The financial protection of national health insurance: Evidence from a cross section of State and Federal workers in Akwa Ibom, Nigeria

Background: Public health insurance schemes can offer households financial protection against health care costs and help to resolve inequality in health care provision. The current study evaluates the impact of the Nigerian National Health Insurance Scheme (NHIS) in reducing financial hardship for a sample of Nigerian households working in the health and higher education sectors. The data allows us to examine the variation in the financial protection effects across different income groups and explore differences in standard of living by households with coverage and those without. Methods: Data was gathered in Akwa Ibom state, Nigeria. A cluster sampling technique is used to compare participants and non-participants in the NHIS and within this, identify equivalent groups with regard to household characteristics such as education level, income and household composition. A propensity score matching approach examines variations in out-of-pocket expenditure (OOPE), catastrophic health expenditure (CHE) and number of household assets across the insured and uninsured groups controlling for cofounding factors. Results: T he likelihood of experiencing CHE for a household that is insured is estimated to be 82% lower than that of an insured household, even after controlling for our variety of observable characteristics. OOPEs are ₦50,000 lower in households with insurance compared to those without. We additionally find a significant difference in standard of living, as measured by household asset ownership across the insured and non-insured groups Conclusions: There is a statistically and practically significant association between participation in the NHIS scheme and household financial protection. This provides support to policy-makers seeking to design and extend equitable health-financing policies.

Health insurance can play a critical role reducing inequality in health care and supporting economic growth through stabilising household finances [1][2][3]. However, across many developing countries, significant inequalities exist in access to health insurance, health care utilisation and health outcomes [4]. In 2005, Nigeria launched the National Health Insurance Scheme (NHIS) to help promote access to quality healthcare and protect households against the financial effects of ill-health.
This was viewed as a first step towards universal health coverage (UHC), an important aspect of the UN Sustainable Development Goals that emphasises the importance of providing affordable, quality health services [5]. Catastrophic health expenditure varies widely across developing countries and is linked to three preconditions; the availability of health services requiring payment, low capacity to pay, and the lack of prepayment or health insurance [6]. Most developed countries have advanced social institutions such as social insurance or tax-funded health systems that protect households from catastrophic spending.
Nigeria is Africa's most populous and largest economy and prior to the pandemic, it also had the highest out-of-pocket expenditure (OOPE) on healthcare in the African Union, estimated at 77% of current health expenditure in 2017 [7]. Moreover, Catastrophic Health Expenditure (CHE), de-fined as spending in excess of 40% of monthly non-food expenditure, was experienced in twentyseven percent of households in southeast Nigeria [8]. Such unforeseen expenditures can eliminate savings, pushing households towards greater indebtedness and poverty [9]. There is also evidence of unequal effects of healthcare spending as households in the lowest income quintile in south-east Nigeria have a higher incidence of CHE relative to higher-income households [10]. Furthermore, the continuing  pandemic has revealed underlying inequalities in access to health care, as it has had a greater impact on lower income groups [11]. Previous studies in Africa and Mexico have evidenced the financial protective effect health insurance schemes may have, particularly for lower income households [12][13][14][15][16]. In this context, we use data on a sample of federal and state workers in Akwa Ibom, Nigeria to examine the relationship between enrolment in a national health insurance scheme and household health expenditures.

