1 Authors: Adam Wagstaff, DPhil†*, Sven Neelsen, PhD†* †World Bank, Washington DC, USA * Contributed equally Date: April 8, 2019 Short title: Universal Health Coverage in 115 Countries and Territories Key words: universal health coverage; service coverage; financial protection; catastrophic health spending; out-of-pocket payments Corresponding author: Adam Wagstaff, Development Research Group, World Bank, Washington DC 20433, USA, awagstaff@worldbank.org 2 Abstract Background. The goal of universal health coverage (UHC) requires that everyone receive needed health services, and that families who get needed services do not suffer undue financial hardship. Tracking progress towards UHC requires measurement of both dimensions, and a way of trading them off. Methods. We measure service coverage (SC) by a weighted geometric average of four prevention and four treatment indicators, financial protection (FP) by the incidence of ‘catastrophic’ health expenditures (those exceeding 10% of household consumption or income), and a country’s UHC performance as a geometric average of the SC index and the complement of the incidence of catastrophic expenditures. We adjust SC for inequality, penalizing high-inequality countries. We obtain the bulk of our data from over 1,700 household surveys. Findings. A low incidence of catastrophic expenses sometimes reflects low SC (often in low-income countries) but sometimes occurs despite high SC (often in high-income countries). At a given level of SC, too, FP varies. UHC index scores are higher in higher-income countries, but there are variations within income groups. Adjusting the UHC index for inequality in SC makes little difference in some countries, but not all. Most countries have increased their UHC index, sometimes sharply, many by improving both UHC dimensions. Some have increased their UHC index despite reductions in FP by substantially increasing their SC. Some countries have seen their UHC index fall, mostly because FP has worsened with stagnant or declining SC. The SC and UHC index are significantly associated with GDP per capita and the share spent on health. Spending by social health insurance and government schemes is significantly associated with the SC index, FP and the UHC index, but spending by other schemes is not. Interpretation. Progress towards UHC can be tracked using an index that captures both SC and FP. Funding. None. 3 The last few years have seen a growing commitment around the world to universal health coverage (UHC) with many countries embarking on UHC-inspired health reforms, commentators urging others to do so, and UHC being adopted as one of the new Sustainable Development Goals (SDGs). 1 UHC means that everyone, irrespective of their ability-to-pay, gets the health services they need without suffering undue financial hardship in the process.2 Measuring progress towards UHC thus requires simultaneous measurement of progress on both dimensions of UHC (‘service coverage’ (SC) and ‘financial protection’ (FP)). Yet, with the exception of two studies 3, 4, work to date has examined each dimension of UHC in isolation. Such studies are, as has been acknowledged 5-7, potentially misleading8, since countries may do well on one UHC dimension but not on the other. A low level of out-of-pocket expenditure on health could reflect people not getting services they need or people getting them but not paying for them out-of-pocket. A high level of use of health services may be associated with a high level of out-of-pocket expenditure or not. This paper builds on our earlier work3, 4 that measures progress on both dimensions of UHC simultaneously. This comprehensive approach to tracking progress towards UHC uses an index that allows progress on one UHC dimension to be traded off against progress on the other – a desirable feature given that policymakers seem likely to be willing (up to a point) to accept worse performance on one dimension (e.g. FP) in exchange for better performance on the other (e.g. SC). In addition, the approach captures the explicit concern about equity in the concept of UHC – shortfalls from UHC matter more if they are systematically associated with a family’s ability-to-pay. This paper extends the geographic 4 coverage from the 19 countries covered in our previous work 3, 4 to 115 countries, including high-income countries. In the measurement of FP, we use essentially the same methods used in our recent work on global trends in catastrophic out-of-pocket expenses. 7 In the measurement of SC, we stick fairly closely to our previous two papers.3, 4 These build on prior work on SC measurement in UHC2, 9, which lay the foundations for the SC index by Hogan et al. 5 Our approach differs, however, from the latter study in two ways. First, we use only indicators of service coverage; we exclude ‘upstream’ proxy indicators of policy intent and resource availability, and ‘downstream’ proxy indicators of health status. Second, we rely entirely on household survey data. Wherever possible, we use our own harmonized estimates from the raw microdata rather than published data in reports and on websites, since indicator definitions can vary from one survey ‘family’ to another, and even within a survey family. We avoid using administrative data, in part because they do not lend themselves to disaggregation by socioeconomic status, and in part because of concerns about accuracy, especially where governments do not face incentives to report accurate numbers. 10-12 In line with the growing concerns about the use of modeling in global health datasets 13, 14, we also avoid using modeled data. We also do no modeling of our own except to replace missing values with assumed values for selected indicators in low- and high-income countries; we highlight which countries are affected by this process. The downside is that our dataset is full of gaps. The upside is that, insofar as the surveys we use are reliable, differences over time or across countries ought to reflect reality rather than modeling assumptions; conversely, when real changes occur on the ground, they ought to get reflected in our numbers, rather than being smoothed away by the modeling process. 5 Evidence before this study Several previous studies reported estimates for multiple countries of either service coverage or financial protection. Only two studies reported estimates for both service coverage and financial protection and combined them in an inequality-adjusted UHC index. The index allowed tradeoffs between service coverage and financial protection and captured socioeconomic inequality in service coverage. It operationalized financial protection by two indicators, ‘catastrophic’ out-of-pocket medical spending, defined as exceeding 10% of household consumption, and ‘impoverishing’ out-of-pocket medical spending that pushes households below the poverty line. The eight service coverage components – utilization of antenatal care (4+ visits) and skilled birth attendance, full childhood immunization, treatment of respiratory infections and diarrhea among children under 5, hospital admissions, and use of cervical and breast cancer screening – encompassed preventive as well as curative care, and services related to both infectious and non-communicable diseases. Due to a lack of data, the two studies could only estimate the UHC index for 19 developing countries. On a scale of 0 to 100, with 100 representing UHC, countries in the 75-85 range included Brazil, Colombia, Costa Rica, Mexico, South Africa and the Philippines, while countries in the 35-50 range included Ethiopia, Indonesia and Peru. The studies found that increases in service coverage have often come at the expense of worsening financial protection, although the net effect has mostly been an increase in the UHC index. 