Health Policy and Planning Advance Access published December 26, 2016

Health Policy and Planning, 2016, 1–6

doi: 10.1093/heapol/czw156
Original Manuscript

National income inequality and ineffective

health insurance in 35 low- and middle-income
countries
Francisco N. Alvarez,1,2 Abdulrahman M. El-Sayed3
1
College of Physicians and Surgeons, Columbia University, New York, NY, USA, 2Detroit Health Department, USA and
3
Detroit Health Department, City of Detroit, Detroit, MI, USA

Downloaded from http://heapol.oxfordjournals.org/ at University of Newcastle on January 3, 2017

Corresponding Author: Francisco N. Alvarez, Department of Medicine, Johns Hopkins University School of Medicine,
1830 E Monument Street, Baltimore, MD 21287, USA. E-mail: falvarez@jhmi.edu
Accepted on 19 October 2016

Abstract
Global health policy efforts to improve health and reduce financial burden of disease in low- and
middle-income countries (LMIC) has fuelled interest in expanding access to health insurance cover-
age to all, a movement known as Universal Health Coverage (UHC). Ineffective insurance is a meas-
ure of failure to achieve the intended outcomes of health insurance among those who nominally
have insurance. This study aimed to evaluate the relation between national-level income inequality
and the prevalence of ineffective insurance. We used Standardized World Income Inequality
Database (SWIID) Gini coefficients for 35 LMICs and World Health Survey (WHS) data about insur-
ance from 2002 to 2004 to fit multivariable regression models of the prevalence of ineffective insur-
ance on national Gini coefficients, adjusting for GDP per capita. Greater inequality predicted higher
prevalence of ineffective insurance. When stratifying by individual-level covariates, higher inequal-
ity was associated with greater ineffective insurance among sub-groups traditionally considered
more privileged: youth, men, higher education, urban residence and the wealthiest quintile.
Stratifying by World Bank country income classification, higher inequality was associated with inef-
fective insurance among upper-middle income countries but not low- or lower-middle income
countries. We hypothesize that these associations may be due to the imprint of underlying social
inequalities as countries approach decreasing marginal returns on improved health insurance by
income. Our findings suggest that beyond national income, income inequality may predict differ-
ences in the quality of insurance, with implications for efforts to achieve UHC.

Keywords: Health insurance, income inequality, ineffective insurance, LMIC, universal health coverage

Key Messages

• Greater income inequality predicted higher prevalence of ineffective health insurance when adjusting for GDP per capita.
• The association between income inequality and ineffective health insurance is stronger among sub-groups traditionally
considered more privileged: youth, men, higher education, urban residence, wealthiest quintile.
• Stratifying by country income, higher income inequality is associated with higher ineffective health insurance among
upper-middle income countries but not low- or lower-middle income countries.
• We hypothesize that these associations may be due to the imprint of underlying social inequalities as developing coun-
tries approach decreasing marginal returns on improved health insurance by income.

C The Author 2016. Published by Oxford University Press in association with The London School of Hygiene and Tropical Medicine.
V
All rights reserved. For permissions, please e-mail: journals.permissions@oup.com 1
2 Health Policy and Planning, 2016, Vol. 0, No. 0

