WPS6324
Policy Research Working Paper 6324
The Impact of Health Insurance Schemes
for the Informal Sector
in Low- and Middle-Income Countries
A Systematic Review
Arnab Acharya
Sukumar Vellakkal
Fiona Taylor
Edoardo Masset
Ambika Satija
Margaret Burke
Shah Ebrahim
The World Bank
Development Economics Vice Presidency
Partnerships, Capacity Building Unit
January 2013
Policy Research Working Paper 6324
Abstract
This paper summarizes the literature on the impact indicate welfare improvements and (2) clarity in the
of state subsidized or social health insurance schemes consideration of selection issues. They find the uptake
that have been offered, mostly on a voluntary basis, of insurance schemes, in many cases, to be less than
to the informal sector in low- and middle-income expected. In general, we find no strong evidence of an
countries. A substantial number of papers provide impact on utilization, protection from financial risk,
estimations of average treatment on the treated and health status. However, a few insurance schemes
effect for insured persons. The authors summarize afford significant protection from high levels of out-of-
papers that correct for the problem of self-selection pocket expenditures. In these cases, however, the impact
into insurance and papers that estimate the average on the poor is weaker. More information is needed
intention to treat effect. Summarizing the literature to understand the reasons for low enrollment and to
was difficult because of the lack of (1) uniformity in explain the limited impact of health insurance among the
the use of meaningful definitions of outcomes that insured.
This paper is a product of the Partnerships, Capacity Building Unit, Development Economics Vice Presidency. It is part
of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy
discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org.
The corresponding author may be contacted at Arnab.Acharya@lshtm.ac.uk.
The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.
Produced by the Research Support Team
The Impact of Health Insurance Schemes for the
Informal Sector in Low- and Middle-Income
Countries: A Systematic Review
Arnab Acharya1*, Sukumar Vellakkal2, Fiona Taylor3, Edoardo Masset4, Ambika
Satija2, Margaret Burke5, Shah Ebrahim3
JEL Classifications: I10 and I15
Key terms: Health insurance, health-care utilization, out-of-pocket expenditures
Word Count: Text 5500; References 1200; Tables: 3100
1
Department of Public Health Policy, London School of Hygiene & Tropical Medicine, London,
UK
2
South Asia Network for Chronic Disease, Public Health Foundation of India, New Delhi, India
3
Department of Epidemiology & Population Health, London School of Hygiene & Tropical
Medicine, London, UK
4
Institute of Development Studies, Sussex, UK
5
Department of Social Medicine, University of Bristol, Bristol, UK
Funding: This work was supported by the UK Department for International Development-DFID
[PO 40031461].
*Corresponding Author: Arnab Acharya, Dept. of Global Health and Development, London
School of Hygiene and Tropical Medicine, 15-17 Tavistock Place, London, WC1H 9SH,
Arnab.Acharya@lshtm.ac.uk, (44)20 7927 2681.
i
Acknowledgements: We would like to thank Rachel Blackman, Billy Stewart, Julia Watson, and
Robert Yates from Department for International Development, United Kingdom for their
comments. We would also like to thank Sarah Alkenbrack, Kara Hanson, Jorge Hombrados,
Anne Mills, Chris Millet, Timothy Powell-Jackson, Ajay Tandon, Richard Smith, Hugh
Waddington, Howard White, and Winnie Yip as well as seminar participants at the London
International Development Centre, London School of Hygiene and Tropical Medicine at a
session at the 8th World Congress on Health Economics, Toronto and from the South Asia
Network for Chronic Disease. We greatly appreciate Wendy Wisbaum’s editorial assistance. All
errors remain the sole responsibilities of the authors.
ii
SYS-REVIEW/HI-LMIC
<>Introduction
In a seminal study, Townsend (1994) showed that in rural India, a health crisis in a household
induced significant declines in the level of household consumption that were more severe than
those associated with other financial crises. Townsend examined households’ ability to “smooth
consumption,� to maintain a stable consumption level over a period of time. The inability to
smooth consumption over time because of a health crisis has been found in other developing
countries, partly as a result of the inability to afford appropriate and effective care to ensure
the recovery of health and partly as a result of the reduced labor supply (Gertler and Gruber
2002; Deaton 1997). Recent accounting methods indicate that for as many as 1.3 billion people
in low- and middle-income countries, financial constraints are a major barrier to access to
health care (Xu et al. 2003; Preker et al. 2004).
Among the solutions proposed within various low- and middle-income countries to reduce
costs to households at the point of care are the establishment or extension of national or social
health insurance (SHI) in which service providers are paid from a designated government fund,
which is partially funded through taxes. These SHIs are primarily intended for those employed
outside of the formal sector. The WHO (2010) and the World Bank (Hsiao and Shaw 2007) have
endorsed the restriction of out-of-pocket (OOP) expenditures for health care at the time of use
through the prepayment of insurance as an important step toward averting the financial
hardship associated with paying for health care. The impact of insurance, even when properly
implemented, is not clear from the outset in low- and middle-income countries. Awareness of
1
SYS-REVIEW/HI-LMIC
and trust in public programs, distance to health care facilities and institutional rigidities within
the health care system can play major roles in limiting insurance enrollment and its related
effects (Wagstaff 2007; Basinga et al. 2010).
We conduct a systematic review of the literature on the extent to which insurance schemes
enhance access to care and offer protection from financial risk to households in the informal
sector. We also report on how these schemes may improve health. Previous reviews of the
impact of health insurance in low- and middle-income countries for the poor include studies by
Ekman (2004) and Lagarde and Palmer (2009). Both of these reviews focused on community-
level insurance. The latter examined the impact of removing user fees. Ekman (2004) focused
on community-based health insurance in low-income countries and concluded that community-
based health insurance provides some financial protection by reducing OOP spending. Several
countries have gone beyond community risk sharing, and many new insurance schemes have
been introduced since 2004. In parallel, there is a growing interest in evaluating the impact of
SHI programs in low- and middle-income countries (Wagstaff 2009). Thus, a new review of the
impact of health insurance for people in the informal sector could inform policy on the extent
to which insurance provides greater access to necessary care, reduces the financial burden of
care, and improves health.
The paper is organized as follows. In Section 2, we explain the review methodology of the
current study. Section 3 describes methodologies that should be used to assess the impact of
health insurance when it is offered on a voluntary basis. In Section 4, we describe the schemes
2
SYS-REVIEW/HI-LMIC
found in the literature. Section 5 examines enrollment into health insurance schemes and its
related impact. Section 6 concludes the paper.
<>Methodology
The methodology aims to critically examine the evidence on whether health insurance
schemes, once implemented, (1) are adopted, (2) provide greater access, (3) provide financial
protection, and (4) improve health status among the intended beneficiaries. The last three
points describe welfare impact. We summarize the methodology by defining the types of
insurance that are of interest.
<>Types of Insurance
If universal health care coverage is to be financed through prepayment for insurance, health
insurance should have the following characteristics: (1) compulsory contributions to the risk
pool (otherwise, the rich and healthy will opt out); (2) large numbers of people in the risk pool
because pools with a small number of people cannot spread risk sufficiently and are too small
to handle large health costs; and (3) pooled funds that are generally subsidized from
government revenue (WHO 2010) due to the large numbers of poor people. The standard SHI
schemes in developed countries mandate enrollment for people who are fully employed by
imposing a mixture of tax on the employer and direct payment of insurance premiums. This
effort is accompanied by requiring enrollment from people who are self-employed or
unemployed, with varying degrees of revenue funding and cross-subsidies (OECD 2004).
3
SYS-REVIEW/HI-LMIC
SHI has also been mandated for formal sector workers in a number of developing countries. To
achieve universal health care coverage, an institutional structure that emphasizes payment to
providers for services delivered has been offered to those beyond the formal workforce in
Vietnam, Nigeria, Tanzania, Ghana, India, and China over the last 15 years. Although schemes
for the poor may share the same administrative structure as SHI for the formal sector, the
former usually offer a reduced package or restrictions on providers. Alternatively, we find free-
standing schemes (separate from SHI) that offer financial protection to the poor through
subsidized, usually voluntary household enrollment into a defined benefits arrangement
(Wagstaff 2009). Insurance is offered at premiums that are considerably below the actuarially
fair price. Given that most employment in low- and middle-income countries is informal,
governments manage compulsory insurance in the formal sector with limited avenues to cross-
subsidize the informal sector. Thus, health insurance is tax financed, although funding for it may
be ring-fenced.