6 Added value of this study This study uses the same service coverage indicators as the previous two studies, but with some modifications in the age range and frequency of screening in the case of the cancer prevention indicators, and improved harmonization across all indicators; for ease of interpretation, and to stick close to the SDG indicators, the present study uses just one financial protection indicator (catastrophic expenditures at the 10% threshold – the official SDG UHC indicator). Like the previous two studies, the present study also looks at changes in UHC at the country level, albeit using an improved methodology. The study’s principal added value over the previous two studies comes from the increase in country coverage. It extends the country coverage for the inequality-adjusted UHC index from 19 countries to 51. In addition, given the fact that data on socioeconomic inequality in service coverage are not always available, the paper estimates a simpler UHC index that is not adjusted for inequality in 115 countries, and compares the adjusted and unadjusted indexes in the countries where both can be computed. The 115 countries covered span the entire income range, from low- to high-income countries, across all continents, and together cover 92% of the world’s population in 2017. We also estimate trends in UHC in 52 countries. The paper thus provides the first ever comprehensive assessment of progress toward UHC, capturing both service coverage and financial protection, for the bulk of the world’s population. Implications of the available evidence While we find positive trends in UHC achievement in recent years for most countries for which data are available, large variations in UHC achievement remain, much of which is explained by differences in GDP. Nevertheless, there is substantial heterogeneity in UHC 7 achievement within income groups, and some countries perform better than countries in the income group above theirs. This, and the positive associations of the UHC index with the share of GDP that a country spends on health, and the shares of a country’s health budget that is channeled through government and SHI schemes, suggests that there is potential for national health policy-makers to accelerate progress towards UHC. Our UHC index (Figure 1) has two ‘dimensions’: financial protection (FP) and service coverage (SC). Each is weighted equally via a geometric average to allow FP and SC to be traded off against each other at a diminishing rate. The version of the index used in this paper has just one FP ‘domain’ (catastrophic out-of-pocket expenses), with just one indicator (the incidence catastrophic expenses at the 10% threshold). The SC dimension, by contrast, is divided into two ‘domains’ (prevention and treatment), weighted unequally via a geometric average, with each domain a geometric weighted average of four indicators. The indicators and their weights are shown in Figure 1. Choice of indicators The definitions and rationales for the indicators are summarized in Table 1. Six broad principles underpin the choice of SC indicators. First, they should be indicators of services delivered by health providers. We thus exclude ‘downstream’ indicators, notably health behaviors and health outcomes; these are influenced by health services, but also by other factors beyond the health sector. We also exclude ‘upstream’ indicators, such as health expenditures, health policies, and health infrastructure; these influence service delivery 8 (and thereby also indicators that are downstream of health services), but service delivery is also influenced by other factors beyond the health sector such as the level and distribution of household incomes. Second, the indicators should collectively cover a wide range of services relevant to a wide range of users and delivered by a wide range of providers; the services reflected in the indicators should not be overly focused on specific conditions or overly targeted toward a specific demographic group or delivered by a specific set of providers. Third, there should be broad agreement that the services captured by the indicators are medically necessary, as reflected in targets, guidelines, influential studies, etc. Fourth, for the reasons indicated in the Introduction, the data should come from household surveys, not from administrative or modeled data. Ideally, the surveys should contain sufficient socioeconomic data to allow for disaggregation by socioeconomic status. Fifth, each indicator should be able to be related to the population in need of the service, so that it can be transformed into an indicator measuring the proportion of the population in need getting the service, i.e. a measure of coverage. Sixth, the indicators should be widely available in surveys. This is a pragmatic consideration. As surveys become more comprehensive, this will become less of a constraint. The FP indicator is simply (100 minus) the percentage of the population incurring out-of- pocket expenses in excess of 10% of their consumption or income, which is SDG indicator 3.8.2. We essentially follow methods outlined in our recent global study 7 with one major difference: for high-income countries we relate health expenditures to household income not household consumption. 9 Two of our SC indicators (antenatal care (ANC) and treatment for acute respiratory infection (ARI)) were included in the recent SC index of Hogan et al. 5 Four of our SC indicators (full immunization, cervical cancer screening, skilled birth attendance and inpatient admissions) were proposed for inclusion by Hogan et al. in their SC index, but were excluded on the grounds of limited data availability. We also include two SC indicators not included by Hogan et al. The first is breast cancer screening, which Hogan et al. rejected on the grounds that guidelines are less clear than they are for cervical cancer screening.15 This may be true, but a recent high-profile review panel in the UK 16 concluded that “breast screening programs confer significant benefit and should continue”, and in practice most high-income countries, including all European Union countries 17, offer (at least) women aged 50-69 a mammogram every 2-3 years. The other indicator included in our index but excluded by Hogan et al. is diarrhea treatment, which they mentioned but excluded without explanation. This seems an important indicator of treatment of childhood illness; our focus, per a recent recommendation 18, is on treatment with oral rehydration salts (ORS). There are also some indicators in Hogan et al.’s index that we do not include. We list them here and indicate in parentheses our principal reason for excluding them: satisfied demand for family planning and non-use of tobacco (health behaviors); adequate sanitation (health infrastructure); the prevalence of non-raised blood pressure (a health outcome); health professionals per person (health resources); a core capacity index for international health regulations (health policies); tuberculosis effective treatment coverage and the number of people with HIV receiving antiretroviral therapy (modeled data); and the population at risk who sleep under insecticide-treated bednets (capturing need 10 requires adjusting household survey data for geographic heterogeneity in malaria risk, a task beyond the scope of this paper). Our indicator coverage can also be compared with other indices and indicator lists. Our index includes seven of the eight indicators in the composite coverage index (CCI) developed by the Countdown to 2030 for Maternal, Newborn, and Child Survival; 19 the missing indicator is unmet need for family planning, excluded for reasons given above. Our index includes only one of the 25 indicators in the global monitoring framework on NCDs, namely cervical cancer screening. Nine of the other 24 capture health behaviors, eight capture health outcomes, two capture policies, and one captures health resources (specifically, medicines and technologies). We exclude the remaining four indicators due to lack of coverage in household surveys: these include two prevention indicators (vaccination against human papillomavirus and hepatitis B) and two treatment indicators (consumption of strong opioid analgesics and drug therapy and counselling to prevent heart attacks and strokes). Choice of weights There is no right or wrong set of SC weights, and indeed no right or wrong way to choose them. The various SC indicators could be weighted equally. Or they could be computed using a statistical method such as factor analysis or principal components. Or they could be set according to best estimates of their effects on health, ideally measured using a measure that captures the quality and length of life. Or the weight-setting could be seen less as a technical exercise and more as a social valuation exercise, with the weights been set by citizens or by policymakers as their representatives. Or, in line with the social valuation 11 approach, one could use weights that broadly reflect the health expenditure shares of each service on the grounds that government and private expenditures reflect the choices that society makes. This is essentially what drove the choice of weights in this exercise. The 25% weight on prevention is somewhat higher than the share of prevention in total health spending in the countries of the Organization for Economic Cooperation and Development (OECD) and in Asia (around 5-10 percent according to the OECD and the Institute for Health Metrics) but not much larger than the average share spent in sub-Saharan Africa. 