The financial protection effect of Ghana National Health Insurance Scheme: evidence from a study in two rural districts

With a gross domestic product per capita of US$ 306 in 2007, Ghana is a low income country in the sub-Saharan Africa region. The country compares positively with peers in sub-Saharan Africa in a number of important health indicators, such as maternal and child mortality, fertility rate, and HIV/AIDS prevalence among adults ages 15-49 [11]. However, Ghana also pays more for health as percentage of gross domestic product (6.2 percent in Ghana compared to the sub-Saharan Africa's average of 5.3 percent in 2006) [11].

Similar to many countries coming out of the colonial regime, after independence in 1957, the Ghanaian government adopted a tax-based health financing system in which the government paid for services in the public sector. By the early 1970s, general tax revenue in Ghana, with its stagnating economy, could not support a tax-based health financing system. Ghana started to introduce nominal user fees in the public sector, which in 1985 were raised significantly with the aim to recover at least 15% of recurrent expenditure [7]. This user fee system, known as "cash and carry," was shown to leave negative consequences in access to health services, especially by the poor people [12, 13]. To offset the adverse effects of user fees, the government introduced exemption policy for children, pregnant women, the elderly, the extreme indigents, and persons suffering from certain communicable diseases. In practice the exemption did not work well, and many of those who should have been exempted were not [14]. A number of adverse effects of the "cash and carry" system have been reported, including long delays in seeking health services when ill and incomplete prescription purchases [15]. In the beginning of the years 2000s, the share of household OOP payment in total health expenditure in Ghana was considerably higher than the regional average for the sub-Saharan Africa (50 percent versus 39 percent respectively in 2006) [11].

The establishment of the NHIS in 2003 is a major step by the Ghanaian government to address the perverse problems of the "cash and carry" system. NHIS was fueled partly by the relative success of the numerous mutual health organizations (MHOs), which existed in 67 out of 138 districts in Ghana with very diverse management structures and benefit packages [16]. However, NHIS marks an important departure from the MHO prepayment mechanism. It mandated the establishment of an MHO in every district and contracted existing private health insurance schemes. Although the administration of NHIS remains decentralized at the district level, its financing is centralized and the benefit package is standardized across the whole country. NHIS's benefit package covers a wide range of outpatient services with associated drugs and lab tests, inpatient care, treatment of cervical and breast cancers, basic oral health services, eye care, maternal care, and all emergency conditions. No coinsurance, copayment, or deductible is required at the point of service [17]. All Ghanaians, from both the formal and informal sectors, are in principle required to enroll. Public health facilities in the country are automatically accredited to contract with the NHIS, and private health facilities can apply for accreditation. It is estimated that in 2008, 479 private facilities were accredited while a larger number remained without accreditation [18].

Funding for NHIS comes mainly from a National Health Insurance sales tax of 2.5 percent, a transfer of 2.5 percent of formal sector contributions to the Social Security and Pension Scheme Fund, and from member contributions. In the beginning years of the scheme, 70 - 75 percent of total insurance revenue came from tax revenue, 20-25 percent from formal sector contribution, and only about 5 percent from contribution of members in the informal sector [19]. Premium is exempted for several member categories, including Social Security and National Insurance Trust Fund pensioners, children under age 18, adults above age 70, and the indigents [19]. As of 2008, the NHIS provided 41 percent of the total public resource envelope and about 45 percent of the Ghanaian population were NHIS cardholders [18]. This coverage level actually outperforms the targets set in the policy document, which aims to insure 30-40 percent of the population by 2010 [16].

Early experience in the implementation of NHIS reveals a number of problems. Despite the explicit policy to exempt premium for the indigents, poor people were much less likely to be insured than the better off [8, 9]. There has been serious delay in issuing cards for members, as well as in reimbursing the providers. Delay in provider reimbursement and inadequate supply side incentives are deemed responsible for the lower quality of care received by insured patients compared to non-insured, paying patients. Anecdotal examples of poor quality of care include provider discrimination against insured patients, long waiting time, low likelihood of being seen by a doctor and of receiving all drugs prescribed [18]. In particular, it has been reported that providers commonly solicited informal payments from the patients by charging for services out-of-hours, asking patients to pay for drugs, which are said not to be in stock, and asking patients to pay for "better" drugs, said to be not provided under the NHIS [18].