Starting from a point of low utilization by the formal sector, both free care and prepayment of
financing for care are implemented to encourage use for illness or to increase contact with
health workers to facilitate better delivery of preventive care. From the perspective of the user,
prepayment insurance schemes, whether highly subsidized or zero-entry fee, should be
understood differently than free care. First, even if the entry fee is small, some households may
not be able to afford the fee. For example, Sparrow, Suryahadi, and Widayanti (2008) report
that in Indonesia, it may be costly to provide photos of household members for insurance
cards. Second, at the point of care, there may be copayments to limit frivolous care-seeking
behavior because of the extremely low marginal cost of seeking care if zero copayments prevail.
4
SYS-REVIEW/HI-LMIC
Reimbursement mechanisms may play a role; many families have little cash, are credit
constrained (Pitt and Khandkar 1998) and cannot take on the financial burden at the point of
care when reimbursement is often delayed (Shi et al. 2010).
From a governmental perspective, insurance allows a separation between the provision and the
funding of care. Despite a public sector that offers care largely free of cost to recipients, the
insurance system can take advantage of the pluralism in the supply of medical care that prevails
in most low- and middle-income countries. The use of the private sector through government
financing of health insurance would reduce the administrative burden within the allocation and
subvention processes, the incidence of side payments and, perhaps, corruption.
When insurance is offered free, it often involves prescribed care and care givers. Some of these
may be offered at the national level and some at the community level. Some community-level
insurance with subsidized entry fees may have limited risk pooling because a specific
community may be small. Most insurance that we examined only offered a set of well-defined
interventions; thus, limited risk pooling at the community level may yield the same coverage as
nationally sponsored insurance in terms of illness. Nationally sponsored insurance may allow
for wider access to providers.
We examine studies of schemes that meet all of the following criteria:
(1) Schemes that seek to offer financial protection for people facing health shocks to cover
health care costs; these schemes involve some tax financing
(2) Schemes that include a component in which poorer households can or must enroll
through some formal mechanism at a rate much lower than the actuarial cost of the
package or even free of charge; in return, these households receive a defined package
of health care benefits
(3) Schemes that are offered in one of two ways:
5
SYS-REVIEW/HI-LMIC
a. Nationally managed and considered an extension of existing SHI
b. Managed at the community level (limiting the risk-pooling population), either
through a local government or nongovernmental organization (usually with
government sponsorship), often called community-based health insurance
<>Defining the Impact of Insurance
We presume that the impact evaluation of a project should provide two essential pieces of
information to policy makers: (1) Is the program implementable? (2) Once implemented, does
the program achieve a set of desirable outcomes? In the case of a health insurance program, if
the adoption of a program is high, then the program at least approached proper
implementation. Furthermore, policy makers are interested in the impact on those adopting
the insurance, the average treatment on the treated, or the average impact on those who
actually adopted the program. Insurance may affect those who do not adopt it by, for example,
affecting the price of health care, which would be part of the total impact of the insurance
program. Morduch (2006)[[This reference is not in the reference list. Please complete the
information at the place holder we have left there]] reports that if richer individuals adopt a
social program disproportionately at zero private cost, the program can be considered a large
income transfer that may affect the prices of all goods within a relevant economy. The impact
on prices is smaller when the poor adopt social goods because the adoption may not replace
previous large expenditures. This type of general equilibrium impact has been largely ignored,
probably because of the stable unit treatment value assumption where only those intended to
6
SYS-REVIEW/HI-LMIC
be affected by the program become the subject of evaluation (see Imbens and Wooldridge
2009), which focuses on the intended target of a program.
The intention to treat effect can also be measured as the impact of insurance on individuals
offered the insurance, regardless of whether it was adopted. An intention to treat measure
would not approximate the impact of health insurance for those who adopt insurance if
insurance uptake is not at a very high level, a common situation in most of the studies.
We report on three outcomes: (1) utilization of health care, (2) financial protection, and (3)
health status. Willingness to pay for health insurance schemes, obtained ex post once the
benefits have been realized, can be used to measure welfare impact. Ex ante willingness to pay
for insurance is likely to be positive (Gustafsson-Wright, Asfaw and van der Gaag 2009).
However, recipients of health services may not always be able to accurately assess benefits.
Furthermore, the severity of income constraints for the poor may not elicit well-considered
responses.
<>Inclusion Criteria
The studies on which we report must measure or report impact through a comparator, using
either a contemporaneous control or a constructed control from data containing similar
information collected over a similar time period. Inclusion criteria are as follows:
(1) Randomized controlled trials
(2) Quasi-randomized controlled trials in which methods of allocating are not random but
create a matched control group through either
a. a propensity score matching method or
b. a regression discontinuity design
7
SYS-REVIEW/HI-LMIC
(3) Controlled before-and-after studies or difference-in-differences; the pre- and
postintervention periods for study and control groups should be the same, and the
choice of the control site should be similar in terms of socioeconomic characteristics
and/or should have no major differences in the baseline
(4) Regression studies that consider the probability of selection into treatment through the
instrumental variable method
(5) Qualitative studies focused on exploring the impact of health insurance and meeting a
checklist
However, no qualitative studies that explored the impact of insurance were found.
<>Search Method
A number of electronic databases 1 were searched using keywords related to health insurance,
health care, and low- and middle-income countries. This search yielded 4756 references,
including numerous duplicates and studies detailing general health issues in low- and middle-
income countries. We filtered by titles and abstracts to reduce the number of relevant studies
to 64. Of these, 35 were related to the impact of SHI on low- and middle-income countries.
Inclusion criteria were met in 24 studies. Further examination found that five studies used poor
identification strategies when measured against the standard methods recommended for
impact evaluation studies (Imbens and Wooldridge 2009).
<>Summarizing the Results
It is difficult—and, more important, misleading—to aggregate the outcome measures that we
found into some form of meta-analysis. This difficulty arises for three reasons: (1) many of the
1
Databases from the 1950s to September 2010 were searched: the Cochrane EPOC group Specialised Register and
Library (3. 2010), MEDLINE, EMBASE, ECONLIT, ISI Web of Knowledge, CAB Abstracts, CENTRAL, DARE, ELDIS, and
IDEAS as well as websites from the World Bank, the World Health Organization, and the U.S. National Bureau of
Economic Research. Expert opinions and searches in key journals yielded findings of additional studies published
before July 2011.
8
SYS-REVIEW/HI-LMIC
outcome measures are different (for example, the time intervals varied); (2) the insurance
schemes were different, as outlined in table 2; and (3) the estimations of the impact depended
on the functional form or the estimation method and the unit of measure, such as the period in
which data were measured.
The unit of measure can shape the results, especially with regard to health expenditures (Das,
Hammer, Sánchez-Paramo 2011). It can also dictate different estimation methods. For example,
OOP expenditures as a share of income can be modeled through probit, whereas OOP
expenditures may be modeled using linear estimation methods. In addition, when magnitudes
are reported, they should be understood within the context of the study; magnitudes have
limited generalisability outside of a study, even for the same insurance scheme within the same
region. Thus, only trends are reported.
<>Identification Issues
Although low enrollment fees should have attracted universal adoption, in most cases,
enrollment rates were low. Low enrollment may induce selection effects, and selection into
insurance may ultimately affect the outcome. One way that selection may affect outcome is
through adverse selection: ill individuals select themselves into insurance at premium levels,
which individuals in good health find the premium too costly given their expectation of their
health care needs (Rothschild and Stiglitz 1976). 2 The pool of the insured may be sicker than
the pool of the uninsured. The expectation of becoming healthy influences the adoption of
2
Adverse selection is one of the reasons mandatory insurance is prescribed in many instances. Voluntary
enrollment increases the possibility of adverse selection, which is one of the reasons that enrollment fees must be
low and the cost of the program must be subsidized.
9
SYS-REVIEW/HI-LMIC
insurance, which can be an efficient way to obtain care. In contrast to the possibility of adverse
selection, given the low costs of entry for most insurance, it may simply be that better-
informed individuals enroll. Better-informed individuals may also be more educated, may have
larger incomes, and may be healthier than those who do not adopt insurance.
Thus, if one examines the average impact of insurance on those who adopt insurance (i.e.,
average treatment on the treated), then a simple comparison of insured and noninsured
individuals does not yield appropriate results. The comparison is flawed because the
noninsured group may not have had an opportunity to enroll, and this group includes those
who would adopt insurance if offered as well as those who would not adopt insurance. The
insured and noninsured groups may differ with regard to the factors that may affect outcomes.
In light of the possibility of selection into insurance, the threat to validity is high. As a result, the
impact may differ significantly when no adjustment is made for this type of selection, especially
if average treatment on the treated is reported (Imbens and Wooldridge 2009). 3 Inclusion
criteria have focused on studies with counterfactuals, we now discuss identification issues
applied to the inclusion criteria. We briefly describe statistical procedures to obtain average
treatment on the treated in the present context. As indicated, 19 studies properly addressed
identification issues.