20 The 50% weight on inpatient admissions within the treatment domain is in line with the equal spending split between inpatient and outpatient care in the OECD countries. In the event, the results are robust to the choice of weights: for example, the rank correlation between the UHC index with the SC indicators weighted as in Figure 1 and an alternative UHC index with all eight SC indicators weighted equally is 0.979 (p<0.0001). Adjustment for inequality To capture socioeconomic inequality in service coverage, we compute, where the data permit it, an ‘inequality-adjusted’ version of the UHC index. In this paper, each SC indicator is adjusted downwards according to the degree of inequality favoring the better off; in contrast to our earlier studies, we do not adjust the score upwards for inequalities favoring the worse off. So, for example, if ANC is not systematically associated with socioeconomic status, or rates are higher among the worse off, no adjustment is made, and the score for the ANC indicator is the percent of pregnant women having 4 or more visits. If, on the other hand, rates are higher among the better off, the inequality-adjusted ANC indicator score is less than the percent of women having 4 or more visits, the downward adjustment being 12 larger the larger the inequality. Specifically, when there is inequality favoring the better off, the score is equal to the population in need receiving the intervention multiplied by the complement of a measure of inequality across the socioeconomic distribution known as the concentration index.21, 22 Where there is no pro-rich inequality, the score is simply equal to the population in need receiving the intervention. Estimating trends in UHC As we use only survey data, our dataset has many gaps. Our approach is to estimate the average annual change in a country’s UHC index and its two components. The average annual change of FP is estimated by regressing the log of the country’s FP score on time in years; the coefficient gives the annual average percentage increase in FP. Such a regression requires at least two datapoints, of course; we insisted on a minimum of three datapoints. We run the same regression for each component of the SC index, and then take a weighted average of the coefficients (using the weights in Figure 1) to get the annual average percentage change in SC. The annual average percentage change in the UHC index is then an unweighted average of the annual average percentage changes in FP and SC. This process results in us often having trend estimates for some but not all UHC indicators. To increase the number of countries in this trend analysis, we make a couple of simplifications. First, we assume flat trends for cancer screening in low-income countries with missing cancer screening data and for MCH indicators in high-income countries with missing MCH data. Second, we kept countries with incomplete trend data providing they have a trend for FP and trends for at least five SC indicators; implicitly we assumed a flat trend for indicators with missing trend data. In low-income countries, we required there be real 13 trend data for at least three SC indicators, so a country with assumed flat trends for the cancer screening indicators and real trend data for three other indicators would qualify. In high-income countries, we required there be real trend data on at least one SC indicator, so a country with assumed flat trends for the five MCH indicators but real trend data for one SC indicator would qualify. Further details of the methodology are in Annex 1. Aggregate correlates of UHC As in our work on global trends in financial protection 6, 7, we use multiple regression to explore the partial relationship between a country’s UHC indicator, on the one hand, and various macroeconomic indicators and health system characteristics, on the other. We use essentially the same model and the same variables as in our previous work 6, 7, with some changes due to the recent refinements in WHO’s Global Health Expenditure Database. We include one macroeconomic indicator, GDP per capita. We also include total health expenditure (THE) as a share of GDP. To capture the overall share of THE that is prepaid, and the mix across different prepayment ‘schemes’, we include the shares of THE spent by: (i) social health insurance (SHI) schemes; (ii) government agencies other than social health insurance schemes (referred to hereafter as ‘government schemes’); (iii) compulsory contributory health insurance schemes (e.g. compulsory private health insurance schemes and compulsory medical savings schemes); (iv) nonprofit institutions serving households (NPISH); and (v) voluntary health care payment schemes excluding (iv) (including voluntary private health insurance and enterprise schemes). 23 The omitted category is payments made out-of-pocket through no scheme, so the coefficients are to be interpreted as effects relative to this omitted category. We expect to see all of (i)-(iv) having positive 14 effects on SC indicators and negative effects on FP. We expect compulsory schemes to have larger impacts on UHC indicators than voluntary schemes. It is unclear a priori whether, on balance, government schemes will have larger or smaller impacts than contributory health insurance schemes, or how the effects of NPISH might compare with the effects of other schemes (NPISH schemes might focus more on some areas, e.g. MCH care, and less on others, e.g. inpatient care). Further details of the regression model are in Annex 2. Service coverage and financial protection data Our data are drawn almost entirely from the 2019 version of the World Bank’s Health Equity and Financial Protection Indicators (HEFPI) database. 24 This database draws on over 1,700 household surveys, with the raw microdata re-analyzed wherever possible to maximize consistency of indicator definitions across surveys and over time within surveys. Full details of the dataset construction are provided elsewhere. 24 Some datapoints in the HEFPI database have been excluded, and some datapoints not in the HEFPI dataset have been used in this paper; full details are provided in Annex 3. Table 2 summarizes the data gaps by World Bank country income groupings (as of 2017). In total, 63 countries have complete data for all nine UHC indicators. For low-income countries, the commonest missing indicators are the cancer screening variables. The rates for cancer screening in low-income countries that have data are very low: median rates are 3% for cervical cancer screening and 0.5% for breast cancer screening. We have assumed that low-income countries without any cancer screening data had values in 2017 equal to 15 these values. For high-income countries, the commonest missing indicators are the MCH indicators. We have assumed that high-income countries (and Argentina, Bulgaria, Mauritius and Romania) with missing MCH data had values of 90% for the missing indicator in 2017. We have not replaced any other missing values. The effect of these changes is to increase the number of countries with a complete set of UHC indicators to 115 (20 low-income countries, 29 lower middle-income countries, 28 upper middle-income countries, and 38 high-income countries). Data on the macroeconomic and health system indicators GDP and THE were taken from the World Bank’s Open Databases. The shares of THE channeled through the four aforementioned prepayment schemes were taken from the December 2018 update