Background: An adequate amount of prepaid resources for health is important to ensure access to health services and for the pursuit of universal health coverage. Previous studies on global health financing have described the relationship between economic development and health financing. In this study, we further explore global health financing trends and examine how the sources of funds used, types of services purchased, and development assistance for health disbursed change with economic development. We also identify countries that deviate from the trends. Methods: We estimated national health spending by type of care and by source, including development assistance for health, based on a diverse set of data including programme reports, budget data, national estimates, and 964 National Health Accounts. These data represent health spending for 184 countries from 1995 through 2014. We converted these data into a common inflation-adjusted and purchasing power-adjusted currency, and used non-linear regression methods to model the relationship between health financing, time, and economic development. Findings: Between 1995 and 2014, economic development was positively associated with total health spending and a shift away from a reliance on development assistance and out-of-pocket (OOP) towards government spending. The largest absolute increase in spending was in high-income countries, which increased to purchasing power-adjusted $5221 per capita based on an annual growth rate of 3.0%. The largest health spending growth rates were in upper-middle-income (5.9) and lower-middle-income groups (5.0), which both increased spending at more than 5% per year, and spent $914 and $267 per capita in 2014, respectively. Spending in low-income countries grew nearly as fast, at 4.6%, and health spending increased from $51 to $120 per capita. In 2014, 59.2% of all health spending was financed by the government, although in low-income and lower-middle-income countries, 29.1% and 58.0% of spending was OOP spending and 35.7% and 3.0% of spending was development assistance. Recent growth in development assistance for health has been tepid; between 2010 and 2016, it grew annually at 1.8%, and reached US$37.6 billion in 2016. Nonetheless, there is a great deal of variation revolving around these averages. 29 countries spend at least 50% more than expected per capita, based on their level of economic development alone, whereas 11 countries spend less than 50% their expected amount. Interpretation: Health spending remains disparate, with low-income and lower-middle-income countries increasing spending in absolute terms the least, and relying heavily on OOP spending and development assistance. Moreover, tremendous variation shows that neither time nor economic development guarantee adequate prepaid health resources, which are vital for the pursuit of universal health coverage.
Health spending stems from four sources: the government, which includes general government budgets and social health insurance; prepaid private spending, which includes private insurance and non-governmental organisation spending; OOP payments; and development assistance for health. The sum of these sources make up total health-care spending. Government, prepaid private, and OOP spending data were extracted from the WHO Global Health Observatory. These data measure the sum of all outlays for health maintenance, restoration, or enhancement paid for in cash or supplied in kind.23 This excludes indirect health spending, such as lost wages due to illness or transportation costs; spending on informal care, such as care provided by a family member; spending on traditional healers; and illegal so-called black market or under the table transactions, such as bribes. Spending estimates were extracted in national currency units and divided by gross domestic product (GDP), also reported in national currency units and reported by WHO. This fraction was multiplied by GDP per capita reported in inflation-adjusted 2015 PPP $.24 WHO data tracks spending by agent, such that it is unclear if government and prepaid private spending were sourced domestically. To differentiate between domestically and internationally financed spending, development assistance for health provided to the government was removed from WHO’s government spending estimates and development assistance for health provided to non-government providers or organisations was removed from WHO’s prepaid private estimates.25 For the 184 countries, between 1995 and 2014, 1·7%, 14·8%, and 1·7% of the government, prepaid private, and OOP health spending estimates were missing, respectively. These estimates were imputed in R using Amelia II: A program for missing data (version 1·7·4), and more information about these methods is provided in the appendix (p 19).26 The result of this approach is four mutually exclusive, collectively exhaustive spending estimates by source and time—government as source, prepaid private (excluding donor financing), OOP health spending, and development assistance for health. These four series were summed to form annual estimates of total health spending for each of the 184 countries, from 1995 through 2014. Development assistance for health is the financial and in-kind resources transferred from development agencies to low-income and middle-income countries with the primary purpose of maintaining or improving health.27 Many of the methods used to estimate development assistance for health have been used and published previously, although the input data and some methods have been updated and improved for this study.25, 27, 28 These estimates are based on data from all publicly available databases tracking development assistance, including project-level records from the Organisation for Economic Co-operation and Development (OECD) and other development agencies, such as the World Bank, the Global Fund, Gavi, and the Bill & Melinda Gates Foundation. In addition, audited budget statements and annual reports are used to estimate development assistance for health through 2016. When disbursement data were not available, commitment data are adjusted to reflect disbursements. Disbursements are tracked comprehensively from source to disbursing agency, also known as the disbursing channel, to recipient to avoid double counting associated with development agencies transferring resources among themselves. These data are disaggregated based on the source of development assistance for health, disbursing channel, health focus area, and country recipient. Development assistance for health is disaggregated by health focus area—newborn and child health, maternal health, HIV/AIDS, malaria, tuberculosis, non-communicable diseases, other infectious diseases, health system strengthening, which includes discretionary grants to the health sector, and other and unallocable—and tracks disbursements from 1990 to 2016. Other includes projects that are for health projects that do not fit into any of the other health