<>Randomized Studies
Even if insurance is offered through random means when uptake is low, it cannot be assumed
that people who adopt insurance are similar to those who do not. An adjustment is needed
3
Yip and Berman’s (2001) study was among the early empirical papers on health insurance for developing countries that
recognized the selection problem. They addressed the issue through simulation.
10
SYS-REVIEW/HI-LMIC
even in this randomized setting. The most standard approach is to determine the local average
treatment effect. If the stable unit treatment value assumption holds, as the instrument, mainly
the assignment to treatment, is exogenous, then the local average treatment effect estimates
the impact of those who comply with the offer but would not be treated otherwise (Angrist and
Pischke, 2009). Thornton used the instrumental variable approach, although different
approaches have also been used.4
<>Matching
If insurance uptake occurs in a nonexperimental setting, a popular method known as propensity
score matching can be used. Impact is measured by comparing the outcomes of insured
individuals with the outcomes of nonparticipants. This measurement derives weights for the
outcomes for nonparticipants according to the degree of similarities between the two groups as
judged through observed factors, which are reduced to a single metric (Rosenbaum and Rubin
1983): the propensity to be enrolled in a program. Wagstaff et al. (2009) emphasize that
unobservable heterogeneity may be stronger between those who adopt insurance and those
who refuse insurance when both groups are offered insurance than between those who have
never been offered insurance and those who adopt insurance when offered. In this case, the
matching method should use comparators chosen from those who have never been offered
insurance.
<>Instrumental Variable Approach
4
Studies with random allocation at the cluster level using informal matching methods cannot be justified; see Devadasan et al.
(2010).
11
SYS-REVIEW/HI-LMIC
A number of authors have used instrumental variable methods to determine that individuals
who adopt insurance are not easily comparable to those without insurance. Insurance status is
dependent on a variable that affects only entry into insurance, not any of the outcomes that
may be affected by insurance. For studies that use the instrumental variable method,
participating in insurance can be considered a problem of endogeneity or of selecting into
insurance. Wagstaff and Lindelow (2008) and Sosa-Rubi, Galarraga and Harris (2009a) model
selection into insurance as a problem of endogeneity; individuals anticipate the impact of
insurance, and this expectation of the impact shapes the uptake of insurance. These studies use
instrumental variable methods to correct endogeneity. Instrumental variable methods can
correct policy endogeneity (Dow and Schmeer 2003) by including policy with the expectation of
certain types of results.
<>Regression Discontinuity Design
When programs are targeted to a group at a measurable threshold income, the regression
discontinuity design approach compares health care-related outcomes for those who are
eligible at the margin with those who are just above eligibility. Individuals who do not qualify
for enrollment in insurance because they are marginally on the other side of eligibility
constitute the control group. The impact of regression discontinuity design yields an intention
to treat estimation; some individuals who are eligible may not actually have insurance although
they intended to receive it.
12
SYS-REVIEW/HI-LMIC
Finally, a study by Wagstaff (2010) subtracts two previous difference-in-differences outcome
measures from two later difference-in-differences measures (using available data for three
periods) and regresses this variable with similar differences in the independent variables and
insurance status.
Several studies model the insurance effect through multiple observations of individuals in the
sample and individual heterogeneity over time. This model is usually performed by inputting
factors for a specific individual effect with the underlying assumption that any correlation
between the error term and the insurance status arises from the correlation between time-
variant unobservable factors (perhaps such as health) and insurance status. However, the time-
invariance assumption is unlikely to hold because health conditions indeed fluctuate to
influence insurance uptake. 5
<>Description of the Studies
Table 1 provides descriptions of the health insurance schemes from the 19 studies in addition
to the corresponding data and methodologies. No study attempted to link the various
outcomes of interest to any specific insurance features. However, it is instructive to note which
types of schemes were evaluated. Reports of the impact of health insurance are from Burkina
Faso (one study) Costa Rica (one study), Georgia (one study), Ghana (one study), India (one
5
Two of the reviewed studies attempted this approach. Sepehri, Sarma and Simposon (2006) attempt to control selection
through the use of fixed-effect or random-effect models for individuals in a panel. Similarly, Sparrow, Suryahadi, and Widyanti
(2010) model insurance impact with baseline self-reported health status, which is nearly akin to a fixed-effect model. They
report that self-reported illness is likely to be unreflective of actual illness status.
13
SYS-REVIEW/HI-LMIC
study) Nicaragua (one study), Colombia (two studies), Mexico (three studies), Vietnam (four
studies), and China (four studies). Three studies, from Burkina Faso, China, and India, reported
on community-based health insurance with government support.
Not all studies reported enrollment. Studies on impact evaluations obtained results through (1)
a randomized trial (three studies); (2) propensity score matching (nine studies); (3) instrumental
variable estimation, to consider either endogeneity at the individual level or regional program
placement (four studies); (4) the use of a regression discontinuity design on eligibility to obtain
intention to treat (two studies); and (5) double difference-in-differences from three periods
with regression (one study).
The data used in these studies ranged from program-designated data sets to routinely collected
available data at the national level gathered to measure a range of indicators of wellbeing.
<>
<>Findings
We first report enrollment, and then, we report intention to treat or average treatment on the
treated estimations of whether insurance is likely to have resulted in welfare improvement.
Table 2 summarizes the outcomes.
<>Enrollment and Its Determinants
The enrollment rate partially reflects whether a health insurance program can be implemented.
Our review did not conduct a systematic search to identify studies that report enrollment. For
14
SYS-REVIEW/HI-LMIC
three papers (Gnawali et al. 2009; King et al. 2009; Thornton and Field 2010), the evaluation
was conducted for programs that were designed to enhance enrollment. The activities did not
enhance enrollment.
Enrollment rates varied. For the Vietnam Health Care Fund for the Poor (VHCFP), introduced in
2003, which includes free enrollment and no copayment with specified access to care, country-
wide enrollment reached 60 percent by 2006 (Wagstaff 2010). The New Cooperative Medical
Scheme (NCMS) in China showed regional variations of 48 to 99 percent (Wagstaff 2009). By
2007 (i.e., within the first four years), national enrollment in Ghana was at 55 percent (Mensah,
Oppong and Schmidt 2010). Bauhoff, Hotchkiss and Smith (2010) report low enrollment in
Georgia in a collected sample.
The enrollment patterns and determinants of enrollment in health insurance schemes are
similar to those observed for enrollment into insurance schemes to provide protection from
adverse shocks in general (Gine and Yang 2007). We summarize the factors affecting
enrollment from studies that reported determinants: 6
(1) No clear demographic patterns emerge; in some cases, positive enrollment factors
include female-headed households and elderly headed households, family size, and
composition
(2) Positive effect of education (except in Colombia; Miller, Pinto and Vera-Hernandez,
2009)
(3) No influence of initial conditions, such as chronic illness (except in Colombia; Miller et al.
2009)
(4) No influence of distance to health centers or rural residency (except in Mexico, where
people in rural areas sign up more frequently; Sosa-Rubi et al. 2009a [[Please include
“a� or “b� after the year, and replace “et al.� with the second and third authors’ last
names, depending on which reference you mean to cite here.)
6
Reporting the determinants of insurance involved straightforward identification issues. We include enrollment results from a
few studies for which we did not include impact results: Sun et al. (2009), Dror et al. (2009), Schneider and Diop (2001), and
Msuya (2004). No studies had high enrollment at the time reported by the study.
15
SYS-REVIEW/HI-LMIC
The initial health condition did not matter. A detailed study by Gine, Townsend and Vickery
(2008) on the uptake of agricultural insurance against bad weather showed that adverse
selection played a small role in uptake. None of the studies we examined explicitly included a
variable for trust in government or financial institutions, levels of risk aversion, availability of
care, or understanding of insurance. Not all studies reported enrollment, even though the issue
of who enroll influences the outcome of a social program.
<>Utilization
The studies report whether the use of overall or specific types of health care was higher for
insured people than for uninsured people within a specific time interval. Studies conducted
across multiple time periods compare changes across two groups. To measure any incidence of
utilization, studies use logit or probit; to measure the impact of insurance on the number of
incidences per household or person, count data models can be used. Most studies reported on
both inpatient and outpatient care. Choice of facility, which has cost implications, was also
reported through a multinomial model.
Membership in health insurance schemes may lead to overuse of health care as a result of two
types of moral hazards: overuse because the cost of any given point of contact with the health
care system for the insured is low or nearly zero and overuse because insurance involves a
third-party payer, which can encourage greater health care utilization. Thus, the utilization rate
may not reflect actual welfare gains. There was no estimation of unnecessary care in any of the
studies. Where there was a financial barrier to care, increased care gained through insurance is
likely to indicate unambiguous welfare improvement.