of WHO’s Global Health Expenditure Database (GHED); these data are available only from 2000 to 2015. Role of funding source The funding sources did not have any role in the design, conduct, analysis, or writing up of the study. The corresponding author had full access to all study data and had final responsibility for the decision to submit for publication. UHC: a snapshot Figure 2 maps the two components of the UHC index and the UHC index itself. In each map, countries are divided into five equal-size groups or quintiles. Figure 3 plots the two 16 dimensions of the UHC index against one another. Overlaid on the chart are contours showing the combinations of SC and FP that produce the same value of the UHC index; higher values of FP are bad, so countries doing best are those at the top left corner, with low rates of catastrophic expenses (i.e. high rates of FP) and high rates of SC. In producing Figure 2 we have taken the average of the country’s two most recent datapoints, or the only datapoint if only one is available. Data for different indicators may well refer to different years, and we have made no attempt to guess the trend to “line up” the data to the same year. Instead, we use whatever data are available, and indicate in the chart the median year that the data for the nine UHC indicators refer to. Markers with holes are used to highlight those low- and high-income countries where missing data were replaced by assumed values for one or more indicators. Figures 2 and 3 reveal several insights. First, as speculated in the Introduction, some countries (those at the bottom left in Figure 3) do indeed achieve a low incidence of catastrophic expenses because service coverage is very low (mostly low-income countries) while others (those at the top left in Figure 3 – mostly high-income countries) achieve a low incidence of catastrophic expenses despite achieving a high rate of SC. Likewise, the incidence of catastrophic expenses varies across countries with similar levels of SC. For example, China, Mexico and South Africa all have a SC score of around 60. Yet, South Africa has a lower incidence of catastrophic spending than Mexico, which has a (much) lower incidence of catastrophic spending than China. In short, one does indeed need to look at both SC and FP to get a sense of how close a country is to achieving UHC. Second, a country’s UHC score tends to be higher the higher the country’s income group: the high- income countries tend to cluster at the top left of Figure 3, while the low-income countries 17 tend to cluster at the bottom left. However, there are variations within income groups. Some low-income countries (e.g. Comoros and Zimbabwe) achieve higher UHC scores than several lower middle-income countries. Some lower middle-income countries (e.g. Bangladesh and Nigeria) achieve lower UHC scores than several low-income countries, while other lower middle-income countries (e.g. Kazakhstan and Ukraine) achieve higher UHC scores than several upper middle-income countries. Among the middle-income countries, there are weaker and stronger performers, with countries like Azerbaijan and Armenia achieving a lower UHC score than even some low-income countries, and countries like Bulgaria, Croatia and the Russian Federation achieving UHC scores on a par with high- income countries. Among high-income countries, Chile, Japan and Trinidad and Tobago stand out as weaker performers on the UHC index. Third, countries vary in their mix of SC and FP for a given level of UHC. For example, Brazil and Thailand (both upper middle- income countries) have the same UHC index value (75). Brazil’s SC score far exceeds Thailand’s (76 vs. 58) but this is counterbalanced in the UHC index by Brazil’s substantially higher incidence of catastrophic expenses (26 vs. 3). Portugal and Taiwan (both high- income territories) achieve a UHC score of 88: Portugal’s SC score exceeds Taiwan’s by 10 points (94 vs. 84) but this is counterbalanced by its higher incidence of catastrophic spending (17 vs. 7). Accounting for inequality Figure 4 shows the UHC index before and after adjusting for inequality in service coverage for countries where data on service coverage inequalities are available. The adjustment makes relatively little difference in some countries, but a sizable difference in others: in 18 Bangladesh, Chad, Cote d’Ivoire, Ethiopia, Guatemala, Indonesia and Lao PDR, accounting for inequality reduces the UHC index by more than 10%. In some cases, the adjustment results in countries changing their international ranking: for example, Guatemala is ahead of Tajikistan before the inequality adjustment but behind it afterwards; both Indonesia and Lao PDR slip behind Malawi once inequality in service coverage is captured. Trends in UHC Figure 5 plots the average annual improvement in FP against the average annual improvement in SC. The chart is divided into four quadrants and overlaid on the chart are contours showing combinations of average annual changes in FP and SC that produce the same average annual change in the UHC index. Green dots label countries where the UHC index has, on balance, increased, while red dots label countries where the UHC index has, on balance, decreased. Figure 5 reveals several insights. First, the majority of countries have seen increases in their UHC index. Many countries have seen improvements in both SC and FP (the top-right quadrant). However, many other countries have increased their UHC index by increasing SC but simultaneously reducing FP (the top-left quadrant). By contrast, in almost none of the countries is the opposite true – virtually no country has increased its UHC index by increasing FP at the expense of SC. Second, some countries have seen their UHC index fall. In some cases, this is because improvements in FP have been more than offset by deteriorations in SC (the bottom-right quadrant). But mostly it is because FP has worsened with stagnant or declining SC (the bottom-left quadrant). Third, several countries have seen their UHC index increase by more than one percent per annum. One country 19 (Rwanda) has exceeded two percent. In some cases (e.g. Ethiopia and Indonesia), UHC has increased by more than one percent per year despite reductions in FP. Correlates with health system features and macroeconomic indicators GDP per capita was positively and significantly associated with all SC indicators, the SC index and the UHC index (table 3). The share of GDP spent on health was positively and statistically associated with all nine UHC indicators, the SC index and the UHC index itself. The shares of THE channeled through compulsory contributory health insurance (SHI) schemes and government financing arrangements were both positively and significantly associated with all five MCH SC indicators and with the UHC index itself; both were also positively associated with the SC index, but spending channeled through government schemes was significantly associated with the SC index only at the 10% level while spending associated with SHIO schemes was significantly correlated at the 5% level. For all five MCH indicators, the effect of spending channeled through SHI schemes is larger, but significantly so only for ANC and SBA, and the effect is not statistically significant for the SC index. Expenditure by both ‘schemes’ is negatively associated with the incidence of catastrophic expenses; only in the case of the government scheme is the effect significant, but the difference between the two is not significant. Spending by compulsory private health insurance schemes is positively and significantly associated with five SC indicators but negatively and significantly associated with inpatient admissions but is not associated with the incidence of catastrophic expenses or with the UHC index itself. Spending by nonprofit schemes is positively associated only with the treatment of ARI and diarrhea, is 20 negatively associated with inpatient admissions and is not significantly associated with the SC index; it is, however, negatively associated with the incidence of catastrophic expenses. Spending by voluntary schemes is associated only with ANC, is not significantly associated with the SC index and is negatively associated with the UHC index. This paper measured achievement of UHC for 115 countries and territories with an index that is based on eight health service coverage and one financial protection indicator and penalizes unequal service coverage rates disfavoring the poor. Indicators underlying the index come largely from household survey data. Our results indicate large variations in UHC achievement across the world. We find a substantial degree of heterogeneity in financial protection at given levels of service coverage and vice versa – a finding that underscores that, to be meaningful, measures of UHC must consider both these dimensions simultaneously. Where data are available to observe country trends, we almost always find increases in the UHC index that generally occur in the presence of improved service coverage. The increases in the index, however, do not necessarily coincide with improvements in financial protection, highlighting a potential trade-off between the two UHC dimensions. Finally, we find UHC achievement to be clearly positively related to national income levels, the GDP spending share on health, and the share of healthcare spending that is channeled through SHI and government schemes. 