Policy Background
The Nigerian case provides a unique opportunity to examine the impacts of a national health insurance plan on household expenditures. Onoka et al. 2013 [17] point to the partial adoption of the NHIS programme largely based on employment sector. Federal employees and their households make up nearly the entirety of the five million participants in the NHIS [18]. However, the programme has been unable to expand coverage beyond federal government employees as planned [19][20][21]. Ozili 2020 [22] points to the low roll-out of the NHIS as evidence of the frailty in the health care infrastructure in Nigeria that is exacerbating the impact of the  pandemic. The critical question about the level of financial protection provided by public health insurance has recently been explored internationally by Erlangga et al. 2019 [23], however Nigeria's NHIS has received scant attention, most likely due to data constraints. Ijeoma et al.'s 2019 [24] review paper on the incidence and determinants of CHE in Nigeria, identifies significant variation in the incidence of CHE in Nigeria across different studies. Aregbeshola and Khan 2018 [25] is to our knowledge the only study that explores the relationship between health insurance coverage and CHE in Nigeria. A significant negative relationship is found between participation in health insurance and experiencing CHE. The study categorises households as non-participants in the NHIS if they did not incur OOPE. This approach is likely to overstate the numbers with health insurance.
Health insurance participation among their sample was 22.1%, far higher than the national estimate of 5% [25] and so further analysis is required. Prior to this study, no available data directly related The current study uses a unique dataset of over five hundred federal (insured) and state (noninsured) employees to evaluate the effects of NHIS on household health expenditures and standards of living for a sample of Nigerian households working in the health and higher education sectors. The construction of such a dataset allows us to explore the benefits of expanding the NHIS programme to State employees as originally envisaged and help inform broader health policy. We are also able to examine the differences in the number of assets a household holds for those with coverage and those without.
The NHIS regulates, monitors, enforces quality control and administers the Social Health Insurance (SHI) programmes operated by the federal government in Nigeria. We set out a short overview of key aspects of the NHIS here, while provide a comprehensive description of the NHIS and its implementation can be found elsewhere [15,24]. Federal government employees were mandated to enlist in the programme, while political and institutional factors inhibited adoption of the programme at the state level [15,24]. As a consequence, state employees remain largely outside the scheme in all but three of the thirty-six states in Nigeria. 1 The NHIS is comprised of three packages, the USSHIP, the RCSHIP and the FSSHIP with the latter the focus of this study. The FSSHIP is directed towards employees in the public sector, organised private sector (employers with more than 10 employees), the armed forces, the police, para-military institutions, students in tertiary institutions and voluntary contributors. It is implemented by NHIS-registered Health Maintenance Organisations (HMOs) and accredited Healthcare Providers (HCPs). The HMOs are publicly-or privately-owned limited liability companies responsible for the collection of premiums from enrollees, payment for the services accessed by enrollees and quality control of healthcare services offered by the accredited HCPs. The HMOs operate within a competitive structure under a model of managed care, with the objective of enabling costeffective delivery of health care [25]. The HCPs are clinics, private and public hospitals at the primary, secondary and tertiary levels that provide healthcare services to enrollees. The HCPs play the role of gate-keepers as they are the first point of contact before referrals are made for secondary or tertiary care [28].
Inclusion in the FSSHIP requires an annual premium equal to 15% of the annual basic salary of an enrollee, with 5% of the contribution coming from the enrollee and 10% from the enrollee's employer. The premium can be paid in lump sum or in monthly instalments to cover health care benefits as specified by the policy for the employee under the age of 65 years, a spouse and four biological children below the age of 18 years. 2 The benefits available on the FSSHIP are summarised in Table 1 and are subject to a waiting period of five months after the payment of the first premium. The NHIS pays capitation for primary care and fee-for-service upon referral through an accredited Health Management Organizations (HMOs) to an accredited secondary or tertiary healthcare facility.
[INSERT TABLE 1 ABOUT HERE] 2 Where the number of children below age 18 is more than four, an additional payment is required.
The rest of this paper is organized as follows. The next section explains the data sources and methods. The following section presents the results. The final two sections present a discussion of the results, policy implications and lessons learned for other low and middle-income countries currently undertaking similar health financing reforms and then concludes.