16
SYS-REVIEW/HI-LMIC
In the case of Ghana, Mensah, Oppong and Schmidt (2010) report a higher utilization rate for
pregnancy care among the insured, although the sample is small. For Nicaragua (Thornton and
Field 2010) and Georgia (Bauhoff et al. 2010), insurance targeted mostly to the poor did not
induce higher utilization, although the study in Georgia reported higher utilization by those with
higher assets. In Burkina Faso (Gnawali et al. 2009) and India (Aggarwal 2010), the two
community-managed schemes, there were overall increases in health care use, but there was
no impact on inpatient utilization.
Trujillo, Portillo and Vernon (2005) and Miller, Pinto and Vera-Hernandez (2009) indicate
positive effects for the same insurance program in Colombia at different times. The insured
received care more often, and the latter study reported a higher use of preventive care after
changes to the payment structure for the provider. Both studies report no difference in
inpatient care for insured and noninsured groups.
For Mexico’s Seguro Popular (SP), studies report differing results: King et al. (2009) report no
higher utilization for the insured for all health care, whereas Sosa-Rubi, Galarraga and Lopez-
Riduaura (2009) report that diabetics insured under SP have better access to diabetic care
compared to the corresponding figures for diabetics who are uninsured.
Three studies of VHCFP from Vietnam report conflicting results. Wagstaff (2007) reported
higher utilization rates for inpatient and outpatient care, with substantially higher inpatient
care. Axelson et al. (2009) reported a small increase in overall utilization, mostly because of
increased outpatient care. Although both papers use propensity score matching, they use
different data. A subsequent study by Wagstaff (2010), which used a different data and
17
SYS-REVIEW/HI-LMIC
methodology, found no effect of insurance on utilization. The results from both papers by
Wagstaff are not robust to functional specifications. For insurance prior to VHCFP, Jowett,
Deolalikar and Mattinsson (2004) use instrumental variables on the decision to seek care and
the type of health center used. They report that the insured are more likely to use health
services and public services than the uninsured.
Contradictory results emerge from two studies on China´s NCMS. Wagstaff et al. (2009) show
that in China, the insured, including the insured poor, use health services more often in
comparison to the noninsured. Lei and Lin (2009) show no overall effect for utilization but find a
drop in the use of traditional care and an increase in preventive care.
We cannot claim that insurance yields a higher probability of care seeking. It is particularly
telling that different results can be obtained for the same insurance. Of the 15 studies reporting
utilization, nine studies report a higher utilization rate among the insured. Recall that increased
usage may not always indicate welfare improvement.
<>
<>Financial Protection
Insurance should protect the insured from incurring high levels of health care costs. An
effective health care system that includes insurance and other forms of social protection should
provide much broader financial risk protection. None of the insurance schemes offered
protection for financial loss due to reduced labor supply, which is among the main reasons for
18
SYS-REVIEW/HI-LMIC
the lack of consumption smoothing, as noted by Gertler and Gruber (2002). Miller et al. (2009)
mention, in passing, the important issue of whether health insurance can go beyond reducing
health care costs to eliminate significant adverse effects of health shocks so that households
can maintain their standard consumption and saving bundles (Townsend 1994; Chetty and
Looney 2006). Most studies addressed only the issue of OOP expenditures and do not include
insurance premiums or entry fees into insurance. Some studies used the measure of
catastrophic payment, defined as a threshold proportion of all expenditures (or some type of
income measure, which is usually imprecise), that is spent on health. The denominator varied
across studies, as did the threshold levels. Because a reduction in the average level of OOP
expenditures for a household would reflect a reduction in high-level OOP expenditures, many
studies reported on this value. One way of describing the impact of insurance on financial risk
protection is to examine the right tail of the distribution of OOP expenditures. Distributional
analysis may require the use of quantile regression methods that help to analyze the
occurrence of high levels of expenditures at different income levels.
One indicator of improved wellbeing is found by measuring the reprieve from high levels of
OOP expenditures by the poor in comparison to populations without insurance. This indicator
was not clearly identified in any of the studies. Low levels of increased spending may actually
indicate greater contact with health services, which may occur through insurance. Comparisons
between insured and noninsured groups at the average level of OOP expenditures may not
yield a clear measure of welfare.
19
SYS-REVIEW/HI-LMIC
Nonetheless, the studies compared the average expenditure between the insured and the
noninsured at the household level as well as the incidence or probability of incurring high or
catastrophic expenditures, measured at different thresholds. Some studies on the determinants
of expenditures for hospital care noted the large fraction of zeroes because many people do
not use hospital care. Although a two-part model can be used by first considering the likelihood
of the use of health services, this model was not incorporated in most of the studies reported
here. It was difficult to discern whether some studies reported costs only for those who
adopted health care, which may be an observed indicator for being ill. One disadvantage of
such an approach is that among the poor, some people may not use health care at all, even
when they are ill.
For Georgia, Bauhoff et al. (2010) report lower levels of OOP expenditures for the insured, with
a larger impact for inpatient care. Thornton and Field (2010) use baseline data to show that
insurance does not provide cost savings in Nicaragua when the cost of insurance is taken into
account. Aggarwal’s (2010) study of community insurance in India shows a favorable impact for
overall care among the insured but found no effect for inpatient care. Miller et al. (2009) find
overall lower OOP expenditures and a lower incidence of high-level expenditures for the
insured.
Sosa-Rubi et al. (2009a) report that pregnant women with Mexican SP insurance use SP-
sponsored state services, the care with the lowest OOP expenditure. Nevertheless, the
evidence is unclear because there is a preference among the insured for expensive private care
over cheaper types of state-sponsored care. The urban poor seem to have benefitted the most
20
SYS-REVIEW/HI-LMIC
from SP. King et al. (2009) confirm that for all types of care, OOP expenditures are lower for the
insured under SP insurance.
For Vietnam´s VHCFP, Wagstaff (2007) shows no overall impact on OOP expenditures for the
insured. However, there is some protection for high levels of expenditures, with the poor
experiencing a small effect. The results are susceptible to the matching methods used. Axelson
(2009) uses data from two periods and finds a protective effect of insurance; however, there is
no impact when a single cross-sectional data period is examined. Wagstaff (2010) uses data
from three periods and finds strong and robust measures of a greater decline in OOP
expenditures for the insured.
Lei and Lin (2009) do not find a significantly lower level of OOP expenditures for people insured
under China’s NCMS. Wagstaff et al. (2009) note weak evidence for lower OOP expenditures for
the insured under the NCMS; however, this evidence is sensitive to matching methods. For
deliveries, the insured received protection, although this protection was weaker for the poorer
population. Wagstaff and Lindelow (2008) use a number of econometric specifications through
instrumental variable analyses and report that people insured in a (now discontinued) Chinese
health insurance scheme actually experienced higher levels of catastrophic payments,
measured at various threshold levels.
Only four of 16 studies reporting on costs provided conclusive indications of lower average OOP
expenditures for the insured. Seven studies provided mixed results, and two showed no effect.
Five studies reported a lower incidence of catastrophic OOP expenditures.
21
SYS-REVIEW/HI-LMIC
<>Health Status
Surprisingly, only six studies reported on health measures. It is presumed that health insurance
would induce greater access for the insured and thereby lead to better health. With the
exception of the study on the health insurance scheme from Colombia (Miller et al. 2009), no
study reported that supply-side improvement accompanied the introduction of insurance. If
health insurance implementation is not accompanied by a significant improvement in the
quality of supply and does not lead to greater utilization, then we should not expect health
improvement. Financial protection is the main aim of insurance. However, if a range of health
outcomes improves or death rates decline for the insured, then it is possible that we can
attribute better health outcomes to health insurance.
Mensah et al. (2010) show lower levels of infant death, although these levels are not
statistically significant. Wang et al. (2009) use EQ-5D, a standardized index value instrument for
use as a measure of a wide range of health conditions, to report on a community-based health
insurance program in China. They find that the scheme had positive effects on health status for
all insured people and for the poor. Measuring regional changes, Dow and Schmeer (2003) find
no correlation in changes in infant mortality as regional insurance uptake improves. For
Nicaragua, Thornton and Field (2010) show no improvement in health. Sosa-Rubi et al. (2009b)
examine Mexico´s SP insurance and show improved glucose control among diabetics with
insurance than those without insurance. Lei and Lin (2009) find no improvement in health
status for the China’s NCMS.
<>Discussion and Conclusion
22
SYS-REVIEW/HI-LMIC
We now summarize our conclusions and note some methodological issues. We offer very little
in terms of broad results regarding the impact of insurance, once implemented, on the
intended beneficiaries.