21 Limitations Our list of SC indicators is, undeniably, shorter than desirable. A major constraint to extending the list is the limited number of household surveys that capture the need for health interventions and their receipt, and the socioeconomic characteristics of the individual’s household. DHS surveys contains rich socioeconomic data but are limited in their health data beyond the traditional MCH indicators. Some ask whether the respondent has been treated for TB, but to assess need the choice is between a question on whether the respondent has coughed blood (which overstates need) or whether they have been diagnosed with TB by a health provider (which understates need). By contrast, some DHS surveys test for HIV and hence get at need, but do not inquire about treatment. Other surveys, such as WHO’s STEPS surveys and TB prevalence surveys, contain rich data on need for and receipt of NCD care or TB treatment, but only a few contain socioeconomic data.25 Thus extending the list of indicators in our UHC index to include, say, HIV, TB and NCDs would mean either dramatically reduced country coverage or settling for the version of the UHC index that does not capture inequality. A second limitation of our study is that our list of SC indicators capture coverage rather than effective coverage. Ideally, we would like to capture, for example, not just whether a pregnant woman had 4 or more ANC visits, but whether she received the recommended interventions during these visits. Similarly, we would like to know not just whether someone was admitted to hospital but whether they were correctly diagnosed and received the most appropriate intervention. Making quality adjustments to convert coverage indicators into effective coverage indicators is far from straightforward. In the literature to 22 date, as far as we are aware, quality adjustments on the indicators we have used have been attempted only in the case of ANC.26, 27 Conclusions The last few years have seen a growing commitment around the world to universal health coverage (UHC) that has culminated in its adoption as an SDG. Despite the increasing prominence of UHC on the global agenda, its measurement is still hampered by severe data gaps, especially for low- and middle-income countries. Addressing these gaps – both in terms of broadening routine health surveys to include NCD indicators and increasing the frequency of data collection – would enhance the meaningfulness of UHC measurement for national policy formulation. Despite the current data shortcomings, several conclusions can be drawn from our analysis. While we find positive trends in UHC achievement in recent years for most countries for which data are available, large variations in UHC achievement remain, much of which is explained by differences in GDP. Nevertheless, there is substantial heterogeneity in UHC achievement within income groups, and some countries perform better than countries in the income group above theirs. This, and the positive associations of the UHC index with the share of GDP that a country spends on health, and the shares of a country’s health budget that is channeled through government and SHI schemes, suggests that there is indeed potential for national health policy makers to accelerate progress towards UHC. 23 Tables and Figures Table 1: Indicator definitions Dimension Domain Indicator Definition Rationale Main data sources MDG indicator 5.5. Included in composite coverage Percentage of most recent births in last two years index (CCI) of Countdown to 2030 for Maternal, (1) 4+ antenatal visits with at least 4 antenatal care visits (women age 18- DHS, MICS, WHS Newborn, and Child Survival.19 Included in Hogan et al.’s 49 at the time of the survey) WHO SC UHC index.5 Percentage of children age 15-23 months who received (a) Bacillus Calmette–Guérin (BCG) against WHO definition. Included in Countdown to 2030 CCI TB, (b) 3 doses of diphtheria-pertussis-tetanus (without polio). In line with SDG indicator 3.b.1. (2) Full immunization (DPT)/Pentavalent, (c) 3 doses of polio (excluding DHS, MICS Proposed for inclusion in Hogan et al.’s WHO SC UHC Prevention polio given at birth), and (d) Measles/Measles- index but excluded due to insufficient data. Mumps-Rubella (MMR), either verified by vaccination card or by recall of respondent DHS, EHIS, Percentage of women who received a mammogram in (3) Breast cancer Eurobarometer, the last 2 years (preferably age 50-69 but age groups Recommended in OECD countries and by WHO20. screening SAGE, STEPS, US- may vary) NHIS, WHS Service coverage Recommended by U.S. Preventive Services Taskforce, DHS, EHIS, Percentage of women who received a pap smear in WHO19, etc. Indicator no. 25 of the global monitoring (4) Cervical cancer Eurobarometer, the last 5 years (preferably age 30-49 but age groups framework on NCDs. Proposed for inclusion in Hogan et screening STEPS, US-NHIS, may vary) al.’s WHO SC UHC index but excluded due to insufficient WHS data. Percentage of most recent births in last 2 years MDG indicator 5.2. SDG indicator 3.1.2. Included in attended by any skilled health personnel (women age (5) Skilled birth Countdown to 2030 CCI. Proposed for inclusion in 18-49 at the time of the survey). Definition of skilled DHS, MICS, WHS attendance (SBA) Hogan et al.’s WHO SC UHC index but excluded due to varies by country and survey but always includes insufficient standardized data. doctor, nurse, midwife and auxiliary midwife. Percentage of children under 5 with cough and rapid Treatment breathing (in MICS case, originating from the chest) (6) Treatment for Acute in the two weeks preceding the survey who had a Included in Countdown to 2030 CCI. Included in Hogan Respiratory Infection consultation with a formal health care provider, DHS, MICS et al.’s WHO SC UHC index. (ARI) excluding pharmacies and visits to ‘other’ health care providers. Definition of formal health care providers varies by country and data source. Percentage of children under 5 with diarrhea in the 2 (7) Treatment for weeks before the survey who were given oral Included in Countdown to 2030 CCI. DHS, MICS Diarrhea rehydration salts (ORS). 24 Proposed for inclusion in Hogan et al.’s WHO SC UHC DHS, ECHP, index but excluded due to insufficient data. Reflects ENAHO, (8) Inpatient care use in concerns over NCDs and underutilization of hospital Eurobarometer, last 12 months (% of Percentage of population age 18 and older using care. WHO has proposed28 a benchmark of 0.1 inpatient EHIS, ISSP, LSMS, population age 18 and inpatient care in the last 12 months. admissions per capita, equivalent to 9% of population MCSS, US-NHIS, older) with inpatient admission in last 12 months. Rate is SLC, SUSENAS, UK- expressed as percent of this benchmark or 100% if rate GHS, WHS is above 9%. Financial Catastrophic Percentage of population with out-of-pocket health (9) Catastrophic protection expenditures expenses exceeding 10% of household consumption SDG UHC financial protection indicator (3.8.2). Various expenditures at 10% or income. 