Introduction Penn World Tables (PWT). Income inequality was assessed utilizing
Gini coefficients for the 35 countries in the WHS datasets obtained
Global health policy efforts to improve health and reduce financial
from the SWIID version 5.0 (Solt 2014). As our exposure variable,
burden of disease in low- and middle-income countries (LMIC) has
Gini coefficients of net inequality for each country were averaged
fuelled interest in expanding access to health insurance coverage to
between 1999 and 2001, the three years before which we had out-
all, a movement for Universal Health Coverage (UHC). The World
come variable data (ineffective insurance as assessed between 2002
Health Report 2010 defines UHC as a health system where all peo-
and 2004). Averages of three years were utilized to reflect the time
ple in need of health services (promotion, prevention, treatment, re-
period of the outcome variable and minimize yearly variation.
habilitation and palliation) receive them without undue financial
National GDP data were obtained from the PWT version 8.1 to
hardship (WHO 2010). Subsequent research, including case studies
control for national wealth (Feenstra et al. 2015). The GDP per cap-
in 13 countries, has underwritten efforts to establish UHC as an out-
ita variable was generated utilizing real GDP at constant 2005 na-
come for the post-2015 development goals (Boerma et al. 2014, UN
tional prices (mil. 2005US$) divided by population (in millions) for
General Assembly 2015).
each country in each year as provided by the PWT. GDP per capita
There is substantial interest in the capacity for UHC to reflect its
was then averaged between 1999 and 2001 for the 35 countries of
goals, particularly with respect to improving health equity (Giedion
interest to match the exposure variable.
et al. 2013). For example, one study of Mexico’s new public insur-
The WHS is a cross-sectional study conducted by the World
ance program found heterogeneity in financial protection for rural
Health Organization (WHO) between 2002 and 2004 in 70 countries

Downloaded from http://heapol.oxfordjournals.org/ at University of Newcastle on January 3, 2017

residents based upon existing access to well-staffed health facilities
representing each United Nations sub-region and comprising over
(Grogger et al. 2015). ‘Ineffective insurance’ is defined as nominal
300 000 individuals with the aim of compiling comprehensive base-
insurance coverage that does not protect beneficiaries from having
line population health information, monitoring health outcomes and
to borrow or sell personal items to pay for health services, having an
informing future health system investments. The survey was con-
untreated medical condition, or delivering a child outside a skilled
ducted at the household level and used sampling weights for national
health facility (El-Sayed et al. 2016). Using World Health Survey
representation to allow country-level inference. Household-level
data of 42 LMICs, they found a 13% prevalence of ineffective insur-
characteristics, including insurance and wealth and individual-level
ance, of which 69% had to borrow or sell personal items to pay for
characteristics, including socio-demographic information, health
healthcare. Furthermore, ineffective insurance was associated with
state descriptions, health care utilization and health system respon-
lower household wealth and residence in rural areas.
siveness, were collected. The SWIID attempts to develop a cross-
Concurrently, there has been renewed interest in income inequal-
nationally comparable database of Gini indices across time. It cur-
ity and its association with social outcomes, including economic
rently includes Gini estimates for gross and net income inequality for
growth, political stability and health and social well-being (Ortiz
171 countries from 1960 to 2009 utilizing all major sources of in-
and Cummins 2011). The relation between income inequality and
equality data including the United Nations University’s World
population health (mortality rates, life expectancy) has been well-
Income Inequality Database (UNU-WIDER WIID), the World
documented, independent of individual-level socioeconomic status
Bank’s PovcalNet, the OECD Income Distribution Database, the
or societal-level wealth (Biggs et al. 2010, De Vogli et al. 2005,
World Top Incomes Database, as well as national statistical offices
Wilkinson and Pickett 2008, Kondo et al. 2009). However, the de-
and other sources (Solt 2009). The PWT provides data on real GDP
bate about the potential causal effect of income inequality on popu-
to measure standard of living across countries (Feenstra et al. 2015).
lation health persists. First, the association may be confounded by
Utilizing prices collected by the International Comparisons Program
education, race/ethnicity, or individual income. Second, reverse
to construct purchasing power parity (PPP) exchange rates, the PWT
causation may occur through effects of good health that contribute
converts GDP at national prices to a common currency (U.S. dollars)
to higher socioeconomic status (e.g. increased earnings, educational
to make them comparable across countries. Data is available for 167
attainment and labour force participation) and may in turn contrib-
countries between 1950 and 2011.
ute to societal income inequality (Ellison 2002, Gravelle 1998,
The initial WHS sample includes 288 431 households in 70
Lynch et al. 2004). However, authors have argued that these alter-
countries. As our focus was LMICs, our analysis excluded those
native explanations can only explain a portion of the observed rela-
countries classified as high-income by 2003 World Bank country in-
tion and that societal-level income inequality remains causally
come classifications (n ¼ 20) (World Bank 2003). We also excluded
predictive of poor health (Babones 2008, Bezruchka et al. 2008,
countries that did not provide WHS survey weights (n ¼ 1;
Pickett and Wilkinson 2015, Zheng 2012).
Guatemala), did not collect insurance information (n ¼ 1; Latvia), or
Access to healthcare may be one causal pathway by which in-
that do not have data in the SWIID or PWT during the years of
come inequality may influence population health – and insurance is
interest (n ¼ 4; Chad, Comoros, Republic of the Congo, Myanmar).
an important predictor of health access. Therefore, we were inter-
Ineffective insurance includes, by definition, a subset of only the
ested in the relation between income inequality and the quality of in-
population that has health insurance. Therefore, we excluded coun-
surance and sought to examine the relation between national-level
tries that had a negligible prevalence of health insurance among
income inequality and ineffective health insurance in LMICs. This
those surveyed, which we define as a prevalence of health insurance
work may both shed light on the relation between income inequality
<1% among the un-weighted country WHS survey population
and population health and improve our understanding of potential
(n ¼ 9; Bangladesh, Burkina Faso, Ethiopia, Laos, Malawi, Mali,
policy challenges to achieving UHC.
Mauritania, Pakistan, Swaziland). The final sample therefore
included 35 countries of which 10 were low income, 16 lower-
middle income and 9 upper-middle income (Table 1). In addition,
Materials and Methods
Turkey was excluded for sub-analyses regarding marital and educa-
Data tional status due to missing WHS household data.
We used data from the World Health Survey (WHS), the Individual country-level data were extracted using the dataset
Standardized World Income Inequality Database (SWIID), and the design published previously (El-Sayed et al. 2016). Briefly,
Health Policy and Planning, 2016, Vol. 0, No. 0 3