Studies reporting on enrollment showed that low enrollment is commonly observed for many
of the insurance schemes; enrollment seems to be related to perceptions, education, and
cultural factors rather than to factors related to health and health care, such as initial health
status and distance to health centers. The study from Nicaragua indicated that there was
considerable confusion about coverage. We do not observe a pattern regarding enrollment and
outcome; for example, China and Vietnam had high enrollment. Nevertheless, there is no
indication that insurance worked well for the participants, although more recent analysis shows
positive results from Vietnam. Given the low coverage, policies could include incentives for
insurance or could even mandate required enrollment. Enrollment may also be low because the
administrative implementation process may be poor.
It is perhaps most important to prevent high levels of OOP expenditures through insurance.
There is some evidence that this may be the case. Some of the studies that report only average
expenditures could not capture this effect. The present method of setting catastrophic
expenditures at various levels of income is arbitrary and complicates comparisons among
studies. One option is to examine the expenditure distributions of the insured and the
noninsured, particularly at high levels of expenditures. However, this technique is of limited use
if there are high levels of selection into insurance. The possibility of quantile regression
methods can be explored (Angrist and Pischke 2009).
23
SYS-REVIEW/HI-LMIC
Counterintuitively, for most of the health insurance schemes, the poorest among the insured
fared less well. One reason that average expenditures may not be lower for the insured poor
than for the uninsured poor is that the latter may not seek any care or may give up on care
altogether when the appropriate care is well beyond reach without insurance. However, if this
were the predominant reason why the poor do not receive the full effect of insurance, then
insurance should induce higher levels of health-seeking behavior. We do not find this to be the
case. One reason for the low level of health-seeking behavior may be a lack of understanding of
insurance or the existence of hidden charges other than those covered by insurance.
Two studies explicitly note that, although a causal link was not established, features of the
payment scheme may have affected the outcome. In the older insurance scheme in China, fee-
for-service may have given rise to cost (Wagstaff and Lindelow 2008). In the case of Colombia,
the incentive structures in the providers’ contract may have given rise to higher use of
preventive care (Miller et al. 2009). Studies should identify health system and household
economic factors that may determine impact. For example, mechanisms for copayments,
expectations from reimbursement policies, and the presence of various financial mechanisms
have been shown to produce variations in uptake, utilization, and health improvement in the
U.S. market (Newhouse and the Insurance Experiment Group 1993; Deb, Trivedi and Zimmer
2006). In the low- and middle-income country study settings, the full range of variations may
not exist as it does in the U.S. market. Qualitative studies may shed some light in this regard.
Health insurance schemes differed sufficiently from one another in this review to avoid
implying any relationship between the specificities of health insurance schemes and outcomes.
24
SYS-REVIEW/HI-LMIC
In this review, we found that many studies used data collected for purposes other than the
evaluation of insurance schemes. Thus, important questions may be missing from general living
standard surveys to allow the assessment of the welfare implications of insurance, such as
detailed questions on illnesses. Longitudinal surveys would be more robust in capturing
selection effects and the extent to which health insurance schemes provide risk protection
against health shocks.
Both the development of rigorous impact evaluation methodology for social programs (Imbens
and Wooldridge 2009) and the introduction of health insurance schemes for the poor in low-
and middle-income countries are new phenomena. Our report of 19 studies may be
encouraging. However, for impact studies to be useful for future considerations of health
insurance schemes, greater attention must be given to the rigor and uniformity of welfare
measurements, especially in terms of risk protection and evaluation methodologies. In the
future, examining a larger number of studies would allow for meta-analyses (regressions),
which would facilitate more conclusive remarks regarding program features and outcomes.
Further, given that health insurance schemes can differ and exist in different contexts, studies
should emphasize pathways through which programs affect outcomes. Although we did not
identify such studies, both qualitative and quantitative methods can be used to trace these
pathways. Policy makers would benefit from a greater number of rigorous studies that examine
the pathways through which programs are likely to affect welfare.
25
SYS-REVIEW/HI-LMIC
References
Aggarwal, A. 2010. “Impact evaluation of India’s ‘Yeshasvini’ community based health insurance
programme.� Health Economics 19 (S1): 5-35.
Angrist, J., A. Imbens, and D.B. Rubin. 1996. “Identification of causal effects using instrumental
variable.� Journal of American Statistical Association 94 (436): 444-72.
Angrist, J., and J-S. Pischke. 2009. Mostly Harmless Econometrics: An Empiricist’s Companion.
Princeton, NJ: Princeton University Press.
Axelson, H., S. Bales, P. Minh, B. Ekman, and U. Gerdtham. 2009. “Health financing for the poor
produces promising short-term effects on utilization and out-of-pocket expenditure:
evidence from Vietnam.� International Journal for Equity in Health 8 (1): 20.
Basinga, P., P. Gertler, A.L.B. Soucat, J.R. Sturdy, and C.M.J. Vermeersch. 2010. Paying primary
health care centers for performance in Rwanda. The World Bank: WPS5190.
Bauhoff, S., D.R. Hotchkiss, and O. Smith. 2011. “The impact of medical insurance for the poor
in Georgia: A regression discontinuity approach.� Health Economics doi: 10.1002/hec.1673.
Chetty R, and A. Looney. 2006. “Consumption smoothing and the welfare consequences of
social insurance in developing economies.� Journal of Public Economics 90 (12): 2351-6.
Das J., J. Hammer, and C. Sánchez-Paramo. 2011. The Impact of Recall Periods on Reported
Morbidity and Health Seeking Behavior. The World Bank: WPS5778.
Deaton, A. 1997. The Analysis of Household Surveys. Baltimore. Johns Hopkins University Press.
Deb, P., C. Li, P. Trivedi, and D.M. Zimmer. “The effect of managed care on use of health care
services: Results from two contemporaneous household surveys.� Health Economics, 15:
743-60.
Devadasan, N., B. Criel, V. Van Damme, S. Manoharan, P.S. Sarma, and P. Van der Stuyft. 2010.
“Community health insurance in Gudallur, India, increases access to hospital care.� Health
Policy and Planning 25 (2):145-54.
Dow, W., and K. Schmeer. 2003. “Health insurance and child mortality in Costa Rica.� Social
Science and Medicine 94 (2): 975-8.
26
SYS-REVIEW/HI-LMIC
Dror, D.M., R. Radermacher, S.B. Khadilkar, P. Schout, F.X. Hay, A. Singh, and R. Koren. 2009.
“Micro insurance: innovations in low-cost health insurance.� Health Affairs (Millwood) 28
(6): 1788-98.
Ekman, B. 2004. “Community based health insurance in low-income countries: a systematic
review of the evidence.� Health Policy and Planning 19 (5): 249-70.
Gertler, P., J. Gruber. 2002. “Insuring Consumption against Illness.� American Economic Review
92 (1): 51-70.
Gine, X., and D. Yang. 2007. Insurance, Credit, and Technology Adoption: Field Experimental
Evidence from Malawi. The World Bank: Policy Research Working Paper 4425.
Gine, X., R. Townsend, and J. Vickery. 2008. “Patterns of Rainfall Insurance Participation in Rural
India.� The World Bank Economic Review: 1-28.
Gnawali, D.P., S. Pokhrel, A. Sié, M. Sanon, M. De Allegri, A. Souares, H. Dong, and R. Sauerborn.
2009. “The effect of community-based health insurance on the utilization of modern health
care services: Evidence from Burkina Faso.� Health Policy 90 (2-3): 214-22.
Gustafsson-Wright, E., A. Asfaw, and J. van der Gaag. 2009. “Willingness to pay for health
insurance: An analysis of the potential market for new low-cost health insurance products
in Namibia.� Social Science and Medicine 69: 1351-9.
Hsiao, W., and R.P. Shaw. 2007. Social Health Insurance for Developing Nations. Washington DC:
The World Bank
Imbens, G., J. Wooldridge. 2009. “Recent development in the econometrics of program
evaluation.� Journal of Economic Literature 47 (1): 5-86.
Jowett, M., A. Deolalikar, P. Mattinsson. 2004. “Health insurance and treatment seeking
behaviour: evidence from a low-income country.� Health Economics 13 (9): 845-57.
King, G., E. Gakidou, K. Imai, K. Lakin, T. Moore, C. Nall, N. Ravishankar, M. Vargas, M. Téllez-
Rojo, J. Hernández �vila, M. Hernández �vila, and H. Hernández Llamas. 2009. “Public
policy for the poor? A randomised assessment of the Mexican universal health insurance
programme.� The Lancet 373 (9673): 1447-54.
Lagarde, M., and N. Palmer. 2009. “The impact of health financing strategies on access in low
and middle income countries.� The Cochrane Collaboration. Protocol Issue 4, Wiley
Publishers www.thecochranelibrary.com
27
SYS-REVIEW/HI-LMIC
Lei, X., and W. Lin. 2009. “The New Cooperative Medical Scheme in rural China: does more
coverage mean more service and better health?� Health Economics 18 (S2): S25-S46.