25 Table 2: Data availability by country income level No. UHC indicators missing 0 1 2 3 4 5 6 7 8 9 Total Low income 10 1 10 10 1 2 34 Lower middle income 28 3 5 4 1 1 3 2 47 Upper middle income 25 1 4 4 5 4 2 2 8 55 High income 2 1 1 36 6 7 7 19 79 Total 63 7 20 18 8 41 6 9 12 31 215 Notes: Table shows number of countries in each income group with number of UHC indicators missing. For example, 10 low-income countries have zero missing indicators. 26 Table 3: Multiple regressions showing marginal effects of macroeconomic and health systems characteristics on UHC at median per capita income Cervical Breast ANC Full immun. SBA Treatment Treatment Inpatient Service Catastrophic UHC Cancer cancer of ARI of diarrhea admission coverage expenses screening screening Per capita GDP 1.194 1.504 1.710 1.082 1.583 1.743 1.898 0.078 1.396 0.099 1.357 2011 int. $ p<0.0001 p<0.0001 p<0.0001 p<0.0001 p<0.0001 p<0.0001 p<0.0001 p=0.086 p<0.0001 p=0.29 p<0.0001 Total Health Exp. 1.116 1.886 1.476 1.722 2.153 1.633 1.418 0.344 1.674 0.737 1.361 (THE as % GDP p=0.029 p=0.003 p=0.0062 p=0.00028 p=0.00019 p<0.0001 p=0.00023 p=0.0011 p=0.00029 p=0.003 p=0.00053 SHI schemes as % 0.200 0.164 0.398 0.329 0.429 0.293 0.232 0.036 0.234 -0.050 0.235 THE p=0.095 p=0.22 p<0.0001 p=0.0003 p<0.0001 p=0.00018 p=0.03 p=0.1 p=0.010 p=0.095 p=0.0032 Govt. ‘schemes’ as 0.053 0.211 0.208 0.303 0.205 0.159 0.211 0.026 0.161 -0.096 0.271 % THE p=0.63 p=0.14 p=0.022 p=0.00028 p=0.0043 p=0.0067 p=0.0008 p=0.055 p=0.058 p=0.00039 p=0.0042 Comp. PI as % THE 1.255 0.644 2.068 1.788 0.978 1.815 2.302 -0.103 0.380 -0.116 0.227 p=0.017 p=0.09 p=0.03 p=0.00079 p=0.28 p=0.0024 p=0.0089 p=0.076 p=0.070 p=0.47 p=0.82 Nonprofit schemes -0.085 0.727 0.021 0.037 0.09 0.230 0.406 -0.082 -0.460 -0.149 0.122 as % THE p=0.85 p=0.29 p=0.9 p=0.74 p=0.72 p=0.017 p<0.0001 p=0.012 p=0.35 p=0.0016 p=0.49 Vol. Schemes as % 0.197 0.101 0.446 0.033 0.313 -0.062 0.235 -0.031 0.118 -0.053 -0.372 THE p=0.36 p=0.59 p=0.021 p=0.87 p=0.14 p=0.67 p=0.21 p=0.32 p=0.48 p=0.50 p=0.0034 No. observations 294 254 369 388 413 380 377 329 118 520 111 p: SHI= Govt. Fin. p=0.12 p=0.44 p=0.017 p=0.72 p=0.0026 p=0.055 p=0.83 p=0.62 p=0.37 p=0.13 p=0.65 Note: GDP=gross domestic product. THE=total health expenditure. 27 Figure 1: UHC Index with weights in parentheses Dimensions Domains Indicators Catastrophic health Catastrophic expenses Financial Protection (50%) expenses (at 10% (100%) threshold) (100%) 4+ antenatal visits (25%) UHC Index Full immunization (25%) Prevention (25%) Breast cancer screening (25%) Cervical cancer screening (25%) Service Coverage (50%) Skilled birth attendance (16.66%) Treatment for Acute Respiratory Infection (ARI) (16.66%) Treatment (75%) Treatment for Diarrhea (16.66%) Inpatient admission in previous 12 months (50%) 28 Figure 2: Service coverage, financial protection (catastrophic payment incidence) and UHC index Service Coverage 89 − 93 68 − 89 53 − 68 33 − 53 16 − 33 No data Financial Protection: Catastrophic Spending (10% threshold) 13 − 28 9 − 13 5−9 3−5 0−3 No data Universal Health Coverage Index 89 − 95 78 − 89 69 − 78 55 − 69 39 − 55 No data 29 Figure 3: UHC index, latest 1-2 surveys 100 UHC=90 UHC=80 LUX SWE2014 FIN 2016 2016 PRT 2016 AUT FRA DEU CZE 2015 2016 20162016 2016 HRV ITA 2017 ESP 2016 AUS 2013BEL 2015 DNK 2016 HUN SVN IRL 2014 2016 2016 ISR 2017 POL 2017 CAN 2015 SVK 2016 USA 2018 CHE LTU 2016KOR 2017 GRC 2014 LVA 2015 2015 2015 CYP MLT 2015 GBR 2015 EST 2015 HKG 2017 BGR 2016 TWN 2014 ARG 2013 UHC=70 80 URY 2009 CHL 2017 UKR 2011 BRA 2006 Service Coverage DOM 2012 KAZ 2011 MUS 2003 MNG 2012 JPN 2017 MYS 2004 BIH 2009 IRN 2010 NIC 1999 COL 2009 RUS 2008 MDA 2010 CHN 2004 TTOMEX ZAF 2008 20102011 UHC=60 60 NAM 2008 HND 2010 SRB 2012 CRI 2008 JOR 2015 THA 2014 COG 2011 PHL 2015 ECU 2004 PRY 2003 VNM 2012 ROU 2014 LKA 2013 PER 2016 TUR 2004 JAM UZB 2003 SLE 2007 2015 SLV ZWE2011 2011 TJK 2014 XKX 2013 GHA 2011 GEO 2006 UHC=50 40 GTM 2009 COM 2006 2010 IND 2013 ARM RWA 2012 ALB 2012 TZA 2012 NPL 2011 MRT 2011 UGA 2014 AZE 2005 SENBFA 2013 COD 2012 EGY 2011 2008 KEN PAK 2011 2009 AFG 2013 SWZ 2011 LAO 2010 MAR 2000 ZMB 2009 BDI 2013 IDN 2010 MDG 2006 BGD 2013 2009 2008 NER TCD MWI 2014 HTI 2014 20122015 MMR NGA KHM 2012 20 MLI 2014 CIV 2013 ETH 2014 0 10 20 30 40 Financial (Non)Protection (CATA10) High income Upper middle income Lower middle income Low income UHC Index 0 10 20 30 40 50 60 70 80 90 Ukraine Argentina Dominican Republic Mexico Bosnia and Herzegovina Kazakhstan Costa Rica South Africa Namibia Honduras Russian Federation Brazil Jordan Moldova Colombia Nicaragua Congo, Rep. Vietnam Ecuador Uzbekistan Paraguay Peru Before inequality adjustment El Salvador 30 Zimbabwe Philippines Tajikistan Ghana Comoros Figure 4: Effect of adjusting UHC index for inequalities in service coverage Azerbaijan Georgia Burkina Faso India Zambia Kenya Guatemala Senegal Mauritania Nepal Pakistan Egypt, Arab Rep. Eswatini After inequality adjustment Morocco Malawi Lao PDR Indonesia Myanmar Chad Bangladesh Mali Ethiopia Cote d'Ivoire 31 Figure 5: UHC index trends 4 RWA 2.02 Better SC, worse FP Better SC and FP UHC change=2 3 ETH 1.38 NER 1.17 Av. annual % improvement in SC IDN 1.19 PHL 0.96 BDI 1.13 BGD 0.86 UZB 1.35 2 VNM 1.19 BFA 1.23 IND 0.43 JPN 0.65 BOL 0.64 SEN 0.53 KAZ 0.63 UHC change=1 1 TJK PAK 0.46 0.50 MOZ 0.47MLI 0.54 GHA 0.54 UGA 0.31 IRL 0.27 POL 0.24 ISR 0.23 HRV 0.32 MEX 0.37 KOR 0.08 SVN 0.13 BGR -0.37 LVA -0.32 EST -0.11 MDG 0.05 SVK 0.15 HUN 0.14 PER GIN0.16 0.17 0 CHE -0.42 LTU -0.10 TZA -0.10 CAN -0.03 GBR -0.00 KEN -0.07 CMR 0.30 USA -0.07 MNG -0.42 KGZ -0.24 ARM -0.26 THA -0.26 NGA -0.52 MWI -0.54 -1 UHC change=0 -2 ROU -1.45 -3 UHC change=-1 ALB -1.87 Worse SC and FP Worse SC, better FP -4 -1 -.5 0 .5 1 Av. annual % improvement in FP (100-CATA10) UHC index improved UHC index worsened 32 Author contributions SN and AW constructed the service coverage dataset. AW led the analysis and wrote the first draft. Acknowledgements We thank Maxime Émile Armand Roche and Benoît Simon for help preparing MICS and financial protection microdata, and Gabriel Liu, Rachel Lu, Irene Wong and Qinghua Zhao for help analyzing the microdata from Hong Kong, Japan and Taiwan. The findings, interpretations and conclusions expressed in this paper are entirely those of the authors, and do not necessarily represent the views of the World Bank, its Executive Directors, or the governments of the countries they represent. Conflicts of interest disclosure All authors have no conflicts to disclose. 33 References 1. Horton R, Das P. Universal health coverage: not why, what, or when--but how? Lancet 2015; 385(9974): 1156-7. 2. Boerma T, Eozenou P, Evans D, Evans T, Kieny M-P, Wagstaff A. Monitoring Progress towards Universal Health Coverage at Country and Global Levels. PLoS Med 2014; 11(9): e1001731. 3. Wagstaff A, Cotlear D, Eozenou PH-V, Buisman LR. Measuring progress towards universal health coverage: with an application to 24 developing countries. Oxford Review of Economic Policy 2016; 32(1): 147-89. 4. Wagstaff A, Dmytraczenko T, Almeida G, Buisman L, Hoang-Vu Eozenou P, Bredenkamp C, Cercone JA, Diaz Y, Maceira D, Molina S, Paraje G, Ruiz F, Sarti F, Scott J, Valdivia M, Werneck H. Assessing Latin America’s Progress Toward Achieving Universal Health Coverage. Health Affairs 2015; 34(10): 1704-12. 5. Hogan DR, Stevens GA, Hosseinpoor AR, Boerma T. Monitoring universal health coverage within the Sustainable Development Goals: development and baseline data for an index of essential health services. The Lancet Global health 2018; 6(2): e152-e68. 6. Wagstaff A, Flores G, Smitz MF, Hsu J, Chepynoga K, Eozenou P. Progress on impoverishing health spending in 122 countries: a retrospective observational study. The Lancet Global health 2018; 6(2): e180-e92. 7. Wagstaff A, Flores G, Hsu J, Smitz MF, Chepynoga K, Buisman LR, van Wilgenburg K, Eozenou P. Progress on catastrophic health spending in 133 countries: a retrospective observational study. The Lancet Global health 2018; 6(2): e169-e79. 8. McPake BI. Crunching health expenditure numbers: important but treacherous terrain. The Lancet Global Health 2018; 6(2): e124-e5. 9. Boerma T, AbouZahr C, Evans D, Evans T. Monitoring Intervention Coverage in the Context of Universal Health Coverage. PLoS Med 2014; 11(9): e1001728. 