The Ghanaian National Health Insurance Scheme

Data from the 2012–2013 Ghanaian Living Standards Survey (GLSS) showed that 48% of adults aged 18–49 were enrolled in the NHIS, which increases to 58% for older adults aged 50 plus. The NHIS was initiative by the 2003 National Health Insurance Act and was put into operation in 2005.

The NHIS is funded through a range of independent sources, including the National Health Insurance Levy, which is a 2.5% value added tax; returns on National Health Insurance Fund investments; premium fees and contributions to the Social Security and National Insurance Trust (SSNIT; 2.5 percentage points per month). There is also further government funding to complement the scheme.9

The country’s previous ‘cash and carry system’ was seen as a barrier to access to healthcare services for many Ghanaians unable to afford unplanned cash payments when seeking care.10 The NHIS operates under the vision of being ‘a model of a sustainable, progressive and equitable social health insurance scheme in Africa and beyond’.11

Although it is mandatory to be part of an insurance scheme in Ghana, the literature often argues that in practice enrolment is voluntary. No penalties apply for not being a member of any insurance scheme and citizens are not enrolled by default.12 In order to become a member of the NHIS, individuals need to register at the local district office and pay a registration fee and a premium. The health insurance membership ID card, which provides evidence of enrolment, will then be provided for participants after a waiting period. Immediate biometric registration at selected district offices has been launched only recently to optimise the enrolment process.9 NHIS membership expires after 1 year and thus requires renewal to remain eligible.

On reaching 70 years of age, individuals are exempt from the NHIS premium payment if they can provide a proof of age, although they still have to pay the registration fee. Other groups that are exempt from paying the premium include the ‘core poor’, defined as those who cannot show a source of income and have no fixed residence.9 A number of smaller groups, such as those who contribute to the SSNIT or receive a pension from the SSNIT, are also exempt from the premium.

The scheme operates a negative list for the benefit package, with a comprehensive benefit package available apart from specific exclusions. Officially, the NHIS covers 95% of the common disease burden in its insurance package.9 The NHIS package includes outpatient services, inpatient services, oral health, eye care, maternity care and emergency care.9

Cancer treatment, with the exception of cervical and breast cancer treatment, is not covered even though the proportion of deaths due to cancer is increasing.13 The NHIS does not include psychiatric services in its benefit package, but mental disorder treatment is freely available in government psychiatric hospitals as well as through community psychiatric nurses. If, however, these services are not accessible, then alternatives have to be purchased out of pocket.14 Homecare, hearing aids and dentures are excluded from the benefit package as well.15

There are known to be issues with the NHIS with regard to long queues and waiting times.16 This has a disproportionate effect on older adults, who are often physically not able to queue for a long time while waiting for care.17 In addition, large distances to healthcare facilities can be a barrier for less mobile older adults to actually use the care. The requirement to first enrol in the scheme, even if there is an exemption from the premium, discriminates against older adults, as travel to the local district scheme office may be problematic for some due to mobility or transport issues.

Moreover, rural dwellers are poorly served by the health system in Ghana in comparison to their urban counterparts. Sulemana and Dinye5 found that due to lack of healthcare facilities, long travel distances and poor transport systems, rural residents are considerably disadvantaged in terms of accessibility to healthcare services. Alhassan and Nketiah-Amponsah18 reported that in rural NHIS accredited facilities drug and resource availability was inferior to that in urban facilities. Overall, only 30% of qualified doctors and nurses in the country work in rural areas,18 even though 54% of adults age 60 and over live in the same area.4

Previous studies that have analysed the effects of the Ghanaian NHIS have mainly focused on maternal healthcare or the probability of women seeking inpatient and outpatient care.12 15 These have found that people enrolled in the NHIS were more likely to seek formal healthcare and that women were more likely to deliver at a hospital. However, it is uncertain whether these findings can be extended to older adults due to the great complexity of treatment, issues with access as well as the care often being required over a longer time period than in maternity.