Table 1 Characteristics of countries included in study sample

Country World Bank income Nominal insurance Ineffective insurance Total under-insurance Unweighted
classification prevalence prevalence prevalence (uninsured þ n*
(2003) (weighted %) (weighted %) ineffectively insured) (weighted %)

Côte d’Ivoire Low 8.88 15.47 92.49 2496

Georgia 100.00 30.81 30.81 2692
Ghana 2.40 22.75 98.15 3346
India 2.01 22.69 98.45 7340
Kenya 11.17 15.32 90.54 4067
Nepal 6.89 60.17 97.26 305
Senegal 3.38 6.85 96.85 998
Vietnam 22.25 11.18 80.24 3677
Zambia 3.20 12.46 97.20 3914
Zimbabwe 7.62 26.38 94.39 3620

Bosnia and Herzegovina Lower-Middle 89.86 5.48 15.06 1005

Brazil 100.00 12.71 12.71 450

Downloaded from http://heapol.oxfordjournals.org/ at University of Newcastle on January 3, 2017

China (People’s Republic of) 49.95 5.71 52.90 3915
Dominican Republic 21.89 16.22 81.66 4738
Ecuador 16.90 12.13 85.16 1605
Kazakhstan 99.32 11.63 12.23 4332
Morocco 19.37 25.73 85.61 2113
Namibia 13.51 11.24 88.02 3842
Paraguay 22.54 10.83 79.90 5221
Philippines 19.24 29.25 86.39 9913
Russian Federation 98.55 9.46 10.79 4233
South Africa 17.38 15.26 85.28 1849
Sri Lanka 3.97 14.42 96.60 4751
Tunisia 84.53 25.93 37.39 4880
Turkey 97.51 13.74 15.89 8303
Ukraine 5.01 4.49 95.21 1080