10. Sandefur J, Glassman A. The Political Economy of Bad Data: Evidence from African Survey and Administrative Statistics. The Journal of Development Studies 2015; 51(2): 116-32. 11. Lim SS, Stein DB, Charrow A, Murray CJL. Tracking progress towards universal childhood immunisation and the impact of global initiatives: a systematic analysis of three-dose diphtheria, tetanus, and pertussis immunisation coverage. The Lancet 2008; 372(9655): 2031-46. 12. Murray CJL, Shengelia B, Gupta N, Moussavi S, Tandon A, Thieren M. Validity of reported vaccination coverage in 45 countries. The Lancet 2003; 362(9389): 1022-7. 13. Boerma T, Victora C, Abouzahr C. Monitoring country progress and achievements by making global predictions: is the tail wagging the dog? Lancet 2018. 14. AbouZahr C, Boerma T, Hogan D. Global estimates of country health indicators: useful, unnecessary, inevitable? Global health action 2017; 10(sup1): 1290370. 15. World Health Organization. WHO Guidelines for Screening and Treatment of Precancerous Lesions for Cervical Cancer Prevention. Geneva: WHO, 2013. 16. Independent UK Panel on Breast Cancer Screening. The benefits and harms of breast cancer screening: an independent review. The Lancet 2012; 380(9855): 1778-86. 17. International Agency for Research on Cancer. Cancer Screening in the European Union. Report on the implementation of the Council Recommendation on Cancer Screening. Lyon, France: International Agency for Research on Cancer, 2017. 34 18. Fischer Walker CL, Fontaine O, Black RE. Measuring Coverage in MNCH: Current Indicators for Measuring Coverage of Diarrhea Treatment Interventions and Opportunities for Improvement. PLOS Medicine 2013; 10(5): e1001385. 19. Victora CG, Requejo JH, Barros AJ, Berman P, Bhutta Z, Boerma T, Chopra M, de Francisco A, Daelmans B, Hazel E, Lawn J, Maliqi B, Newby H, Bryce J. Countdown to 2015: a decade of tracking progress for maternal, newborn, and child survival. Lancet 2015. 20. Kaplan A, Nakhimovsky S, Won A, Avila C. An Analysis of Spending on Prevention and Public Health in Africa. Bethesda, MD: Abt Associates Inc.; 2013. 21. Kakwani N, Wagstaff A, Van Doorslaer E. Socioeconomic inequalities in health: Measurement, computation and statistical inference. Journal of Econometrics 1997; 77(1): 87-104. 22. Wagstaff A, Paci P, Van Doorslaer E. On the Measurement of Inequalities in Health. Social Science & Medicine 1991; 33(5): 545-57. 23. OECD. A System of Health Accounts 2011. Paris: OECD; 2011. 24. Wagstaff A, Eozenou P, Neelsen S, Smitz M-F. Introducing the World Bank's 2018 Health Equity and Financial Protection Indicators database. The Lancet Global Health 2018. 25. van Leth F, Guilatco RS, Hossain S, van't Hoog AH, Hoa NB, van der Werf MJ, Lönnroth K. Measuring socio-economic data in tuberculosis prevalence surveys. The International Journal of Tuberculosis and Lung Disease 2011; 15(6): S58-S63. 26. Lozano R, Soliz P, Gakidou E, Abbott-Klafter J, Feehan DM, Vidal C, Ortiz JP, Murray CJL. Benchmarking of performance of Mexican states with effective coverage. The Lancet 2006; 368(9548): 1729-41. 27. Hodgins S, D'Agostino A. The quality-coverage gap in antenatal care: toward better measurement of effective coverage. Global health, science and practice 2014; 2(2): 173- 81. 28. World Health Organization. Service Availability and Readiness Assessment (SARA): An Annual Monitoring System for Service Selivery - Reference Manual. Geneva: WHO; 2013. Annex 1: Computing UHC Trends If we had the relevant data, we could compute the annual average percentage point change in a country’s UHC index from the coefficient b in the equation (1) = + The problem is we cannot compute UHC for many country-year combinations, since the data for different indicators often come from different surveys conducted at different dates. Rarely, if ever, do we have all 9 UHC indicators for the same year for a given country, let alone for several years. So, we cannot compute UHC for multiple years for each country, and therefore we cannot run the above regression. What we can do instead is compute the rate of change of each of the components of UHC and then infer b in the equation above. This can be done as follows. Given our assumption about the UHC index, we can write: (2) ≡ + lnSC and lnFP change over time, so we can write: (3) = + (4) = + Substitute eqns (3) and (4) in eqn (2) to get: (5) = ( + ) + ( + ) = + So, given and are assumptions, once we estimate and using eqns (2) and (3), we can compute b as + . Of course, while we can estimate eqn (4) if FP is captured simply by the incidence of catastrophic expenditures, we cannot estimate eqn (3), because we cannot compute SC because some indicators come from some surveys and some from others. But the same procedure can be used. We want the coefficient bSC in eqn (3). Suppose for simplicity that SC is a weighted average of just two SC indicators SC1 and SC2. (The extension to eight is immediate.) We have: (6) ≡ + Both change over time, so we can write: (7) = + (8) = + Substituting eqns (7) and (8) into eqn (6) gives us (9) = ( + ) + ( + )= + So, given and are assumptions, once we estimate and using eqns (7) and (8), we can compute bSC as + . Annex 2: Regression Analysis We estimate the following regression for the individual components of the UHC index: (1) = + + + + ∙ + + ∙ + + ∙ + + ∙ + + ∙ + + ∙ + + , where Y is the outcome variable (i.e. the component of the UHC index), GDP is gross domestic product per capita, THE is total health expenditure, SHI is social health insurance, GS is government schemes, CPI is compulsory health insurance, NPS is nonprofit schemes, VHI is voluntary health insurance, i denotes the country, t the year, the ’s are coefficients, i is a country-specific fixed effect, and it is a random error term. We report the marginal effects of the various components of the index, evaluated at mean GDP per capita. For example, in the case of the share of THE in GDP, we report: (2) = + . For the UHC index as a whole, we have just one observation per country. We estimate eqn(1) above but without the country-specific fixed effects, and report the marginal effects as above. Annex 3: Datapoints Included in Sample Not Included in the HEFPI Database Our sample includes a number of datapoints – listed in Table A3.1 – which are not part of the 2019 HEFPI dataset:  Two datapoints – Austria1 and Hong Kong2 – on catastrophic expenditure from reputable studies where, unlike for all other catastrophic expenditure data points, we did not have access to the micro-data;  Datapoints which are not representative at the national level: o From published articles: the diarrhea treatment data point from China 3; the mammography data points for Albania4, Armenia5, and Iran6; the pap smear data points for Cambodia7, Iran8, and Nigeria9; o From countries where the WHS did not draw a nationally representative sample,10 namely the inpatient admission data point from Congo Republic; the antenatal care visits, skilled birth attendance, ARI treatment, pap smear and mammogram data points from China; the mammogram datapoints from Australia (the Australian WHS used a drop-and-collect survey in some parts of the country and a telephone survey in others; We only use data from the drop-and-collect survey in our analysis), the Comoros, Congo Republic, and Ivory Coast; and the pap smear data points from Australia, China, the Comoros, Congo Republic, and Ivory Coast.  Two mammography data points where we assume population means based on circumstantial evidence: o An overview of oncology in Cambodia 11 reports that no screening program exists for either cervical cancer or breast cancer. Since the study we use for i our cervical cancer screening rate7 found that despite this lack of program for either type of cancer, 7% of women had ever had a pap smear, and since breast cancer screening rates are typically lower than cervical cancer screening rates, we assume 2.5% of women have ever had a mammogram in Cambodia. o For Nigeria, a recent study12 found that only 2.8% of women over the age of 40 had ever had a mammogram, while an RCT13 aimed at increasing the uptake of breast (and cervical) cancer screening reported a baseline value of 4.4%. We have therefore assumed that 5% of women have ever been screened for breast cancer in Nigeria. Under assumptions detailed in the HEFPI database’s methodological background paper14, we adjust the Nigeria and Cambodia mammography data to 2-year recall periods, obtaining rates of 0.5% and 0.25%, respectively.  