Mensah, J., J. Oppong, C. Schmidt. 2010. “An evaluation of the Ghana national health insurance
scheme in the context of the health MDGs.� Health Economics 19 (S1): 95-106.
Miller, G., D.M. Pinto, and M. Vera-Hernandez. 2009. High-powered incentives in developing
country health insurance: evidence from Colombia’s regimen sobsidiado: NBER Working
Paper, 15456.
Morduch. 2006. “Micro-insurance: The Next Revolution?� In Abhijit Banerjee, Roland Benabou,
and Dilip Mookherjee, eds., Understanding Poverty. New York: Oxford University Press..
Msuya, J., J.P. Jutting and A. Esfaw, 2004. “Impacts of community health insurance schemes on
health care provision in rural Tanzania.� Bonn: ZEF-Discussion Papers on Development
Policy (82).
Newhouse, J., and the Insurance Experiment Group. 1993. Free for All? Lessons from the RAND
Health Insurance Experiment, Cambridge, MA: Harvard University Press.
OECD. 2004. OECD Study on Private Health Insurance. OECD Health Project. Organisation for
Economic Cooperation and Development.
Pitt, M.M., and S.R. Khandker. 1998. “The impact of group-based credit programs on poor
households in Bangladesh: Does the gender of participants matter?� Journal of Political
Economy 106 (5): 958-96.
Preker, A., G. Carrin, D. Dror, M. Jakab, W. Hsiao, and D. Arhin. 2004. “Rich-Poor Differences in
Health Care Financing.� In Health Financing for Poor Pople: Resource Mobilisation and Risk-
Sharing, pp. 3-51, eds. A.S. Preker, and G. Carrin. Washington DC: The World Bank.
Rosenbaum, P., and D. Rubin. 1983. “The central role of the propensity score in observational
studies for causal Effects.� Biometrika 70 (1): 41-55.
Rothschild, M., and J. Stiglitz. 1976. “Equilibrium in Competitive Insurance Markets: An Essay on
the Economics of Imperfect Information.� Quarterly Journal of Economics 90 (4): 630–49.
Schneider, P., and F. Diop. 2001. Synopsis of results on the impact of community-based health
Insurance on financial accessibility to health care in Rwanda. HNP Discussion Paper.The
World Bank.
28
SYS-REVIEW/HI-LMIC
Sepehri, A., S. Sarma, and W. Simposon. 2006. “Does non-profit health insurance reduce
financial burden? Evidence from the Vietnam living standards survey panel.� Health
Economics 15 (6): 603-616.
Shi, W., V. Chongsuvivatwong, A. Geater, J. Zhang, H. Zhang, and D. Brombal. 2010. “The
influence of the rural health security schemes on health utilization and household
impoverishment in rural China: Data from a household survey of China.� Int J Equity Health
9: 7: doi:10.1186/1475-9276-9-7.
Sosa-Rubi, S.G., O. Galarraga, and J. Harris. 2009a. Heterogeneous impact of the ‘Seguro
Popular’ program on the utilization of obstetrical services in Mexico, 2001-2006: A
multinomial probit model with a discrete endogenous variable. MA: National Bureau of
Economic Research Cambridge.
Sosa-Rubi, S.G., O. Galarraga, and R. Lopez-Riduaura. 2009b. “Diabetes treatment and control:
the effect of public health insurance for the poor in Mexico.� Bulletin of World Health
Organisation 87 (7): 512-9.
Sparrow, R., A. Suryahadi, and W. Widayanti. 2008. Public Health Insurance for the Poor:
Targeting and Impact of Indonesia’s Asheskin Programme. Mimeo. Institute of Social
Studies.
Sun, Q., X. Liu, Q. Meng, S. Tang, B. Yu, and R. Tolhurst. 2009. “Evaluating the financial
protection of patients with chronic disease by health insurance in rural China.�
International Journal of Equity in Health 9: 42.
Thornton, R., and E. Field. 2010. “Social Security Health Insurance for the Informal Sector in
Nicaragua: A Randomised Evaluation.� Health Economics 19 (S1): 181-206.
Townsend R. 1994. “Risk and Insurance in Village India.� Econometrica 62 (3): 539-91.
Trujillo, A., J. Portillo, and J. Vernon. 2005. “The impact of subsidized health insurance for the
poor: Evaluating the Colombian experience using propensity score matching.� Journal of
Health Care Finance and Economics 5 (3): 211-39.
Wagstaff, A. 2009. “Social health insurance re-examined.� Health Economics 19 (5): 503-17.
Wagstaff, A. 2007. Health insurance for the poor: initial impacts of Vietnam‘s health care fund
for the poor. The World Bank, Policy Research Working Paper Series: 4134 2007.
Wagstaff, A. 2010. “Estimating health insurance under unobserved heterogeneity: The case of
Vietnams Health Care Fund for the poor.� Health Economics 19 (2): 189-208.
29
SYS-REVIEW/HI-LMIC
Wagstaff, A., and M. Lindelow. 2008. “Can insurance increase financial risk? The curious case of
health insurance in China.� Journal of Health Economics 27 (4): 990-1005.
Wagstaff, A, M. Lindelow, G. Jun, X. Ling, and Q. Juncheng. 2009. “Extending health insurance to
the rural population: an impact evaluation of China's new cooperative medical scheme.�
Journal of Health Economics 28 (1): 1-19.
Wang, H., W. Yip, L. Zhang, and W.C. Hsiao. 2009. “The impact of rural mutual health care on
health status: evaluation of a social experiment in rural China.� Health Economics 18: S65–
S82. doi: 10.1002/hec.1465.
WHO. 2010. The world health report - Health systems financing: the path to universal coverage.
Geneva: World Bank.
Xu, K., D. Evans, K. Kawabata, R. Zeramdini, J. Klavus, and C. Murray. 2003. “Household
catastrophic health expenditure: a multi-country analysis.� The Lancet 362 (9378): 111-7.
Yip, W., and P. Berman. 2001. “Targeted health insurance in a low income country and its
impact on access and equity in access: Egypt's school health insurance.� Health Economics
10 (3): 207-20.