Croatia Upper-Middle 98.39 5.27 6.80 956

Czech Republic 100.00 3.72 3.72 849
Estonia 96.21 6.19 9.75 924
Hungary 99.82 3.67 3.84 583
Malaysia 40.89 5.65 61.42 5873
Mauritius 6.93 5.49 93.45 3763
Mexico 43.81 12.21 61.54 38292
Slovak Republic 99.67 2.40 2.72 1613
Uruguay 92.72 5.20 12.13 2835

*Unweighted n: Number of households by country from which health insurance data was extracted from World Health Survey.

households were considered nominally insured if they answered Analysis

affirmatively regarding the insurance status of the primary We performed logarithmic transformations on all variables to correct
household respondent (‘Is this person covered by any kind of for non-normal distribution of the data (not shown). First, we fit mul-
health insurance plan?’). Nominally insured respondents were tivariable linear regression models of ineffective insurance on Gini co-
considered ineffectively insured if they fulfilled any of the follow- efficient, adjusting for GDP per capita, and on each constituent
ing criteria: lack of treatment for a diagnosed chronic condition, component of ineffective insurance. Second, we fit further multivari-
failure to deliver a child in a health facility (women only), or sale able linear regression models of ineffective insurance by the Gini coef-
of household assets or borrowing money from someone other ficient stratified by each of our covariates. Third, we fit multivariable
than a friend or family member in order to pay for health care in linear regression models of ineffective insurance by the Gini coefficient
the past year. In addition, we also collected the following covari- stratified by World Bank country income classifications in 2003.
ates: age in years (categorical: 13–34, 34–65, 65þ), sex, marital The analysis utilized secondary data in the public domain and
status (binary: married/cohabitating versus other), education was therefore exempt from IRB review requirements. All analyses
(binary: any secondary education versus no secondary educa- were conducted using Stata version 14 (StataCorp, College Station,
tion), urban residence (binary: urban versus rural), and country- TX, USA).
specific wealth quintile (Hosseinpoor et al. 2014). Within-
country relative wealth indices were created within each country
using principal components analysis of country-specific house-
hold asset questions and subsequently divided into quintiles as Results
described previously (Filmer and Pritchett 2001). National sur- Multivariable linear regression models of ineffective insurance on
vey sampling weights as provided by the WHS were used to allow Gini coefficients adjusting for GDP per capita are shown in Table 2.
for country-level inferences. Simple regression of the prevalence of ineffective insurance on Gini
4 Health Policy and Planning, 2016, Vol. 0, No. 0

Table 2 Survey-weighted multivariable linear regression models of ineffective health insurance and its sub-categories on national income
inequality

Simple regression Multiple regression

Gini coefficient Gini coefficient GDP per capita

Coefficient [95% CI] Coefficient [95% CI] Coefficient [95% CI]

Ineffective insurance 1.759 [0.844, 2.673] ** 1.274 [0.429, 2.119] ** 0.437 [0.696, 0.178] **
Sold/borrowed 2.573 [1.270, 3.877] *** 1.984 [0.720, 3.248] ** 0.531 [0.919, 0.143] **
Untreated chronic condition 0.901 [0.080, 1.882] 0.736 [0.310, 1.782] 0.148 [0.469, 0.173]
Non-facility delivery 3.260 [0.692, 5.828] * 1.491 [0.559, 3.540] 1.595 [2.224, 0.967] ***

*P  0.05;
**P  0.01;
***P  0.001.

Downloaded from http://heapol.oxfordjournals.org/ at University of Newcastle on January 3, 2017

Table 3 Survey-weighted multivariable linear regression models of ineffective health insurance on national income inequality stratified by
covariates

Simple regression Multiple regression

Gini coefficient Gini coefficient GDP per capita

Coefficient [95% CI] Coefficient [95% CI] Coefficient [95% CI]