Six datapoints that use indicator definitions which are different from the rest of the sample: o The ARI treatment data points for China, Mauritius, and Russia come from the WHS which asks respondents for their first response to a child’s respiratory problems – i.e. not for all responses like the DHS and MICS surveys where the rest of our ARI treatment datapoints come from; o The inpatient admission data point from Uganda is generated from a household level indicator of inpatient admission – and not individual level data like the rest of our inpatient care data points. To compute the data point, we assume that the total number of admissions for households reporting an ii inpatient admission over the last year is one. Under the additional assumption that the admitted person was an adult, division by the number of adults in the household generates the average number of inpatient admissions per adult household members over the last year, which we subsequently aggregate to the country level. The resulting number overstates the true adult inpatient admission probability to the degree that children are admitted and understates it to the degree that households reporting any admissions over the previous year had more than one admission. o The diarrhea treatment data points from the Mauritius and Russia 2003 WHS, where treatment is defined as visiting a healthcare provider as the first response to the child’s diarrhea, rather than as giving ORS like for the rest of our sample.  Finally, our sample includes 64 data points which were dropped from the HEFPI dataset for lying below our HEFPI sample size threshold of 100 observations. iii Table A3.1: Data points included in sample which are not in HEFPI database Differi N<1 Sub- ng Publicat Indicator Country Year Survey 00 national definiti ion on Catastrophic Austria 2009 HBS Y Expenditures Catastrophic Household Hong Kong 1999 Y Expenditures Expenditure Survey Bosnia and 4+ ANC visits 2003 WHS Y Herzegovina 4+ ANC visits China 2002 WHS Y Y 4+ ANC visits Georgia 2003 WHS Y 4+ ANC visits Russia 2003 WHS Y 4+ ANC visits Ukraine 2002 WHS Y ARI treatment Albania 2005 MICS Y ARI treatment Armenia 2015 DHS Y ARI treatment Azerbaijan 2000 MICS Y ARI treatment Azerbaijan 2006 DHS Y Bosnia and ARI treatment 2000 MICS Y Herzegovina Bosnia and ARI treatment 2011 MICS Y Herzegovina ARI treatment China 2002 WHS Y Y Y ARI treatment Georgia 2005 MICS Y ARI treatment Jamaica 2000 MICS Y ARI treatment Jamaica 2005 MICS Y ARI treatment Jamaica 2011 MICS Y ARI treatment Kazakhstan 1995 DHS Y ARI treatment Kazakhstan 1999 DHS Y ARI treatment Kazakhstan 2006 MICS Y ARI treatment Lao PDR 2000 MICS Y ARI treatment Moldova 2000 MICS Y ARI treatment Moldova 2012 MICS Y ARI treatment Mongolia 2010 MICS Y ARI treatment Mauritius 2003 WHS Y Y ARI treatment Russia 2003 WHS Y Y ARI treatment Tajikistan 2000 MICS Y ARI treatment Tajikistan 2005 MICS Y ARI treatment Tajikistan 2017 MICS Y Trinidad and ARI treatment 2006 MICS Y Tobago Trinidad and ARI treatment 2011 MICS Y Tobago ARI treatment Uzbekistan 1996 DHS Y Diarrhea treatment Albania 2005 MICS Y Diarrhea treatment Albania 2008 DHS Y iv Differi N<1 Sub- ng Publicat Indicator Country Year Survey 00 national definiti ion on Diarrhea treatment Armenia 2015 DHS Y Diarrhea treatment China 2005 Specialized Y Y Diarrhea treatment Jamaica 2005 MICS Y Diarrhea treatment Jamaica 2011 MICS Y Diarrhea treatment Jamaica 2011 MICS Y Diarrhea treatment Moldova 2000 MICS Y Diarrhea treatment Mauritius 2003 WHS Y Y Diarrhea treatment Russia 2003 WHS Y Y Trinidad and Diarrhea treatment 2002 MICS Y Tobago Trinidad and Diarrhea treatment 2006 MICS Y Tobago Trinidad and Diarrhea treatment 2011 MICS Y Tobago Diarrhea treatment Uzbekistan 1996 DHS Y Dominican Full immunization 1999 DHS Y Republic Russian Full immunization 1996 RLMS Y Federation Russian Full immunization 1998 RLMS Y Federation Russian Full immunization 2000 RLMS Y Federation Russian Full immunization 2001 RLMS Y Federation Russian Full immunization 2002 RLMS Y Federation Russian Full immunization 2003 RLMS Y Federation Russian Full immunization 2004 RLMS Y Federation Russian Full immunization 2005 RLMS Y Federation Russian Full immunization 2006 RLMS Y Federation Russian Full immunization 2007 RLMS Y Federation Russian Full immunization 2008 RLMS Y Federation Russian Full immunization 2009 RLMS Y Federation Inpatient admission Congo, Rep. 2003 WHS Y National Service Inpatient admission Uganda 2008 Y Delivery Survey Mammography Albania 2016 Specialized Y Y Mammography Armenia 1999 Specialized Y Y Mammography Australia 2003 WHS Y Mammography China 2002 WHS Y Mammography Cambodia 2012 Assumed Y v Differi N<1 Sub- ng Publicat Indicator Country Year Survey 00 national definiti ion on Mammography Comoros 2003 WHS Y Mammography Congo, Rep. 2003 WHS Y Mammography Cote d'Ivoire 2003 WHS Y Mammography Iran, Islamic Rep. 2011 Specialized Y Y Mammography Luxembourg 2003 WHS Y Mammography Nigeria 2017 Assumed Y Mammography Slovenia 2003 WHS Y Pap smear Australia 2003 WHS Y Pap smear Cambodia 2016 Specialized Y Y Pap smear China 2002 WHS Y Pap smear Comoros 2003 WHS Y Pap smear Congo, Rep. 2003 WHS Y Pap smear Cote d'Ivoire 2003 WHS Y Pap smear Iran, Islamic Rep. 2009 Specialized Y Y Pap smear Nigeria 2013 Specialized Y Y Skilled birth Bosnia and 2003 WHS Y attendance Herzegovina Skilled birth China 2002 WHS Y Y attendance Skilled birth Georgia 2003 WHS Y attendance Skilled birth Russia 2003 WHS Y attendance Skilled birth Ukraine 2002 WHS Y attendance References 1. Sanwald A, Theurl E. Out-of-pocket payments in the Austrian healthcare system–a distributional analysis. International journal for equity in health 2015; 14(1): 94. 2. van Doorslaer E, O'Donnell O, Rannan-Eliya RP, et al. Catastrophic payments for health care in Asia. Health economics 2007; 16(11): 1159-84. 3. Gao W, Yan H, Wang D, Dang S. Oral rehydration salt use and its correlates in low-level care of diarrhea among children under 36 months old in rural Western China. BMC public health 2013; 13(1): 238. 4. Kamberi F, Sinaj E, Lalo R. Breast cancer health beliefs and the use of mammography among women randomly selected in Vlora, Albania. EC Gynaecology 2017; 4(6): 204-11. 5. Harutyunyan T. Breast self-examination practice in Yerevan women: American University of Armenia (AUA); 2014. 6. Moodi M, Rezaeian M, Mostafavi F, Sharifirad G-R. Determinants of mammography screening behavior in Iranian women: A population-based study. Journal of research in medical sciences: the official journal of Isfahan University of Medical Sciences 2012; 17(8): 750. 7. Touch S, Oh J-K. Knowledge, attitudes, and practices toward cervical cancer prevention among women in Kampong Speu Province, Cambodia. BMC Cancer 2018; 18(1): 294. vi 8. Jalilian F, Emdadi S. Factors related to regular undergoing Pap-smear test: application of theory of planned behavior. Journal of research in health sciences 2011; 11(2): 103-8. 9. Ekine A, West O, Gani O. Awareness of female health workers and non health workers on cervical cancer and cervical cancer screening: south–south, Nigeria. Int J Med Sci Clin Inventions 2015; 2(2): 713-25. 10. World Health Organization. World report on disability 2011: Technical appendix C, Design and implementation of the World Health Survey. Geneva: World Health Organization; 2011. 11. Eav S, Schraub S, Dufour P, Taisant D, Ra C, Bunda P. Oncology in Cambodia. Oncology 2012; 82(5): 269-74. 12. Olasehinde O, Boutin-Foster C, Alatise OI, et al. Developing a Breast Cancer Screening Program in Nigeria: Evaluating Current Practices, Perceptions, and Possible Barriers. Journal of Global Oncology 2017; 3(5): 490-6. 13. Chigbu CO, Onyebuchi AK, Onyeka TC, Odugu BU, Dim CC. The impact of community health educators on uptake of cervical and breast cancer prevention services in Nigeria. International Journal of Gynecology & Obstetrics 2017; 137(3): 319-24. 14. Wagstaff A, Eozenou P, Neelsen S, Smitz M. The 2018 health equity and financial protection indicators database: overview and insights. Policy Research Working Paper 2018; 8577. vii