Study setting, sampling and empirical approach
This study was conducted in Akwa Ibom, a state in southern Nigeria with an estimated population of over 5.45 million. Akwa Ibom is rich in crude oil and has thirty-one local government areas with Uyo as the state capital. Enrolment on the FSSHIP is compulsory for federal employees within the state and unavailable to state employees. Similar to Nguyen et al. 2012 [29], we are not comparing households before and after insurance coverage, but instead we compare households with and without insurance. The variation in NHIS coverage across federal and state employees in Akwa Ibom provided an appropriate sample.
We undertake a cluster sampling technique to select the participants and obtain cross-sectional data [29]. Purposive sampling is used to select four sample clusters, namely; employees from two federal and two state institutions. We limit our sample to federal and state employees in health and education institutions to best identify equivalent groups with regard to household characteristics such as education level, income and household composition. The federal institutions were; the University of Uyo and the University of Uyo Teaching Hospital with state employees were drawn from, Akwa-Ibom State University and Etinan General Hospital. Twenty postgraduate students from the University of Uyo conducted the interviews. The interviewers were provided with two days of training on the data collection process and interviewing skills. Face-to-face surveys were then undertaken with participants at their place of work.
To more closely evaluate the financial protection effect of NHIS participation, our survey was focused on the sub-population where a household participant had experienced at least one episode of illness (using a 4-week recall period). This approach is consistent with the previous literature [24,29,30]. We additionally gather information on their NHIS participation status, medical care utilization, expenditure on healthcare, food and other non-food expenditures, total income and other household demographics 3 . In total 560 respondents participated in the survey. Of these, twenty-two were incomplete leaving an estimable sample of 538, 273 state employees and 265 employed in federal institutions.

Variables
The outcome variables in this study are out-of-pocket expenditure (OOPE) and Catastrophic Health Expenditure (CHE). OOPE is the total monthly outlays on healthcare costs, including consultations, drugs, hospital costs, transportation to and from where treatment was received and other cost directly related to the restoration, improvement, and maintenance of health [10]. CHE is measured using a threshold of 40% monthly non-food expenditure [31,32]. Total household expenditure by broad category was gathered and food expenditure was subtracted from this to obtain household non-food expenditure. The treatment variable of interest is whether a household, by having a member employed in the federal or state sector, is enrolled or not enrolled on the NHIS.
Household characteristic data also collected in in the study are; employment sector, education level of the household head as well as the location, income group, size and self-reported frequency of health care utilization of the household. A description of these is included in Table 2.

Propensity Score Matching
While a descriptive analysis of OOPE and CHE is presented in the next section, to better understand the relationship of interest, we utilize propensity score matching (PSM) [33]. This method estimates the treatment effect of being enrolled on the NHIS by comparing the treatment group (federal workers) and the control group (state workers) based on matching the two groups using a vector of confounding variables. Specifically, the matching pairs each treated unit with a control unit based on the estimated propensity scores. The method is widely used in health evaluations, including those studying the effects of health insurance [34] [35] and noted by Erlangga et al. 2019 [23] as one method to help avoid selection bias in observational studies evaluating health insurance schemes. With PSM, we are able to construct a control group that comprises of households that do not participate in NHIS but who have the same probability of participating based on a set on observable factors and compare them with those who participated in NHIS and estimate the effect of participation More formally, the propensity score is defined as the conditional probability of enrolment on NHIS given household characteristics (confounders): where D = {0, 1} is the NHIS enrolment status and X is the vector of confounders. From (1), it can be shown that if NHIS enrolment status is random within cells defined by , then NHIS enrolment status is also random within cells defined by the value of the one-dimensional variable, ሺ ሻ. Given a population of households, denoted by if the propensity score ሺ ሻ is known, then the Average Treatment effect on the Treated (ATT) can be estimated as: where is the ATT, 1 are the potential outcomes in the situations of enrolment on NHIS and 0 , are the potential outcomes in the counterfactual situations of non-enrolment on NHIS. It is noteworthy that to derive (2) given (1), it must be established that enrolment on NHIS is un-confounded given the propensity score, that is, This simply means that the matched sample must be balanced given the covariates. To generate a balanced, matched sample with PSM, different matching methods are proposed in the literature but each method implies a trade-off between quality and quantity of matches and none of the methods is believed to be superior to another. This study utilizes the kernel method of PSM to construct the counterfactual outcomes for households that are insured (treatment group) using all the uninsured households (control group) 4 .