Age 13–34 years 3.196 [1.629, 4.764] *** 2.731 [1.110, 4.353] ** 0.419 [0.916, 0.078]
35–65 years 1.471 [0.561, 2.382] ** 1.011 [0.156, 1.866] * 0.415 [0.678, 0.153] **
>65 years 1.306 [1.945, 4.557] 2.245 [1.126, 5.616] 0.847 [0.187, 1.881]
Sex Female 1.122 [0.091, 2.152] * 0.555 [0.384, 1.493] 0.511 [0.799, 0.223] ***
Male 2.473 [1.005, 3.942] ** 2.239 [0.671, 3.807] ** 0.212 [0.692, 0.269]
Marital Status Unmarried 1.872 [0.988, 2.757] *** 1.495 [0.624, 2.366] *** 0.339 [0.609, 0.069] *
Married/Cohabitating 1.648 [0.670, 2.626] ** 1.109 [0.216, 2.003] * 0.484 [0.761, 0.207] ***
Education No secondary education 1.957 [0.406, 3.507] * 1.538 [0.083, 3.160] 0.375 [0.878, 0.127]
Any secondary education 1.891 [0.346, 3.436] * 1.856 [0.181, 3.530] * 0.032 [0.551, 0.488]
Residence Rural 2.391 [1.331, 3.450] *** 1.795 [0.838, 2.751] *** 0.537 [0.830, 0.244] ***
Urban 1.940 [0.463, 3.417] * 1.707 [0.129, 3.285] * 0.210 [0.694, 0.274]
Wealth Poorest 20% 0.192 [3.176, 2.791] 0.151 [3.376, 3.074] 0.037 [0.952, 1.026]
Second poorest 20% 2.137 [0.090, 4.184]* 2.519 [0.342, 4.695]* 0.344 [0.323, 1.012]
Middle 20% 2.309 [0.269, 4.349]* 2.635 [0.457, 4.814]* 0.294 [0.374, 0.962]
Second wealthiest 20% 1.892 [0.285, 3.499]* 1.354 [0.287, 2.995] 0.485 [0.988, 0.018]
Wealthiest 20% 3.428 [1.693, 5.162] *** 3.182 [1.326, 5.039]*** 0.221 [0.790, 0.348]

*P  0.05;
**P  0.01;
***P  0.001.

coefficient showed a significant association (P  0.001) with a coef- associated with prevalence of ineffective insurance among all demo-
ficient of 1.759 (95% CI: 0.844–2.673). After adjusting for GDP graphic subgroups except those >65 years old and the poorest
per capita the association remained significant (P  0.01) with a co- wealth quintile. After adjusting for GDP per capita, associations
efficient of 1.274 (95% CI: 0.429–2.119). continued to be significant among 13–34 year-olds and 35–65 year-
Linear regression models were also fit for each sub-category of olds, males, those of both married and unmarried status, those with
ineffective insurance (sold/borrowed assets or money, untreated for any secondary education, both urban and rural residence, as well as
chronic condition and child delivery outside health facility). In the the second-poorest, middle and wealthiest quintile.
simple regression model national Gini coefficients were significantly Table 4 shows simple and multivariable regression models of the
associated with selling or borrowing assets or money to pay for prevalence of ineffective insurance on Gini coefficients stratified by
healthcare expenditures (P  0.001) and delivering a child outside a World Bank country income classifications in 2003. Gini coeffi-
health facility (P  0.05). Upon adjusting for GDP per capita, the as- cients were significantly associated with ineffective insurance only
sociation with selling or borrowing assets or money to pay for among upper-middle income countries (P  0.05). As expected,
healthcare expenditures remained significant (P  0.01) with a coef- GDP per capita was not significantly associated with ineffective in-
ficient of 1.984 (95% CI: 0.720–3.248). surance when stratifying by country income classification.
Table 3 shows simple and multivariable regression models of
prevalence of income inequality on Gini coefficients stratified by
demographic covariates known to affect health and healthcare util- Discussion
ization (age, sex, marital status, education, residence, wealth quin- This study considered the influence of income inequality on the
tile). In a simple regression model, Gini coefficients were directly quality of health insurance in LMICs. We found that income
Health Policy and Planning, 2016, Vol. 0, No. 0 5

