An Analysis of Health Financial Protection in Ghana

Ghana introduced a state sponsored health insurance in 2005. The objectives of risk sharing through a prepaid mechanism was integral to the design, aimed at addressing the inequities in nancial protection. Progress, however, has been predictably slow and complicated. The Scheme coexists alongside out-of-pocket spending, resulting in the persistence of catastrophic health expenditures and impoverishment. This study adds to the comparative literature on health nancial protection in Ghana. This study analysed data from Round 7 of the Ghana Living Standards Survey using ADePT. Out-of-pocket spending was used as the indicator for living standards. We estimated different headcount thresholds and concentration index for catastrophic health expenditure. Consumption was estimated to determine the poverty headcount and poverty gap. The distributive effect of health care expenditures on the different income quintiles was estimated to determine progressivity of health nancing sources.

results in accordance with underlying principles and assumptions of analysis. Issues such as programming errors associated with dissimilarities in computational methods are drastically reduced. This reduces variability thereby improving validity and authenticity of results. Module 2 of ADePT, the health nancing and nancial protection -inclusive of catastrophic payments and impoverishing effects, and; progressivity and redistributive effect of health nancing -was used for this analysis [22,23].

Analysis
The GLSS 7 dataset was rst prepared in Stata before uploading onto ADePT. The variables and analysis model is shown in Table 1. OOP spending was used as proxy for living standards. The analysis calculated the incidence of catastrophic OOP expenditure and impoverishment and its intensity. The progressivity of the sources of nancing, that is OOP payments and social health insurance contributions, was also computed. In determining the decomposition effects of progressivity, horizontal equalty and reranking, the distributive effect of health nancing, that is the effects of health care nancing on income inequality, is also estimated [22,23]. Note: * = a more advance type of analysis. Prepayments include private insurance premiums, social insurance contributions, and taxes.
Catastrophic health spending OOP payments are direct household cash expenditures at the point of access that when they exceed the 15-20% consumption threshold is deemed catastrophic [15,16,22]. OOPs are not necessarily always catastrophic [16,24]. This threshold had a deterministic effect on the proportion of the population falling into this classi cation, also termed as the headcount. ADePT enabled head count simulations at different CHE thresholds for comparability and intensity (or overshoot). Catastrophic payments between the poor and rich households was differentiated by estimating the concentration index, where a negative index denotes higher CHEs among the poor and vice versa. By calculating the weighted head count index, we determined the degree of overshoot in relation to the place of households in income distribution. A negative concentration index being indicative of a higher weighted head count, vital in tracking CHEs among the poor [22].

Impoverishment and health spending
Health expenditures can be the difference between households either being well-off or poor. However, OOP spending as an indicator of household living standards totally neglects the opportunity cost of the forgone basic household necessities [3,15,22]. We aggregated consumption, inclusive and exclusive of OOP payments, to determine the alternate and actual scenarios. This showed the extent to which OOP payments contributed to the poverty headcount, including comparison of the poverty headcount with the two different consumption types. OOP payments intensify poverty among already poor households. To get the aggregated shortfall in consumption from the poverty line, we calculated the poverty gap attributable to OOP payments [22].
1. The Kakwani index; that is the progressivity of health care payments.
2. Health care payments as a proportion of income -where a higher percentage suggests a larger impact on income inequality.
3. The degree of horizontal inequality in; tax, social health insurance contributions, and OOP payments; as in comparable health care payments for households in the same rankings. Estimating horizontal inequality involved determining the post-payment income inequality for similarly ranked groups -a positive number -represented by HI in the equation below. The higher the degree of post-payment income inequality within groups, the higher the degree of horizontal inequity. The share of the population and post-payment income were weighted to determine the post-payment income inequality for each group. By either decreasing the equalizing or magnifying the disequalizing effects, horizontal inequalty diminishes the redistributive effect of health care payments. 4. The degree of reranking shows the different uses to which households distribute their residual income after health care payments. Stepwise, households within each group were rst ranked by their prepayment income, followed by their post-payment income. The difference between the post-payment Gini coe cient and the post-payment concentration index afterwards showed the reranking [22,23] as shown in the below equation:

RE = [g/(1 -g)] Kakwani -HI -Reranking
Where: RE is the difference between the two Gini coe cients, prepayment and post-payment incomes; g represents the proportion of health spending of prepayment income; Kakwani represents equal household payments of equal ability in the Kakwani progressivity index; HI represents the horizontal inequalty index, and; Reranking corresponds to the reranking index [23].