Increasing use of healthcare is an important step of improving health outcomes in the longer term.19 However, the magnitude of the link between utilisation and outcomes is affected by the quality of the services offered, the timeliness that care is sought and follow-up care, which is outside the scope of this paper.

Data and methods

The data for the analysis were taken from the nationally representative GLSS and its supplementary community survey. The first GLSS was conducted in 1987. Round six, the latest (2012/1013) GLSS, sampled 18 000 households within 1200 Enumeration Areas.20 In round six, a response rate of over 93% led to 16 772 households being interviewed. The GLSS collected data on sociodemographic characteristics, health, employment, household expenditure, income and housing conditions.20

The associated national rural community survey was carried out in 655 rural enumeration areas which had been selected during the 2012/2013 GLSS data collection process.21 The community survey provides information on a household’s environment, including a facility questionnaire including information on the availability and accessibility of local services like healthcare providers. Further information on the community survey can be found in the Community Facility Report.21

The analysis was restricted to individuals aged 50 and above. This group were classified as older adults as it is contended that the commonly used chronological definition of 65 is not appropriate when conducting research in an African context. In an African context, a chronological age of 50 or 55 serves as an appropriate cut-off point when defining an older adult. The average life expectancy in low-income countries for men is 60 years and for women 63 years,22 while in Ghana, the life expectancy at birth is 62 years and the healthy life expectancy at birth 54 years.23 Therefore, to define people aged 50 plus as ‘older adults’ is more appropriate when looking at low-income and middle-income countries.24

In order to understand the impact of the NHIS on healthcare utilisation, it is necessary to control for differences in personal characteristics between the insured and non-insured. As a randomised control trial was not feasible or ethical, statistical matching was one way to control for selection bias.25 This method creates a control group that is as similar as possible to the group that received the intervention in its observable characteristics, allowing the determination of the effect of an intervention. This paper applied PSM to analyse the effect of NHIS enrolment on healthcare usage. Using different matching algorithms to estimate coefficients allowed a measure of robustness in the results. Therefore three matching techniques were compared here in order to ensure the robustness of the results: (1) nearest neighbourhood (NN) with replacement, (2) radius and (3) Gaussian kernel matching.

Radius matching, also referred to as NN matching within a defined calliper distance, is a good alternative to NN matching when failing to find neighbours with the same propensity score.26 Radius matching refers to defining a maximum score distance, also known as ‘calliper’ to match on.26 Specifically, this means that within the defined propensity range, the NN is used as a matching partner. In order to choose an appropriate radius, Rosenbaum and Rubin27 propose that an appropriate calliper width is a quarter of the SD of the estimated propensity score.

Unlike the NN and radius matching approaches which only use selected observations of the control group, kernel matching uses all of the individuals in the comparison group to construct the counterfactual outcome.26 Kernel matching is defined as ‘non-parametric matching estimators that compare the outcome of each treated person to a weighted average of the outcomes of all the untreated persons, with the highest weight being placed on those with scores closest to the treated individual’ (p. 27).28 Due to the increased information used, this approach results in a lower variance, although it also has potentially poorer matches.26

NHIS enrolment was defined as holding a valid NHIS insurance card (treatment group). The control group consisted of all other older adults. The health insurance membership ID card provides evidence of enrolment and entitles members to free use of care at accredited facilities. Overall, the effect of NHIS enrolment was measured using two outcome variables: (1) consultation with health practitioner during the last 2 weeks (outpatient care) and (2) hospitalisation in the last 12 months (inpatient care). Both outcome variables are dichotomous.

The propensity score for this analysis was estimated using a logit function. The predictor variables entered into the model were based on the considerations discussed in the literature, which identified correlates of NHIS enrolment and increased utilisation of care. The availability and accessibility of healthcare services influence the usage of healthcare and insurance enrolment.5 29 Therefore, travel time to the nearest healthcare facility and hospital (reported by the community chiefs) as well as place of residence, region and road access (existence of a motorable road that passes through the community) was included to estimate the propensity score. Sensitivity analysis used the availability of a healthcare facility in the community instead of travel time to the nearest healthcare clinic and hospital which confirmed that availability of a healthcare facility in the community has a positive effect on insurance enrolment.