Covariate Balancing and sensitivity
We first present results of covariate balancing after propensity score estimation. For our sample of 538 households, all households have at least three nearest neighbors to provide a match. Figure 1 below shows the box plot of raw and matched samples while Figure 2 shows the kernel density plots for the distribution of the propensity score before and after matching. Both figures illustrate a satisfactory level in the balance achieved. With regard to the composition of both groups, Table 3 shows that they are similar in terms of education, income, age, household size with the main difference stemming from geographical location and medical care utilization. Specifically, 46.15% of the uninsured households reside in urban areas while 89.91% of insured households are concentrated in urban areas. Insured households in our sample use medical care more frequently than uninsured households, indicating that enrolment on the NHIS induces some degree of moral hazard health [35,36].

Descriptive results
Household standard of living was based on household ownership of a representative basket of ten household assets. This approach is shown to be at least as reliable as conventionally measured consumption expenditures and sometimes more so [32,33]. The objective was to measure each household's ownership of a basket of goods, in this way capturing the benefits that people receive from publicly provided goods. 5

Empirical Results
While these descriptive statistics provide a useful summary of the differences in CHE and OOPE between those with insurance and those without, we also present the average treatment effects on the treated (ATT) estimated by PSM in Table 4. We find that after controlling and balancing for our observable set of characteristics such as income, medical care utilization and urban/rural location that the probability of experiencing CHE for an insured household is 82% lower than that of an uninsured household.
[INSERT  [36], households were asked to indicate ownership of assets from a representative basket of ten items. Similar approaches to evaluating standard of living have been applied in the Nigerian context in the context of the socioeconomic impact of remittances [37]. Quantifying household standard of living using this approach has also been applied to examine the socioeconomic impact of health care utilisation in Palestine and Tunisia respectively [38,39]. From the descriptive data presented in Table 3, on average, households with no insurance had a value of 4.76 in this index compared to 6.4 for those with insurance. This suggests a significant difference in standard of living across the two groups and similar to Hailemichael et al. 2019 [39], to better examine this relationship we also utilized a PSM model with the same household characteristics used for balancing in our previous model. While not presented here for reasons of brevity these results show that households covered by the NHIS own 1.5 more assets relative to the uncovered households 6 .

DISCUSSION
Our analysis presents some noteworthy findings for discussion. First, the results from our sample suggest the NHIS has achieved positive outcomes by reducing the incidence of CHE by 82% and OOP health expenditures by ₦50,000. The results demonstrate the notable protective effect of the NHIS, showing that the structure of the scheme provides a comprehensive range of cover and minimises co-payments by participants. This is a more pronounced effect when compared to Ghana (3% reduction in CHE) and Mexico (54% reduction in CHE) [12,14]. The high level of variation across countries may be attributed to several factors, for example; the level of insurance cover (conditions covered, co-payments) as well as institutional factors (capacity, range of medicines, indebtedness of insurance scheme). The evidence suggests, that participants in the Nigerian NHIS are actively using their insurance cover when participating in the health care system, showing that the NHIS is functioning well, albeit for a relatively small proportion of the total population.
Second, our results indicate that insured households on average spend more on education, food and rent and demonstrate a higher standard of living, proxied by ownership of household assets.
Furthermore, insured households save more during the reference period. Taken together these results suggest a powerful effect of insurance on household expenditure and saving activity.
Our study has a number of limitations in the interpretation of the results. Firstly, the short recall period of 30 days prior to the survey enables respondents to provide reliable payment responses but may underestimate longer term effects. Secondly, we use a representative basket of ten assets to create a proxy for the household's standard of living, however we applied a simple equal weight to ownership of each asset. A more sophisticated approach would have required households to additionally indicate the purchase price and current value for each asset under ownership. This would have allowed the construction of a more precise measure of standard of living. Finally, our survey was limited to federal and state employees in Akwa Ibom. Generalising our results to a broader population and to states other than Akwa Ibom is thus made more difficult as the quality and accessibility of health care institutions is likely to vary across the country. This in turn will impact on the financial protection effect of the NHIS and the requirement for insureds to co-pay to ensure faster access to the health care system.

Funding
This work is funded by a the Kemmy Business School PhD scholarship fund. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Availability of data and material
The data used in the analysis is freely available on request from the authors.