Households spending on health
The dataset analysed from the GLSS covered health care expenditure for 14,009 households. Table 1 presents the summary expenditures by socio-demographic characteristics. Households in urban areas, for both gross and net per capita consumption, spent more than twice on health than rural households. The non-poor households similarly had a higher gross per capita consumption than the poor. Signi cant inequalities exists between the richer and poorer households as for example observed in total health payments between the Upper West, 6.2% on average, and Greater Accra, 33.4%, regions. Catastrophic health spending Table 2 presents measures of the incidence and intensity of catastrophic health payments. As the threshold increased from 5-25% of total expenditure, the estimate of the incidence of catastrophic payments (H) dropped from 24.2-6.2%. Likewise the mean overshoot; from 4.2-1.6%.
The standard errors were small relative to the point estimates considering the large sample size. Unlike the headcount and the overshoot, the mean overshoot among those exceeding the threshold (MPO) need not decrease as the threshold is increased. Households who spent more than 5% of total expenditure on health on average spent 22% (5% + 17.2%). Those that spent more than 25% of the household budget on health on average spent 51.2%. For the different thresholds, both the head count and the overshoot were higher, when catastrophic payments are de ned with respect to health payment relative to non-food expenditure. The total expenditure and non-food expenditure is illustrated in Fig. 1. For any budget share, the OOP or non-food expenditure curve is always to the right of the total expenditure curve. For more than 15% of households, health spending was at least one-fth (1/5) of non-food expenditure, but health spending was about a quarter of total expenditure for 3% of households.
The concentration indices and rank-weighted headcount and overshoot measures are shown in Table 3. The distribution of catastrophic payments depends on whether health payments are expressed as a share of total expenditure or of non-food expenditure. In the former case, catastrophic payments decreased with total expenditure. It showed OOP health payments budget share reduced with total household resources.
As a result, the rank-weighted headcount and overshoot were bigger than the unweighted indices shown in Table 2. When the health payments were evaluated relative to non-food expenditure, it produced negative concentration indices, indicating households with low non-food expenditures were more likely to incur catastrophic health payments de ned in this way. Consequently, the weighted indices are larger than the unweighted indices as shown in Table 4.  Source: Authors.
Households in the lowest quintile allocated a higher share, 12.5%, of their total expenditures to health. About three times the average spent by the top four quintiles (Fig. 2). The results further showed a 15% incidence of OOP health payments. The effect of OOP payments on poverty using the Pen's parade of households based on their consumption expenditure distribution illustrated in Fig. 3 Progressivity and redistributive effect The per capita health care nancing by households is shown in Table 5. The average per capita consumption by the poorest quintile, GH 65.4, was signi cantly lower than for the richest quintile, GH 3,725.70. Expenditure on insurance, including housing and health, was mainly incurred by the richest. Poorer households in the lowest quintile incurred the least total health payments, GH 7.5 on average. Source: Authors.
Progressivity of households' health care nancing is presented in Table 6. It shows the average consumption and nancing shares by quintiles with households ranked from lowest to highest of gross consumption. The information relating to consumption explains the inequality that exists between households in Ghana. The poorest quintile had the least share of consumption compared with the richest; from GH 1.3% to GH 74.6%. The richest households paid a greater share, 46.8%, in insurance premiums, than the poorest households. Source: Authors. Table 7 shows households' nancing expressed as a share of their gross consumption. Tables 6 and 7   Page 10/15 The lowest quintile accounted for, on average, 1.3% of total health consumption whereas the highest quintile consumed 74.6%. Expenditure on insurance, including housing, health, among others, was mostly incurred by the richest households. The rst three quintiles combined contributed just 9.3% on average. Therefore, the nancing share rises the higher the quintile rank. Total expenditure on health represented 1.97% of the households budget, on average. The total redistributive effect of the expenditure on insurance (-0.0002), total expenditure on health (-0.0069) and total payments (-0.0071) showed an increase in households income inequality.

