A map of recent global research on how health systems are financed, built from studies published since 2010, showing where evidence is plentiful and where it is thin.
This project was commissioned and funded by the Wellcome Trust and is led by the International Initiative for Impact Evaluation (3ie).
Health systems everywhere are under pressure. In high-income countries, ageing populations and tight budgets are straining care; in low- and middle-income countries, growing populations must be served as development assistance for health falls. Governments and funders need to know what the evidence says about raising, pooling and spending money for health, and where it is missing.
This project systematically maps the recent global empirical research on health system financing, published since 2010. It covers how money for health is raised, pooled, allocated and spent, as well as upstream investment in facilities, workforce, medicines and supply chains, and what these choices mean for health-system performance and the wider economy.
The maps describe the shape of the evidence base. They do not judge which financing approaches work best, rate study quality or risk of bias, or pool results. Their main purpose is to help researchers, funders and decision-makers see which policy-relevant areas are well studied and which are underexplored, to inform future research priorities and funding.
The evidence is presented as two linked maps. Each offers interactive visuals for exploring where research has been done, on what topics, with which methods, and how this has changed over time.
Research on the core financing functions (revenue raising, pooling, purchasing, benefit decisions and public financial management), plus capital investment and supply-chain financing, and the outcomes linked to them.
Open the Health Financing Functions map →Full economic evaluations that compare two or more ways of delivering care on both costs and health consequences, giving a focused view of value-for-money evidence.
Open the Health Economic Evaluations map →A study is included only if it meets all of the following.
| Criteria | Description |
|---|---|
| Population | Global |
| Interest | Evidence on recurrent financing of health system services; capital investments; links to the supply chain; and the consequences for health-system and societal goals. |
| Context | No limitations based on location (geographic or health setting) or health characteristics. |
| Study design | All types of empirical studies using qualitative, quantitative or mixed-methods approaches are included. For non-economic evaluations, no restriction is placed on methodological rigour. For economic evaluations, only studies comparing both costs and benefits and two or more alternative methods of delivering care are included. |
| Time and scope | Published articles and reviews published since 2010. English-only search terms are used, but no language restrictions are placed on included studies. |
More detailed inclusion and exclusion criteria are available in Extended data, Appendices C and D: Marion, P., Lee, S., Lotfi, T., Hutchinson, B., Sempé, L., Pande, S., et al. (2026), Systematic mapping of recent global research on health system financing using machine learning. View the extended data ↗
Not included
Start with the approach shared by both maps, then open a map's tab for how its dataset was built, its data sources and its limitations.
Loading methods…
The map covers full economic evaluations of health-care delivery published from 1 January 2010 onward: studies that compare two or more ways of delivering care on both costs and consequences. Partial evaluations (cost-minimisation, budget-impact and cost-of-illness studies) are excluded, as are records that remain unclear at extraction.
Titles and abstracts are screened with a machine-learning classifier. Against a human-coded reference set, it identified 93 of 97 eligible studies (sensitivity 0.959; 95% CI 0.899–0.984).
Eligible studies then go through field extraction. Intervention type and care setting are extracted by a single LLM pass and checked against a keyword list, the emergent topic clusters and a second model; they have not been validated by human reviewers. Bibliographic details, disease topics and geography come from publication metadata and external classifications (see Data sources).
Studies whose outcome is not a clinical measure, such as diagnostic accuracy or service use, are kept and flagged. Reviews that synthesise published economic evaluations are also kept: 1,602 of the 38,199 studies (4.2%) are flagged as reviews from their titles. Most of these (65%) name no single country, so country-level views draw mainly on primary studies.
| Pipeline stage | Records |
|---|
| Input | Source |
|---|---|
| Publication identifiers and metadata | Crossref REST API, OpenAlex and PubMed, covering about 93% of eligible studies. |
| Disease classification | PubMed Medical Subject Headings (MeSH) disease headings (2026 vocabulary), mapped to 12 disease groups and linked to Global Burden of Disease Level-2 causes. |
| Research topics | OpenAlex topic taxonomy, plus themes identified by clustering abstracts (UMAP and HDBSCAN). |
| Country income groups and population | World Bank country and lending groups and World Bank population data (latest year available, mostly 2025). |
| Disease burden | IHME Global Burden of Disease 2023 DALYs: country totals and 22 Level-2 causes. |
| Health spending | IHME health-spending estimates, 1995–2022: total health expenditure per capita (PPP, 2022). National totals are per-capita spending multiplied by the latest population. |
Almost all included studies are in English (99.8%). About a third report no study country and are excluded from country-level figures. Metadata coverage ranges from 47% to 93% by field and is reported with each figure. Comparisons with disease burden are ecological. Extracted fields were inconsistent for 8.3% of studies, which were excluded. A further 50,661 records rated uncertain at title and abstract screening are not yet included in the analysis.