year Range |
|---|---|
| Benin | 2019–2021 |
| Burkina Faso | 2014–2021 |
| Cambodia | 2009–2021 |
| Côte d’Ivoire | 2008–2019 |
| Ethiopia | 2014–2019 |
| Ghana | 2013–2017 |
| Guinea-Bissau | 2019–2021 |
| India | 2005–2012 |
| Malawi | 2011–2020 |
| Mali | 2014–2021 |
| Niger | 2011–2019 |
| Nigeria | 2013–2019 |
| Pakistan | 2011–2019 |
| Senegal | 2011–2022 |
| Sierra Leone | 2011–2018 |
| Tanzania | 2009–2021 |
| Togo | 2019–2021 |
| Uganda | 2010–2020 |
Education
World Bank, UNESCO Institute for Statistics, UNICEF, USAID, Bill & Melinda Gates Foundation, & Foreign, Commonwealth & Development Office, The State of Global Learning Poverty: 2022 Update. Conference Edition, 2022, www.unicef.org/media/122921/file/StateofLearningPoverty2022.pdf.
Source for Learning Poverty 2022 simulations:
Azevedo, J. P., Demombynes, G., and Wong, Y. N., “Why Has the Pandemic Not Sparked More Concern for Learning Losses in Latin America? The Perils of an Invisible Crisis,” Education for Global Development, 2023. blogs.worldbank.org/en/education/why-hasnt-pandemic-sparked-more-concern-learning-losses-latin-america-perils-invisible.
Gender equality
The Equal Measures 2030 (EM2030) SDG Gender Index is the most comprehensive global tool to measure progress toward gender equality aligned to the SDGs. The index tracks 56 key gender indicators that provide the “big picture” across and within 14 of the 17 SDGs.
It is the only index that adds a gender lens to each of the goals, including the many SDGs that lack such a lens in the official framework. Going beyond SDG 5 (the single goal dedicated to gender equality) is important in capturing the broader trends that influence progress on gender equality and highlighting how issues such as hunger, poverty, and climate change affect girls and women.
The 2024 index covers 139 countries, which represent 96 percent of the world’s women and girls. The index tracks scores for three reference years: 2015, 2019, and 2022 and forecasts a scenario for 2030 based on current trends.
This is the third edition of the SDG Gender Index—it was previously released in 2019 and 2022. It is one of the few global gender indices to be formally audited by the European Commission Joint Research Centre’s Competence Centre on Composite Indicators and Scoreboards (JRC-COIN). The fourth edition of the SDG Gender Index will be released by Equal Measures 2030 later this year.
The index was developed by a coalition of national, regional, and global leaders from feminist networks, civil society, and international development.
Resources:
- To download 2024 index data and the latest index report and for more information about index methodology, see: www.equalmeasures2030.org/2024-sdg-gender-index
- To access interactive index data visualizations, see: www.equalmeasures2030.org/2024-sdg-gender-index/explore-the-data/
- To view the technical audit conducted by the COIN center of the EU’s Joint Research Centre, see www.equalmeasures2030.org/2024-sdg-gender-index/about-the-index/
Equal Measures 2030. A Gender Equal Future in Crisis? Findings from the 2024 SDG Gender Index. 2024 www.equalmeasures2030.org/2024-sdg-gender-index.
Inclusive financial systems
The “income” comparison refers to what the World Bank calculates as account ownership of the richest 60 percent of households versus the poorest 40 percent of households.
Klapper, L., Singer, D., Starita, L., and Norris, A., The Global Findex Database 2025: Connectivity and Financial Inclusion in the Digital Economy. Washington, DC: World Bank, 2025. doi.org/10.1596/978-1-4648-2204-9.
World Bank. “Account Ownership at a Financial Institution or with a Mobile-Money-Service Provider (% of Population Ages 15+).” Global Findex Database. Data set. 2025. genderdata.worldbank.org/en/indicator/fx-own-totl-zs
For methodology, see:
Klapper, L., Singer, D., Starita, L., and Norris, A. “Survey Methodology.” Appendix A in The Global Findex Database 2025: Connectivity and Financial Inclusion in the Digital Economy, 267–294. Washington, DC: World Bank, 2025. www.worldbank.org/en/publication/globalfindex/methodology.