Discussion
This analysis of households health care nancing in Ghana showed signi cant ndings that merit further attention. The nding of 15% incidence of OOP payments is within the recommended 15-20% threshold, down from the pre-insurance level of 64% in 2002 [19,25]. This is however evidently higher than the global average of 2.3% [17]. Marked disparities still persist in incidence of OOP payments, ten times higher among uninsured households [13]. The 6.2% incidence of catastrophic payments is similarly broadly within the 1.3%-8% previously reported for Ghana [8,10,13,21,26,27]. Although an eight-fold reduction in incidence of CHEs has been recorded since 1995 [14], user fees remains the primary means by which poor Ghanaian households pay for health care, attributed to the low depth of prepayment coverage among this quintile [9,13,14,28]. Prepayment mechanisms also do not provide full immunity against OOP health expenditures [1,8,17,29]. Therefore the exclusion of specialised services from the Ghana Scheme's bene ts package entrenches formal user fees, thus intensifying incidence of CHEs [9,14,28]. A more sustainable long-term policy goal, backed by related strategies, have to target reducing OOP spending as share of total health care nancing among households in the bottom quintiles to lower than the15% threshold. This may involve increasing the tax component of prepayment of the Scheme as a share of total expenditure on health. In general, this is of interest to developing countries implementing universal prepayment mechanisms [17,26].
We further observed though households in the upper quintile paid more in insurance contributions as a share of total payments for health, the poorest paid highest. Premium contributions has previously been reported as a deterrent to insurance enrolment in Ghana [10,11,13,30,31]. Our negative values for the Kakwani index rea rms ndings by Akazili et al, Mills et al and Amporfu suggesting the better-off households do not contribute more to the nancing of health care than the poor and that health care nancing in Ghana is generally pro-rich [9,10,21]. The issue of regressive premium contributions pertains to the at rates charged informal sector enrolees due to an inability of Scheme agents to actuarially determine the contributions for this group. It is also the case of the weak implementation of a free membership policy for the poor [8][9][10]28].
This reiterates the importance of the need for periodic reviews of the price of insurance in poor resource settings and the introduction of caps for the different groups within the informal sector where necessary. Contributions to social health insurance (SHI) must therefore be based on ability to pay [8,16,28,29]. Other complementary policy options such as mixed targeting approaches improves the technical e ciency in identi cation, enrolment and extension of SHI coverage to vulnerable groups [8-10, 16, 29, 32]. Using poverty prevalence maps and pro les, a more ambitious policy and political dilemma will be to go universal altogether [16,17,26,28,29]. The rst two options promote nancial protection [9] and all three can potentially bridge inequity gaps in CHEs thereby offering more secured nancial protection [10].
Households in the lower end and some in the middle half of the distribution were brought below the poverty line by health payments -that is 6.2% headcount -and that the poorest households had the highest share of expenditure allocation for health. This is consistent with the 3-5% and 9.4% headcount impoverishment arising from CHEs recently reported by Aryetey et al and Akazili et al [11,12]. This is suggestive of a strong correlation between OOPs, CHEs and impoverishment in Ghana, where health care expenditures trap and push vulnerable households deeper into poverty [9,27,29]. The adverse effects of prepayment policy reforms in poor resource settings include correlative effects with poverty and inverse care law where those who need health care the most access it the least [15,19]. Such reforms mainly involve the scrapping of OOP payments for consultation which is usually only a small fraction of the many other direct and indirect costs incurred [1,15,33]. Poor households therefore resort to coping mechanisms [3,8,11,15]. For instance, Levie and Xu found that for Ghanaian households in the highest category of inpatient spending, 40% resorted to borrowing and sale of assets to meet health care expenditures; the highest in a comparative study of eleven countries [3]. Ominously, some poor households simply forgo health care even though they need it to avoid the impoverishing costs [1,15,17,25].
For nancial protection to be truly universal and effective, it must go beyond the direct costs to include indirect costs and the elimination of other nancial barriers that shape decision making on health [34][35][36][37]38]. It is imperative that nancial protection policies nd support in transformative change in resource e ciency and expansionary budget policies to expand the domestic scal space for health [3,16,28,33].
Ghana and other developing countries can for instance learn from Indonesia, Sri Lanka and Malaysia where caps on user fees at public facilities and targeted exemptions for the poor have been effective in reducing impoverishment due to health spending [15]. A corresponding related strategy will involve introducing reimbursement policies for both direct and indirect household health expenditures for vulnerable groups [34].
The results also revealed deepening inequalities between households as a result of CHEs. Zhang et al reported a 5.4 percentage points' decrease in inequality in CHE in Ghana since 1995 [14]. While this represents creditable progress, it nevertheless should not distract from the still excessive socio-economic disparities between households attributable to health expenditures [10,14]. Such inequalities intensify the recurrence of the incurrence of CHEs any time poor households come into contact with the health system [8,9,11,20]. This highlights coverage and enrolment as poor indicators of nancial protection [17] and an inability of the current systems of LMICs to adequately track and monitor the shifting dynamics in OOP spending and CHEs in general [37].
Remedial measures, therefore, should use a mix of policy and programmatic tools. Thailand's universal coverage programme used this approach to successfully reduce inequalities arising from health care costs [8,17,24]. The overarching strategy to tackling household inequalities due to health expenditures will require a broader human capital development approach that includes simultaneous strengthening of health and social services, and environments [12,15,39] to build cross-cutting bi-directional synergies that link health and other development policies [6,8,15]. In this regard, enrolment of the poor in social security and poverty alleviation schemes [17,24] and allocation of resources for health based on burden and need rather than cost-effectiveness and population, and facility density are more sustainable medium-to long-term policy strategies [33].