To date the impact of NHIS on health service utilization and OOP expenditure has not been rigorously evaluated at the national level. Several studies conducted in a small number of districts suggested that NHIS has encouraged utilization of curative health services [9, 20]. None of these studies, however, successfully dealt with the issue of adverse selection of NHIS membership.

One of the key values of health insurance is to help people deal with the unpredictability of illness and medical spending [21]. It is well known that health expenditure is extremely variable and skewed, with a disproportionately large amount usually concentrated in a small portion of the population. For example, data from the United States in 1987 show that the top 10 percent of users account for nearly 75 percent of total medical spending [22]. Without a proper risk pooling mechanism, households that are unfortunate enough to have catastrophic illnesses will find it very difficult to cover the medical bills. The financial consequences of ill health can be especially devastating for poor people, who already struggle to cover the basic daily needs such as food and shelter. A recent analysis of 15 African countries revealed that in most of these countries, around 30 percent of all households financed OOP health expenditure by borrowing and selling assets [23]. Health insurance is designed to reduce the amount that households have to pay OOP when they use medical services.

Looking at the absolute amount of OOP payment changed as the result of having health insurance is not adequate to judge its effect on the financial burden of health care. Low OOP payment may just derive from the fact that insurance does not trigger any increase in utilization; hence no additional payment is needed. On the other hand, high OOP payment may not necessarily be bad if it buys substantial improvements in quality and/or quantity of services and if the households can afford to pay [6]. Thus, the issue of financial burden is more directly related to capacity to pay than to the absolute amount.

Along this line of argument, Wagstaff and van Doorslaer [24] defined a concept of "catastrophic" health expenditure - expenditure that exceeds some pre-specified fraction of household income. The idea is that health spending comes at the expense of other household consumption expenditure, and health spending should not exceed a certain threshold z, so that households have at least (1-z) of their income to spend on other needs. The catastrophic expenditure indicator is a commonly accepted tool for measuring the financial burden of health care relative to the household capacity to pay. The concept has been employed widely to portray the health financing profile and to evaluate the impact of health insurance in many countries [5, 25‐27]. Although one may argue that z should vary by income level, because rich people may afford to spend higher portion of their income on health compared to the poor, no theoretical or empirical work has been conducted to date to address this issue.

Following the practice of the existing literature, we adopt two measures of the financial protection effects of insurance in this study. They include the absolute amount of OOP expenditure and the likelihood of high or catastrophic expenditure.

A generic model for estimating the impact of health insurance for an individual i with the outcome of interest Y can be estimated as followed:

where the F function in health service research typically takes the form of linear, probit, logit, or poisson depending on the distribution of the outcome variable, HI denotes insurance status, X is a vector of observable characteristics, Z is a vector of characteristics unobservable to the researchers, and ε is the stochastic term, usually assumed to be normally distributed. Note that included in X and Z could be values shared by all household members.

Because Z is unobservable, it is included in the error term. The estimate of the health insurance (HI) effect on Y will be biased if Z contains at least one element that is systematically correlated with both insurance status and the outcome of interest (HI is endogenous). For example, if Y is a measure of health service utilization, typical candidates for such an element of Z would be the individual's risk aversion or poor health condition, which are likely to be positively correlated with both insurance and Y. In such a case, failure to control for this interfering effect of the omitted variables will lead to an overestimation of the insurance effect. Unfortunately, while tools for dealing with this omitted variable bias are widely known (such as randomized controlled study, individual fixed effects model, or instrumental variable), in reality these tools often are not readily available.

In this study, we obtained information on the individual chronic health condition and whether the study respondents assessed their health status as "bad." While ultimately there will be other confounding factors that we have not controlled for, poor health status is a commonly cited reason for adverse selection in the insurance literature [21]. Thus being able to account for poor health status helps improve the validity of the study findings. More importantly, for the type of outcome studied (health expenditure), failure to control for the omitted variable bias will likely lead to a lower bound estimate of the insurance effect, provided that the signs of correlation between such omitted variable with both insurance status and the outcome are similar. This ability to bound the insurance effect is important - if we find some significant result, we know the true effect will be significant.