Table 4 Survey-weighted multivariable linear regression models of ineffective health insurance on national income inequality stratified by
World Bank country income status

Simple regression Multiple regression

Gini coefficient Gini coefficient GDP per capita

Coefficient [95% CI] Coefficient [95% CI] Coefficient [95% CI]

Low income 0.770 [3.085, 4.625] 0.740 [3.500, 4.980] 0.090 [1.209, 1.388]
Lower-middle income 0.880 [0.427, 2.188] 0.899 [0.489, 2.286] 0.067 [0.966, 0.831]
Upper-middle income 1.287 [0.343, 2.231]* 1.715 [0.338, 3.092]* 0.912 [1.177, 3.001]

*P  0.05;
**P  0.01;
***P  0.001.

inequality was directly associated with ineffective insurance after ad- This is supported by our finding that upon stratifying the associ-

Downloaded from http://heapol.oxfordjournals.org/ at University of Newcastle on January 3, 2017

justing for national income per capita. This was particularly true ation between income inequality and health insurance by individual-
among the more privileged: younger people, men, the more highly level markers of socioeconomic well-being, the influence of income
educated and the wealthiest quintile. Further, we found that this as- inequality was more substantial among privileged groups: younger
sociation was most robust among the wealthiest LMICs. people, men, the more educated and the wealthiest. Among higher-
The general results are consistent with prior studies reporting income societies, where the effects of income inequality on ineffect-
GDP-per-capita-adjusted Gini coefficients as predictors of formal ive insurance is more pronounced, standardized state benefits are af-
health insurance coverage across countries (Feigl and Ding 2013). forded primarily to the poor. This could leave the relatively more
These results are also consistent with the finding that, in the United privileged more exposed to the negative influences of income in-
States, Gini index tracks inversely with life expectancy among the equality. Alternatively, those more poised to reap benefits from a
highest income quartile but not among the bottom income quartile more egalitarian distribution of wealth may be currently disadvan-
(Chetty et al. 2016). Our findings are further supported by another taged members of historically privileged groups, thus allowing them
study showing that the relation between inequality by U.S. state and to more acutely respond to changes in income inequality.
poor self-rated health was stronger among advantaged sub-groups, Research on causal mechanisms of the effect of income inequal-
including a stronger relationship among whites versus blacks and ity on health has focused on a hypothesized pathway of socioeco-
wealthier rather than poorer individuals (Subramanian and Kawachi nomic status differentiation mediated by effects of relative
2006). Similarly, living in states with high income inequality was deprivation and upward social comparisons as well as erosion of so-
associated with increased mortality among people with high in- cial cohesion (Deaton 2003, Kondo et al. 2009, Pickett and
comes but not low incomes (Lochner et al. 2001). Wilkinson 2015). Furthermore, the causal link from income inequal-
We propose that the association between income inequality and ity to health has often been contested among health economists,
ineffective insurance is initially mediated by increases in health insur- who allow only an effect mediated by the purchase of health care
ance with increasing wealth, and therefore, health insurance quality (Deaton 2003). This study provides evidence of another policy-
may be driven largely by increasing societal wealth, as represented by mediated link between income inequality and health. Inferior health
GDP per capita. However, countries eventually arrive at a relative insurance may lead to poorer health outcomes (via decreased access
plateau of decreasing marginal returns on health insurance quality to care, lapses in financial protection, etc.). The quality of health in-
from increases in societal wealth. At this stage, remaining between- surance may have structural social underpinnings independent of
country inequalities in the quality of insurance can be predicted by na- the ability of individuals to purchase health insurance. This effect
tional income inequality and may reflect underlying social inequality. may be one of multiple possible causal pathways by which socioeco-
In other words, at lower societal wealth levels, quality of insurance is nomic status differentiation in unequal societies affects health.
likely to track more or less directly with societal wealth. At higher so- This study had numerous limitations largely resulting from the
cietal wealth levels, where a basic set of health services can be af- source data and exigencies of study design. Since income inequality,
forded by a substantial portion of the population, other factors and therefore Gini coefficient, is a societal-level variable, the ana-
become more important as predictors of insurance quality, such as dis- lysis must be done on a societal—not household—level, and the
tance to a health centre, access to transportation, and employment sta- power from having 288 431 households in the WHS was lost. The
tus. These other factors may be more sensitive to societal-level WHS survey included 70 countries, and after excluding the 20
inequalities. Furthermore, higher income countries with more societal upper-income countries we were left with only 35 countries eligible
inequality may exhibit policy characteristics that may predispose cer- for this study, partially due to the low insurance prevalence in nine
tain segments of the population to higher prevalence of ineffective in- of the LMIC countries included in the WHS. This is, however, a
surance. For example, higher income countries more dependent on common limitation of ecological studies investigating cross national
private health insurance systems may have segments of their popula- income inequality, with the number of countries included ranging
tion exposed to higher deductibles or co-insurance or pre-existing con- between 21 and 24 in numerous studies (De Vogli et al. 2005,
dition exclusions leading to relatively increased prevalence of Wilkinson and Pickett 2007, Biggs et al. 2010). The limited number
ineffective insurance than would be expected for their level of societal of countries is often compensated for by longitudinal analyses, but
wealth. Hence, societies with greater income inequality may be those the lack of time series data from the WHS prohibited us from analy-
where the inter-individual influences of the social determinants of sing trends in ineffective insurance and income inequality over time.
health, including quality of health insurance, are most distinct. The type of data collected in the WHS also limited our case
6 Health Policy and Planning, 2016, Vol. 0, No. 0