focus areas, while unallocable development assistance for health is that that does not have sufficient data to be disaggregated by health focus area. A more thorough presentation of these methods and explanation on how these methods defer from past research is provided in the appendix (p 41). The Institute for Health Metrics and Evaluation’s (IHME) development assistance for health database further disaggregates funds based on whether they are expected to be channelled to the recipient country’s government, or are provided to non-governmental providers or organisations.29 Development assistance for health estimates are reported in 2015 US$, and were converted into 2015 purchasing-power-adjusted US$ to reflect the purchasing power of development assistance for health in the recipient country. Purchasing power parity exchange rates were based on data from the International Monetary Fund (IMF), World Bank, and WHO. Total health spending was also disaggregated by type of goods and services, such as inpatient or outpatient care. For this purpose, we collected all available National Health Account (NHA) reports. NHAs track health spending using an agreed upon accounting framework developed by the OECD, WHO, and Eurostat. The standards were first codified in 2001, but then updated in 2011 to what is now known as the System of Health Accounts (SHA) 2011.30, 31 A systematic review located, extracted, and published data from 872 reports spanning 1996 to 2010.32 We added to this review by searching WHO, OECD, and Eurostat databases, and Google for search terms “National Health Account” and “System of Health Account”, which identified 178 additional NHAs. The newly collected reports are primarily from more recent years. We mapped estimates based on NHA 2001 standards to SHA 2011 with methods described in the appendix (p 20). Because the NHA 2001 and SHA 2011 standards for tracking spending on some preventive health services were irreconcilable, we include only the spending on immunisations and early disease detection. Excluded categories contain spending on education and counselling programmes, epidemiological surveillance, and disaster preparedness. Of the 1050 NHA reports identified for this research, only 964 NHA reports included the necessary data (specific information about the exclusion criterion used to determine the set of used NHAs outlined in the appendix [p 20]). These 964 reports span 108 countries and range from 1995 to 2014. From these, we extracted total and government health spending by type of goods and services, and aggregated spending into eight categories: inpatient curative and rehabilitative care; day and outpatient curative and rehabilitative care; long-term care; ancillary services; medical goods, which includes pharmaceuticals; governance and health-system and financing administration; immunisation and early disease detection programmes; and other care. Estimates by type of service were generated as a share of both total health spending and total government health spending. Other care is a residual category that includes all health spending not included in the other categories. GDP data spanning 1995 to 2015 were based on data collected from the International Monetary Fund, World Bank, the UN, the Maddison Project, and Penn World Tables database.33, 34, 35, 36 These data were combined using regression methods and previously developed for producing a complete GDP time series.24 GDP data were reported in inflation-adjusted 2015 PPP $. We report estimates aggregated by FY2016 World Bank income groups and Global Burden of Disease (GBD) super regions.37, 38 World Bank income groups are four mutually exclusive categories assigned by the World Bank and based on gross national income. GBD super regions are seven mutually exclusive categories based on geography and cause of death patterns. Spending estimates were constructed to reflect the group or region as a whole. For example, the group’s health spending per capita is the group’s total spending divided by total population. Similarly, the group’s government health spending as a share of total spending was the group’s total government spending divided by the group’s total health spending. We completed three primary analyses. For all three analyses, we used multivariate penalised spline regression to allow for flexible and non-linear model fit across all 184 countries and 20 years of data.39 First, we regressed the natural log of total health spending per capita on the natural log of GDP per capita and time. Second, we regressed four health spending by source fractions—development assistance for health and government, prepaid private, and OOP health spending—on GDP per capita and time. Third, we regressed the eight health spending by type of goods and services fractions—inpatient curative and rehabilitative care, day and outpatient curative and rehabilitative care, long-term care, ancillary services, medical goods, governance and health-system and financing administration, immunisation and early disease detection programmes, and other care—on GDP per capita and time. For the second and third analyses, the spending fractions (by source and by type) were each measured as a share of total health spending and centre log-ratio transformed, while GDP per capita was natural log transformed.40 To estimate uncertainty, the underlying data was bootstrapped 1000 times, and all regressions were completed independently on each of the 1000 bootstrap samples.41, 42 Robustness checks included in the appendix (p 29) reinforce our qualitative conclusions, and use subsets of our data and also rely on the Socio-demographic Index and the Human Development Index, both of which track additional dimensions related to socioeconomic development, rather than simply GDP per capita which tracks economic development. To measure countries’ 2014 health spending relative to the expected value as determined by the fitted trend, we extracted the estimated country-year and year-specific residual (or error) from the regression analyses. The residuals measure the difference between the actual 2014 spending levels and the expected 2014 spending levels predicted by the model relative only to the country’s GDP per capita. The residual measures the effect of country characteristics not included in the model, such as health burden, health-system policies, prices, and society’s willingness to spend on health. In addition to this, we tested whether health spending per capita grew exponentially with economic development. To test these we used ordinary least squares to regress the natural log of health spending per capita on the natural log of GDP per capita. More details on all estimation are included in the appendix (p 4). The funder of this study had no role in the study design, data collection, data analysis, data interpretation, or writing of the report. All authors had full access to the data in the study and JLD and CJLM had final responsibility for the decision to submit for publication.