Table 2 presents the breakdown summary of OOP health expenditure by insurance status for the study sample. The sample consists of 4,899 people having NHIS status (members) and 6,718 people without NHIS status (non-members), yielding a weighted insurance rate of 35 percent. This rate is lower than the administrative record of the NHIS shown in Table 1. One reason for this discrepancy could be that members have to wait for six months after registration to begin accessing services, and many received their NHIS card only after a delay. Therefore, the registration information does not give a precise measure of the actual number of cardholders [18]. Because 99 percent of all insured people in the sample have NHIS, we ignore other types of insurance. For simplicity, individuals with NHIS status will be called "insured" in this analysis.

As shown in Table 2, on average, an insured person paid 21 thousand Cedi over the 12 months preceding the survey and an uninsured person paid nearly 30 thousand Cedi (we report old Cedi used before the redenomination of currency in July 2007; exchange rate: US$1 = Cedi 9,302). Thus, despite the generous benefit package, insured people still incurred an OOP expense roughly 72 percent of that paid by uninsured people.

Some interesting observations emerge from Table 2. First, insured people still report paying for items that should be covered by insurance, such as consultation fees, lab expenses, and drugs purchased at the facility (see description of benefit package in Table 1). This pattern suggests that the NHIS does not fully shield its members from OOP payment at the point of service. Second, compared with the uninsured, insured people pay more for items not covered by insurance, such as informal care, unofficial payment to providers, and purchase of drugs outside the facility. The higher amount spent on informal care among the insured suggests that insured people may indeed sicker than the uninsured. On the other hand, the fact that insured people pay more for drugs purchased outside facility (hence not covered by insurance) and for "other expenses" provides evidence supporting the earlier observation that quality of care may be a problem with insured services. The pooled average of OOP expenditure among the insured and uninsured is 26,922 Cedi (table not shown), rather low compared with the country mean annual per capita income of 3,970,000 Cedi reported in the Ghana Living Standard Survey 2005-2006 (0.68 percent). This figure also is rather low compared with that of other developing countries, especially given that insurance coverage was 35 percent at the time. However, it is important to note that this OOP amount only includes expenditures associated with specific illness, hospitalization, antenatal care, or delivery. Hence it does not capture expenditures associated with preventive care and purchase of drugs for health maintenance.

Table 3 presents the summary statistics of relevant variables. For completeness, it also shows the total OOP expenditure reported in Table 2. As Table 3 shows, only 3-4 percent of the study sample actually incurred any positive OOP payment for health services. The incidence of high and catastrophic expenditure ranges from 0.9 percent to 2.8 percent depending on the indicators. Still, the pattern is very consistent; the incidence of high and catastrophic OOP expenditures is noticeably lower among the insured for all measures, suggesting that the NHIS did have some protective effects. Xu et al. [25], using Ghana 1998-1999 Living Standard Survey data and 40 percent of non-food consumption threshold, estimated a catastrophic incidence rate of 1.3 percent. For the purpose of checking comparability, we also computed a catastrophic expenditure variable based on Xu's et al. threshold using our method (i.e., employing quintile-specific income and expenditure data from Ghana Living Standard Survey 2005-2006 as denominators). The incidence rate obtained is 0.95 percent, which is reasonable given that our estimates and those of Xu et al. used data that were 8-9 years apart. Still, our figures could be slightly underestimated because the OOP variable does not cover all expenses, as mentioned above.

Table 3 also shows that the proportions of insured people with reported chronic health conditions and bad health status were twice as large as those among the uninsured, suggesting that adverse selection exists. However, health status is not the only factor relating to insurance. As shown, insured people are more likely to be female, from ethnic Akan, and living in smaller, female-headed households and in households with at least one member joining a local solidarity scheme. There is also a very strong socioeconomic gradient of insurance, with coverage being systematically higher among the urban, wealthy, and highly educated. This differential coverage of NHIS among socioeconomic groups has similarly been documented in many other studies [8, 9].