Further, previous research found that sex, education, religion, ethnicity, marital status, living standard and age were found to determine uptake of the NHIS and significantly influence the utilisation of healthcare.12 29–31 These variables were therefore also included.

In order to fully understand the benefit of NHIS coverage, the difference in healthcare usage between poor and non-poor older adults was measured. In this paper, ‘poor’ corresponds to a household lying below the poverty line of 1314.00 Ghana cedis (around US$300) per adult per year. This method used equivalent household expenditure taking into account that the consumption requirements of children are less than those of adults.32

The data were randomly sorted to remove potential bias in the matching caused by the sorting of the data. Bootstrapping was used to estimate SEs, with 1000 replications run for each matching approach. All analyses were conducted in Stata using the psmatch2 command.

Does insurance enrolment increase healthcare utilisation?

Due to potential selection bias, it is open to question as to whether the differences described in tables 1 and 2 reflect the true effect of NHIS enrolment. Table 3 shows the results for the three different methods of PSM that were estimated. Generally, the findings support the hypothesis that NHIS membership increases the utilisation of healthcare, supporting the unconditional differences reported above. When conditioning on observable characteristics for all matching approaches, a positive effect of NHIS enrolment on the utilisation of care was found. In all instances, significant differences were found in inpatient and outpatient care between insured and non-insured. The difference in outpatient care use between NHIS members and non-members was around 9%, with NHIS members tending to also use more inpatient care compared with non-members. Thirteen per cent of older adults enrolled in the NHIS used inpatient care in the previous 12 months compared with only 7% who were not insured.

Additional investigations disaggregated the analysis into different age groups in order to see the effect of the premium exemption for older adults aged 70 plus. The results supported the above analysis and indicated that insured individuals aged 70 or over were more likely to use both inpatient and outpatient care.

Has Healthcare utilisation increased equitably?

Conditional on all observable characteristics, a positive effect of NHIS coverage on the utilisation of healthcare among poor and non-poor older adults was found (see table 4). Keeping all observable factors constant, the elderly poor enrolled in the NHIS were around 5% more likely to use outpatient healthcare services compared with the unenrolled. The enrolled non-poor older adults, however, were up to 11% more likely to consult a health practitioner compared with their non-enrolled counterparts. Further, the percentage of the enrolled poor who used outpatient care was the similar to the percentage of uninsured non-poor.

The benefit of enrolment for inpatient care services over the last 12 months was similar for both poor and non-poor older adults, at around 7%. However, only 11% of insured poor older adults stayed in a hospital overnight compared with 15% of insured non-poor elderly.

Finally, the differences in impact of NHIS coverage between males and females were assessed (see table 5). Depending on the matching technique, enrolled males were between 6% and 9% more likely to use outpatient care services compared with their unenrolled counterparts and around 7% more likely to use inpatient care service. For women, the likelihood of using outpatient care services increased by over 10% when they were enrolled but by only around 5% when measuring whether they have stayed overnight at a hospital.

Sensitivity analysis

Sensitivity analysis was used to assess the above PSM analysis. The quality of the matching was assessed by evaluating whether it was possible to balance the distribution of observed characteristics across the treated and untreated, conducted by comparing the situation before and after the matching and looking for remaining differences.26 A two-sample t-test showed that after the matching, no significant differences between the control and comparison group exist indicating successful matching.

The main limitation of PSM is that only observed confounders can be included when estimating the propensity score. Unobserved omitted variables bias cannot be controlled for using this method. This study used Mantel-Haenszel test statistics to determine how much positive or negative hidden bias is necessary for an unobservable factor to affect the treatment effect significantly. The results illustrate that, even when an unmeasured variable increases or decreases the odds of NHIS enrolment by as much as 50%, the treatment effect on consulting a health practitioner or visiting an inpatient facility still is significant, validating the results shown above.