30
SYS-REVIEW/HI-LMIC
Table 1: Description of the Studies
Study (1) Name of the Health insurance1 Data Methodology Funding source
country, and scheme (1) Benefit package
year of (2) National or (2) Target beneficiaries
launch regional (3) Premium
(4) Cost-sharing
arrangements
(5) Enrollment rate
1 Aggarwal (1) The Yeshasvini (1) Covers surgical procedures Survey of 4109 Propensity score matching; Bill and Melinda
(2009); Health Insurance of high-cost, low-probability, households; block clear selection equation Gates Foundation
Karnataka, Programme highly catastrophic medical matching to the insured; and balancing results; not through the Global
events and free out-patient clear if the presence of zero Development
India; 2003 (2) Regional (state cross-sectional; sample
department care expenditure is taken into Network
level) scheme (2) Rural farmers of includes households that account
cooperative societies and have never been offered
informal sector workers insurance
(3) Premium INR 120 (USD 2.4)
per person
(4) No copayment
(5) 3.0 million in 2008–09
2 Bauhoff et al. (1) The Medical (1) Most emergency outpatient Survey of 3500 Intention to treat Georgia Health and
(2010); Insurance Programme care and set of planned and households estimation regression Social Project
Georgia; 2006 for the Poor emergency inpatient care discontinuity design where Implementation
(2) National (2) Poor, 20 percent of enrollment is not very high; Center
Georgian population estimations for OOP
(3) Fully funded through the expenditures is presented
general government budget through generalized linear
(4) No copayments model with log link to
(5) Low enrollment account for those who
undertook zero
expenditure
3 Dow and (1) National health (1) Primary and secondary Vital statistics registries, Fixed-effect model of National
Schmeer insurance health care Census data; a panel for health outcomes on a Institute of Child
(2003); (2) National (2) Lower socioeconomic 88 to 97 regional data region for infant mortality Health and
31
SYS-REVIEW/HI-LMIC
Costa Rica; groups rates using Cox binary Human
1970s (3) 73 percent of children by transformation for infant Development
1984 mortality rates as the Grants
(4) Not reported dependent variable for the
(5) Not reported region
4 Gnawali et al. (1) Community-based (1) Consultation, essential and Cluster randomized with Propensity score matching The German
(2009); health insurance generic drugs, laboratory tests, 33 clusters, involving model to account for very Research
Senegal; 2004 (2) Regional limited inpatient hospital stays 4936 households; step- low uptake; no balancing Foundation
to 2006 (2) People in the informal wedge design; complete table, but selection model
sector, including the poor information is found for is present
(3) XAF 1500 per adult and XAF 1309 households; not
500 per child per annum in a clear where the
household (USD 1 = XAF 655) uninsured came from
(4) No copayment
(5) 5.2 percent in 2004, 6.3
percent in 2005, and 5.2
percent in 2006
5 Mensah et al. (1) National Health (1) Specified package of Survey by researcher; Used propensity score Global
(2010); Insurance Scheme general out-patient services, in- 393 insured women and matching; presents clear Development
Ghana; (2) National patient services, oral health, 1689 uninsured women, selection equation and Network and the
eye care covariate balance tables; Bill and Melinda
2003 randomized samples
(2) General population, results from different Gates Foundation
from regions; control matching methods that are
including people in the
group matched in the generally consistent
informal sector
area; reports on only 565
(3) Sliding scale: free for core
women who were
poor, between USD 8 and USD
pregnant; small sample
52
(4) Not reported
(5) 55 percent of the total
national population by August
2007
6 Thornton et al (1) Nicaraguan (1) Preventive, diagnostic, Insurance offered at Local average treatment The United States
(2010)[[Please Social Security maternity, and curative health randomly selected effect; selection into the Agency for
International
confirm services market booth; uses a insurance is modeled as an
Institute’s health Development’s
(2) Informal sector ordinary least squares with
correction of insurance programme pre-experiment baseline, Private Sector
(3) USD 15 per month the insurance offer as the
Partnerships-One
32
SYS-REVIEW/HI-LMIC
names and (Seguro Facultativo de (4) No copayment at the time 2610 households; loss to instrumental variable; the project and the
year or revise Salud): of service follow up 7 percent outcome measure is Global
according to (2)National (5) 20 percent of the sample (in differences-in-difference; Development
the experiment) results for those who Network
intended (study experiment on
enrolled given the
reference]]; increasing enrollment) enrollment procedure; no
Nicaragua; accounting for zero OOP
2007 expenditure
7 Trujillo et al. (1) Subsidized health (1) Basic health care services 1997 Colombia Living Propensity score matching, The University of
(2005); insurance program; (2) Low-income families Standards Survey; 5559 selection equation Central Florida
(2) National (3) Government funded insured through social presented (no balancing
Colombia; (4) A coinsurance rate health insurance system results) and instrumental
1993 that varies between 5 percent and 16,732 uninsured; variable estimation are
and 30 percent according to may not have been compared
the individual’s income insurance
(5) Not reported
8 Miller et al. (1)Régimen (1) Primary care, inpatient care Colombian household Intention to treat The Economic and
(2009); Subsidiado (2) Poor surveys (the Encuestas estimation of constructed Social Research
Colombia; (Subsidised Regime), (3) Fully funded through the de Calidad de Vida and eligibility from a survey; Council; UK
National Health general government budget the Demographic and uses regression Inter-American
1993
Health Surveys); nearly discontinuity design; the Development Bank;
Insurance (4) Low level of coinsurance
4300 families eligible and analysis also uses an National Institute
(2) National (5) Not reported instrumental variable of of Child Health and
marginally ineligible
constructed value for Human
eligibility on actual Development
enrollment; no accounting and the Stanford
for zero expenditure Center on
Demography and
Economics of
Health and Aging
9 King et al. (1) Mexican Seguro (1) A package to treat the Negotiated 74 paired Presents intention to treat National Institute
(2009); Popular de Salud diseases responsible for clusters to participate estimations and the effect of Public Health of
Mexico; 2005 (Universal Health approximately 95 percent of with one from a pair on experimental compliers Mexico; the
of average causal effect; Mexican
Insurance program- the burden randomly assigned to
the outcome measured is Ministry of Health;
33
SYS-REVIEW/HI-LMIC
SP) (2) People in the informal intense insurance uptake differences-in-difference; the National
(2) National sector campaign before national the complier results should Institutes of Aging
(study experiment on (3) Fully government funded enrollment; survey in 50 be understood as specific and the National
(sliding scale by income, free to the study; no accounting Science
increasing enrollment) pairs; 32,515 households
for the poor) for OOP expenditure of Foundation; United
(4) Not reported zero States
(5) Approximately 3.5 million
families
10 Sosa-Rubi et al. See King (2009) National Health and Nutrition Innovative multinomial The Health
(2009a) Survey 2006; Sample of probit that takes into Ministry of
Mexico; 2005 3890 women who account endogenous Mexico (SSA)
delivered between 2001 enrollment through
instrumenting by time of
and 2006; no one with
the introduction of SP,
employer insurance or assuming there is no
private insurance was problem of policy
included; complete data endogeneity.
for all women
11 Sosa-Rubi et al. See King (2009) National Health and Nutrition Propensity score matching The Health
(2009b) Survey 2006; 1491 adults for those with and without Ministry of
Mexico; 2005 with diabetes were SP insurance; presentation Mexico (SSA)
chosen; no pregnant of bias reduction and
selection equation
women or women with
access to social security
services; complete data
for all adults (see Sosa-
Rubi (2009a))
12 Wagstaff (1) Health care fund (1) Covered inpatient and Vietnam Household Propensity score matching Not reported
(2007); for the poor outpatient care only at public Living Standards Survey method on single-period
Vietnam; (2) National providers until 2005; some 2004 data; selection equation
preventive care
2003 and balancing is well
(2) All poor households and
selected other groups presented; not clear
(3) Fully subsidized whether those in the
(4) No copayment control group were offered
(5) As of 2006, the program insurance; results
34
SYS-REVIEW/HI-LMIC
covered approximately 60 dependent on propensity
percent of those eligible, score matching weights
accounting for approximately chosen; no accounting for
23 percent of the population
zero expenditure
13 Axelson et al. (1) Health care fund (1–4) See Wagstaff (2007) Vietnam Household Uses both single-period Not reported
(2009[[Please for the poor (5) Study reported 18 percent Living Standards Survey and differences-in-
(2) National enrollment for 2002 data 2002 (preprogram) difference outcome
confirm
and 2004 (postprogram); measures; uses survey data
correction
cross-sectional 10,232 as in Wagstaff (2007);
of year]]); and 4112 panel selection equation and
Vietnam; balancing results are
2003 presented from propensity
score matching; no
accounting for zero
expenditure
14 Wagstaff (1–5) See Wagstaff (2007) The panel 2002, 2004, Triple difference is the Not reported
(2010); The same as Wagstaff and 2006 Vietnam outcome variable; the
Vietnam; (2007) Household Living triple difference is
2003 Standards Survey regressed on covariates; no
Multipurpose household accounting for zero
survey; 1689 households expenditure
in all three waves
15 Jowett et al. (1) Voluntary health (1) Not reported Data were collected Two-stage multinomial United Kingdom’s
(2004); insurance program (2) Self-employed individuals, through a household logit model to examine the Department for
Vietnam; (2) National farmers, schoolchildren survey designed type of facility used; International
(3) Fully subsidized
1992, specifically to evaluate instrumental variable for Development
(4) A copayment of 20 percent
discontinued, the impact of the selection into insurance is
with exceptions
not currently in (5) 9.7 percent of target group scheme; analysis from used; appropriateness of
place 2631 households instrumental variable
tested; no theory given for
unusual instrumental
variable
16 Wagstaff and (1) Multiple health (1) Not reported Three surveys: (1) China Instrumental variable is The Spencer
35
SYS-REVIEW/HI-LMIC
Lindelow insurance schemes: (2) General population Health and Nutrition used to take account of Foundation Small
(2008); Labour Insurance (3) Not free Survey in 1991, 1993, selection with probit for and Major Grants
China; Scheme and (4) Not reported 1997, and 2000; (2) catastrophic measure and Programme;
then panel data are used; Fogarty
1996 onward Government (5) 90 percent of the Gansu Survey of Children
fixed effect is only used for International
Insurance Scheme population covered in 1970, and Families in 2000 and logit with no instrumental Center at the
(2) National but decreased to 20 percent 2003; (3) World Bank variable because these are
National Institutes
for rural population and 40 Health VIII project OOP expenditures, a
for Health; World
percent for urban population baseline survey in 1998; generalized linear model
with instrumental variable Bank Research
from 1980 onward; increased total sample was 18,200
is used to consider zero Committee
to 90 percent of urban workers adults
by 2003 expenditure by some
17 Wagstaff et al. (1) NCMS (1) Heterogeneity in the benefit Two data sets: (1) The Use propensity score The World Bank
(2009); China; (2)National package across counties and 2003 round of the matching to match the and the United
2003 coverage modes; all counties National Health Service insured with those who Kingdom’s
cover inpatient care, some Survey of the Ministry of have never been insured; Department for
cover outpatient Health; follow up in show balancing results but International
(2) Rural population 2005; no selection equation; Development
(3) The minimum premium (2) Routine Health subgroup analyses are
requirement was a CNY 10 (per Facility Survey from the presented by regressing
person) beneficiary Ministry of Health individual treatment effect
contribution from households, administrative data; total (weighted through
supplemented by government households > 8000 propensity score) on
subsidy income groups; most likely
(4) Deductibles, ceilings, and estimation of cost is for
coinsurance rates those receiving medical
(5) Not reported care.