definition and, therefore, assessment of ineffective insurance. Thus, Feenstra RC, Inklaar R, Timmer MP. 2015. The next generation of the Penn
estimates of ineffective insurance were intended to reflect conserva- World Table. American Economic Review 105: 3150–82.
tive estimates by limiting our definition to the existence of three po- Feigl AB, Ding EL. 2013. Evidenced Formal Coverage Index and universal
healthcare enactment: A prospective longitudinal study of economic, social,
tential issues (having an untreated chronic condition, selling assets
and political predictors of 194 countries. Health Policy 113: 50–60.
or borrowing money to pay for healthcare expenditures, and deliver-
Filmer D, Pritchett LH. 2001. Estimating wealth effects without expenditure
ing a child outside of a health facility).
data – or tears: An application to educational enrollments in states of India.
We found that income inequality was directly associated with inef- Demography 38: 115–31.
fective insurance, especially among groups that are traditionally more Giedion U, Alfonso DA, Dıaz Y. 2013. The Impact of Universal Coverage
privileged and among wealthier LMICs. We hypothesize that this as- Schemes in the Developing World: A Review of the Existing Evidence.
sociation may be due to the imprint of underlying social inequities as Washington DC: The World Bank.
a population approach decreasing marginal returns of societal wealth Gravelle H. 1998. How much of the relation between population mortality
on improved health insurance. Despite the growth in health insurance and unequal distribution of income is a statistical artefact?. BMJ 316:
coverage from UHC expansion and increasing wealth of LMICs, there 382–5.
Grogger J, Arnold T, Le on AS, Ome A. 2015. Heterogeneity in the effect of
can still be inequities in access and financial protection that are reflect-
public health insurance on catastrophic out-of-pocket health expenditures:
ive of underlying social inequities. As UHC is further expanded under
the case of Mexico. Health Policy and Planning 30: 593–9.
the Sustainable Development Goals, our work emphasizes the need
Hosseinpoor AR, Bergen N, Koller T et al. 2014. Equity-oriented monitoring
for investment in health systems strengthening to monitor and ameli-