The socioeconomic gradient observation is corroborated in Figure 1, which plots insurance coverage and incidences of high and catastrophic payments against wealth quintiles. As revealed, there is a clear upward trend in insurance coverage, with the level among richest quintile being more than 2.5 times higher than that among the poorest quintile (50 percent versus 20 percent, respectively). Correspondingly, there is also a clear downward pattern of high and catastrophic expenditure, with poor people having much higher incidence compared with the rich on all measures. Besides the clear concern about equity in coverage and financial protection, the patterns in Figure 1 suggest some relationship between insurance coverage and large OOP payment, which we will test formally in the subsection that follows.

Table 4 presents the first set of estimation results assessing the effects of NHIS on the absolute amount of OOP payment. Model 1 controls for self-reported chronic health conditions and health status while model 2 omits these variables. Both models include individual and household characteristics described in table 3, namely the person's age, age squared, gender, ethnic Akan, if s/he is a member of a solidarity scheme, household size, the household head's gender, education, and occupation, as well as household's assets and living conditions. For clarity of presentation, the effects of these covariates are not shown in the table because they are not of primary interest. The key figures of interest are those for "has HI from NHIS," which show the marginal effects of insurance on OOP expenditure. As explained earlier, these marginal effects, estimated from the two-part model, combine insurance effects on the probability of having any OOP payment and the absolute amount given that a payment is made. The robust standard errors, adjusted for sampling weight and clustered at the municipality level, were estimated by boostrapping with 200 replications.

Both models 1 and 2 show that the insurance effect is in the expected direction, i.e. insurance has been shown to reduce OOP among its members. The marginal effects of insurance are smaller (in absolute value) when negative health status indicators are not controlled for (model 2 compared to model 1), suggesting that the derived estimates could be on the lower bound if adverse selection due to poor health status has not been fully controlled for. If there remain omitted variables that drive adverse selection in the same way as health status (e.g., residual wealth status), the true insurance effect will be larger than that observed in models 1 and 2.

It is noteworthy that in model 1, insurance effect is only marginally significant (at 10 percent level) and rather small (33,821 old Cedi). To examine its magnitude relative to household capacity to pay, we compare that amount with the average figure on household non-food consumption expenditure from Brong Ahafo region, where Nkoranza is located. We take the household non-food consumption expenditure per capita for Brong Ahafo reported in the Ghana Living Standard Survey 2005-2006 (Tables 9.2 and 9.5) [28] and adjust it for inflation and real growth to obtain the comparable figure for 2007, which is 2,701,300 Cedi. Thus, an effect of 33,821 Cedi would be equal to 1.25 percent of non-food consumption expenditure, which is not substantial. Again, it is important to keep in mind the large confidence interval of this estimate (0.24 percent-2.75 percent of non-food consumption expenditure).

Table 5 shows the estimation results for the various catastrophic OOP payment measures. All estimations control for health status and usual individual and household characteristics. A clear and consistent pattern emerges across all measures: having NHIS significantly reduced the probability of catastrophic OOP payment on health services. The estimated reduction ranges from 0.5 percentage point (for expenditure of at least 20 percent of non-food expenditure) to 1 percentage point (for expenditure of at least 5 percent of income). In relative terms, a reduction of 1 percentage point would be 67 percent compared with the mean of the insured and 36 percent compared with the mean of the uninsured (Table 3). Consistent with the finding on OOP expenditure, males have a significantly lower incidence of catastrophic payment. Interestingly, while Offinso is less poor than Nkoranza, Offinso residents were more likely to incur catastrophic payment. Because Offinso has more hospitals, more private and mission clinics, and higher insurance premiums and registration fees, it is possible that the price of services is also higher in Offinso, which explains higher incidence of catastrophic payment. However, further data and investigation are needed to form a rigorous explanation of the difference between the two study districts.

A question of interest is whether the effect of insurance works differently for the poor versus non-poor. We examine this by performing separate estimations for the poorest 20 percent versus the rest of the sample. The results, presented in Table 6, are telling. The effects of insurance are large and strongly significant in all measures among the poor, while marginally or non-significant among the rest of the sample. In fact, most of the average effect observed in Table 4 actually comes from effect among the poor. For the rest of the population, the effects have the expected direction of effects, but are too small to yield any statistical significance.