18 Lei and Lin (1) NCMS (1–4) See Wagstaff et al. (2009) Longitudinal sample Differences-in-difference Not reported
(2009); (2) National (5) For NMS, 85.7 percent of drawn from the China using propensity score
China; the rural population were Health and Nutrition matching along with
2003 covered in 2008 Survey for 2000, 2004, instrumental variable
and 2006; different estimations and fixed-
analyses use different effect panel are used on
panel and thus have panel data; balancing
36
SYS-REVIEW/HI-LMIC
different data sizes; results are presented with
differences-in-difference no selection equations; not
is only for a panel of clear if OOP expenditures
3225 individuals includes zero expenditures
19 Wang et al. (1) Rural Mutual (1) Both outpatient services The Rural Mutual Health Propensity score matching Not reported
(2009); China; Health Care in China; and hospital services Care experiment adopted models with varied
2003 to 2006 a social experiment (2) Villagers, including farmers a pre-post treatment- matching and subgroup
(2) Regional and (3) Annual premium of at least control study design of analyses are presented, as
community-based in CNY 10 those not offered is balancing after matching
rural area (China’s (4) No copayment insurance; panel of 1665
western provinces) (5) 1173 households insured and 1745
uninsured individuals
1
No entry indicates not reported in the study
37
SYS-REVIEW/HI-LMIC
Table 2: Summary of Findings†‡
Utilization Financial Protection Health Status
1. Aggarwal (2010) India (Yeshasvini Community-based Health Insurance)
Those with health insurance Overall, medical expenses were n.a.
decided to use health facility in actually higher for the insured, with
greater numbers and with greater the poor experiencing no change;
frequency; increase from outpatient for hospitalization, expenditures
service usage, including outpatient are significantly lower for the
surgery; no higher usage in insured; also reported is the
frequency of hospitalization; less incidence of burrowing for hospital
usage of government services care, which is smaller for the
insured
2. Bauhoff et al. (2010) Georgia (Targeted Scheme for the Poor)
No impact on utilization from No robust evidence of lower n.a.
intention to treat estimations expenditures among insured
outpatients’ expenditures, except
for the elderly; lower expenditure
among insured for inpatient care
3. Dow and Schmeer (2003) Cost Rica (National Insurance Expansion)
n.a. n.a. No impact on decline in
community infant mortality
rates from increased proportion
of population insured over time
4. Gnawali et al. (2009) Burkina Faso (Community-based Health Insurance)
Overall, there is a significant n.a. n.a.
positive impact on health care
utilization; more outpatient visits,
but no significant impact on
inpatient care utilization; the higher
outpatient utilization is only
significant among the richest group
5. Mensah et al. (2010) Ghana (National Health Insurance Scheme)
38
SYS-REVIEW/HI-LMIC
The insured women who are n.a. Three types of health status are
enrolled are more likely to give reported, two of which (infant
birth in hospitals and to receive death and birth complications)
higher levels of prenatal care, are significant under specific
preventive health check ups, and matching weights; the
attention from trained health difference in infant death is
professionals likely to suffer from small
sample size
6. Thornton and Field (2010) Nicaragua
Local average treatment effect Local average treatment effect
measure; no significant effect on measure; overall decline in OOP
overall health care utilization; fairly expenditures decreased by a
substantial substitution away from smaller amount than the actual
use of public and private facilities to premium for the insured; no
health care facilities covered by significant result for OOP spending
insurance; social security hospitals for the insured reported; the
sample is too small to note effect in
catastrophic spending
7. Miller et al. (2009) Colombia (Targeted Scheme for the Poor)
Intention to treat estimations; Intention to treat estimations; no n.a.
substantial higher use of significant effect on average
traditionally underutilized outpatient expenditures; lowers
preventive services for those with inpatient expenditures and lowers
health insurance incidence of high-end expenditures
among the insured
8. Trujillo et al. (2005) Colombia (Targeted Scheme for the Poor)
Greatly increased medical care n.a. n.a.
utilization among the country’s
poor, including children, women,
and the elderly
9. King et al. (2009) Mexico (SP)
No effect in utilization Intention to treat and complier [n.a.
effects for low-asset holders show
lower OOP expenditures for the SP-
39
SYS-REVIEW/HI-LMIC
insured with low assets for overall,
inpatient, and outpatient care;
female-headed households had
lower inpatient OOP expenditures;
all insured had lower inpatient and
outpatient care, but no significant
effect was found for overall care
(including drug costs)
10. Sosa-Rubi et al. (2009a) Mexico (SP)
n.a. Different types of facility n.a.
utilizations are reported; these
imply different costs; those with SP
prefer SP facilities and the cheapest
care over private care and non-SP
government hospital service care,
which costs more than SP care but
is cheaper than private care; private
care is preferred to non-SP
government care
11. Sosa-Rubi et al. (2009b) Mexico (SP)
Those with SP had better access to n.a. Higher proportion of the insured
diabetes care; they had higher rates with glucose control, and lower
of insulin shots, regular tests, and proportion with very poor
physician visits glucose control
12. Wagstaff (2007) Vietnam (VHCFP)
Increase in both outpatient and Results robust to different n.a.
inpatient utilization but matching techniques and samples;
substantially increased inpatient there is no effect on overall OOP
care utilization; impact on expenditures; however, there is a
utilization among the poor is even lowered risk of high or catastrophic
less noticeable OOP expenditures; even with this
protection among the insured, one-
third still faced catastrophic
expenditures; the poor may have
received more risk protection from
high expenditures
40
SYS-REVIEW/HI-LMIC
13. Axelson (2009[[Please confirm correction of year]]) Vietnam (VHCFP)
Small but positive impact on overall From the cross-sectional study, the n.a.
health care utilization; the insured OOP expenditures were higher for
do not have greater difference in the insured; differences-in-
utilization of inpatient care; difference measure of two periods
statistically significant effect is showed a larger reduction in
present only for outpatient visits in expenditures for the insured for
community hospital inpatient care and a reduction in
catastrophic expenditures of 20
percent; the opposite result was
found for outpatients
14. Wagstaff (2010) Vietnam (VHCFP)
No impact on the use of either Triple-differencing estimates yield n.a.
outpatient or inpatient health care insurance effects with a sizable
services reduction of expenditures; poorer
groups also experienced a
significant reduction
15. Jowett et al. (2004) Vietnam (health insurance not presently in place)
Overall, insured patients are more n.a. n.a.
likely to use outpatient facilities,
public providers, and inpatient
services
16. Wagstaff and Lindelow (2008) China (Basic Medical Insurance Programme)
The results suggest that the insured The results vary for different data n.a.
may use health services more sets and specifications; the general
frequently; analysis suggests that picture that emerges is that
insurance facilitates the use of insurance results in a lower OOP
higher-level services expenditures and is likely to
increase the probability of incurring
catastrophic expenditures at
different threshold levels
17. Wagstaff et al. (2009) China (NCMS)
In the analysis by regions, the The results vary; for delivery n.a.
scheme increased outpatient and services, all regions showed lower
inpatient utilization; households costs for the insured; for the overall
with insurance have more doctor OOP expenditures, a mixed result
41
SYS-REVIEW/HI-LMIC
visits and inpatient spells; results emerges; total OOP expenditures
differ by health centers; the richest increased in most regions, except
quintile responded more favorably two, mirroring the case for
inpatient visits
18. Lei and Lin (2010) China (NCMS)
No significant evidence on increase Authors detect no impact on Among the estimation
in the utilization of formal medical expenditures from any of the presented, the propensity score
service; however, utilization estimations presented in the work matching shows marginally
significantly decreases the use of higher health for the insured
traditional Chinese folk doctors and
increases the utilization of
preventive care, particularly general
physical examinations
19. Wang et al. (2009)[[Please confirm correction of name and year]] China (Rural Mutual Health Care,
Community-based Health Insurance)
n.a. n.a. EQ-5D dimensions and specific
dimensions were reported for
the entire population with the
Differences-in-Difference
measure; decrease in illness is
significantly higher among the
insured for all dimensions and
for pain/discomfort and anxiety
and depression from two types
of propensity score matching
â€
All measures are the average treatment on the treated unless specified
‡
Empty cells indicate no information for the category
42