Limitations
The ndings of this study should be discussed in respect of the following limitations. The data used for this analysis is from a national survey that required respondents recall information about households expenditures and consumption on health and the established period of recall can impact the information on the frequency and magnitude of health care payments thus affecting the actual occurrence of OOP spending and CHEs [37,39]. The recall is also speci c to a period of time hence does not allow for comparison of multiple health expenditures within the same households over a period of time -that is comparative analysis of multiple health expenditures for the same households over a period of time [24]. The type and number of variables included in the survey and the level of disaggregation impact the results, where higher levels of disaggregation tends to elevate estimates of health care expenditure. Hence, the number of expenditure categories and the period of recall used by the survey have a deterministic effect on the reliability and validity of households report on health spending [37]. Other supply-and demandside factors which determine availability and price of health care for which household expenditure data was not collected could have underestimated the composite household spending on health [16]. Importantly also, population level surveys on household spending like the GLSS tend to miss expenditures made to informal providers, including other associated indirect and direct costs in the course of health care seeking. This can underestimate OOP spending and CHEs [9,16]. The probability that households may become poorer as a result of the coping mechanisms they resort to and which may yet still not be captured as part of their spending on health as such expenditures are below the established threshold weakens the predictive power of OOP payments and CHEs as a measure of nancial protection [12,24,25]. Obviously, not every household member that needs health care is able to access it due to nancial limitations; these group assessments on CHEs and poverty tend to miss them [16,20,24,25]. Further, there is no golden standard for setting thresholds for nancial catastrophe. This is usually subject to interpretations, adopted de nition of OOP spending and CHEs used in the study and study objectives as is ours [20,40]. Finally, delity demands OOP payments and CHEs are analysed alongside utilisation data for a much clearer and complete representation [24].
These limitations notwithstanding, the methodologies used in this analysis are statistically proven, and used previously in a variety of studies in different contexts [12,16,19,20,24,25,37,39,40]. This study have assessed nancial catastrophe and impoverishment at different thresholds which presents a broader picture on nancial protection than any one threshold would have. In the absence of general guiding principles of calculating thresholds, based on the expenditure levels, this study nevertheless calculated multiple levels of thresholds for the different households thereby accounting for all possible scenarios [20].

Conclusion
Re ective pauses in the implementation of population level health policies and programmes such as for universal nancial protection enable policy makers, programmers and scholarship alike to interrogate the evidence and evaluate the experiences and lessons learned. In Ghana's case, the results from this study, though mixed, are informative its health nancing policies are failing to provide adequate nancial protection against health expenditures for the poor. They further highlight the multiple technical challenges related to the operationalisation and rationalisation of policies and interventions on nancial protection and risk equalization in poor resource settings in general. It is however pertinent that discussions on nancial protection seek to magnify both the limitations and narrative of opportunities. This will require continuous understanding of the changing systemic and socio-demographic, and -economic factors that are amenable to nancial catastrophe and push households into poverty.
Improving the ecosystem of the systemic and social determinants of health nancing is central to any efforts targeted to nancial protection in health. Following, Ghana can leverage its Scheme to strengthen the implementation of its current health nancing policies and create opportunities for investments in pro-poor interventions and actions. This, when complemented by social protection strategies can reinforce and promote cross-sectoral policy and programmatic coherence to improve the ability and exibility of poor households to cope with the uncertainties of health expenditures.

Funding
This study did not receive funding.
Availability of data and materials