Downloaded from http://heapol.oxfordjournals.org/ at University of Newcastle on January 3, 2017

in the context of universal health coverage. PLoS Medicine 11: e1001727.
orate the development of such inequities. Kondo N, Sembajwe G, Kawachi I et al. 2009. Income inequality,
mortality, and self-rated health: meta-analysis of multilevel studies. BMJ
339: b4471.,
Acknowledgements Lochner K, Pamuk E, Makuc D, Kennedy BP, Kawachi I. 2001. State-level in-
The authors thank the two anonymous reviewers for their feedback. come inequality and individual mortality risk: a prospective, multilevel
study. American Journal of Public Health 91: 385–91.
Lynch J, Smith GD, Harper SAM, Hillemeier M. 2004. Is income inequality a
Funding determinant of population health? Part 1. A systematic review. Milbank
Quarterly 82: 5–99.
This work was supported by a grant from Columbia University College of
Ortiz I, Cummins M. 2011. Global Inequality: Beyond the Bottom Billion – A
Physicians and Surgeons to F.N.A.
Rapid Review of Income Distribution in 141 Countries. New York: United
Conflict of interest statement. None declared. Nations Children’s Fund (UNICEF).
Pickett KE, Wilkinson RG. 2015. Income inequality and health: A causal re-
view. Social Science & Medicine 128: 316–26.
References Solt F. 2009. Standardizing the world income inequality database. Social
Science Quarterly 90: 231–42.
Babones SJ. 2008. Income inequality and population health: Correlation and
Subramanian SV, Kawachi I. 2006. Whose health is affected by income in-
causality. Social Science & Medicine 66: 1614–26.
equality? A multilevel interaction analysis of contemporaneous and lagged
Bezruchka S, Namekata T, Sistrom MG. 2008. Improving economic equity
effects of state income inequality on individual self-rated health in the
and health: The case of postwar Japan. American Journal of Public Health
United States. Health & Place 12: 141–56.
98: 589–94.
UN General Assembly, 2015. Draft outcome document of the United Nations
Biggs B, King L, Sanjay B, Stuckler D. 2010. Is wealthier always healthier? The
summit for the adoption of the post-2015 development agenda. Draft reso-
impact of national income level, inequality, and poverty on public health in
lution submitted by the President of the General Assembly. A/69/L.85.
Latin America. Social Science & Medicine 71: 266–73.
http://www.un.org/ga/search/view_doc.asp?symbol¼A/69/L.85&Lang¼E,
Boerma T, Eozenou P, Evans D et al. 2014. Monitoring progress towards univer-
accessed 31 December 2015.
sal health coverage at country and global levels. PLoS Medicine 11: e1001731.,
Wilkinson RG, Pickett KE. 2007. The problems of relative deprivation: why
Chetty R, Stepner M, Abraham S et al. 2016. The association between income
some societies do better than others. Social Science & Medicine 65:
and life expectancy in the United States, 2001-2014. JAMA doi:10.1001/
1965–78.
jama.2016.4226.
Wilkinson RG, Pickett KE. 2008. Income inequality and socioeconomic gradi-
De Vogli R, Mistry R, Gnesotto R, Cornia GA. 2005. Has the relation between in-
ents in mortality. American Journal of Public Health 98: 699–704.
equality and life expectancy disappeared? Evidence from Italy and top industri-
World Bank. World Development Indicators 2003 – Country Income
alised countries. Journal of Epidemiology and Community Health 59: 158–62.
Classification. Washington DC: World Bank, 2003.
Deaton A. 2003. Health, inequality, and economic development. Journal of
World Health Organization. 2010. The World Health Report – Health
Economic Literature 41: 113–58.
Systems Financing: the Path to Universal Coverage. Geneva: World Health
El-Sayed AM, Vail D, Kruk ME. 2016. Ineffective insurance in 42 lower- and
Organization.
middle-income countries. Manuscript in review.
Zheng H. 2012. Do people die from income inequality of a decade ago?. Social
Ellison GTH. 2002. Letting the Gini out of the bottle?: Challenges facing the
Science & Medicine 75: 36–45.
relative income hypothesis. Social Science & Medicine 54: 561–76.