Incidence, socio-economic inequalities and determinants of catastrophic health expenditure and impoverishment for diabetes care in South Africa: a study at two public hospitals in Tshwane

Since the World Health Assembly of 2005, many countries have committed to ensuring that their health systems offer financial protection from risks of catastrophic expenditure or impoverishment due to healthcare payments [1]. More recently, through target 8 of Sustainable Development Goal 3 (SGD 3), many countries have renewed their commitment to achieving universal health coverage by 2030 as a means to prevent financial hardship and to assure equitable healthcare outcomes. Access to healthcare is considered a basic human right. However, in many African countries, out-of-pocket (OOP) payments remain a hindrance to access healthcare and may result in financial hardships [2]. Globally, up to 150 million people suffer catastrophic healthcare expenditure yearly whilst 100 million are impoverished as a result of OOP healthcare payments [1]. It is reported that catastrophic health expenditure and impoverishment is high in countries in which OOP payments for healthcare are above 20% of total health expenditure (THE) [2]. The World Health Organisation (WHO) African region expenditure atlas shows that in 2012, 37 out of 47 countries had OOP health expenditure as a percentage of THE above 20% [2]. In South Africa, OOP health expenditure as a percent of THE was reported to be 7% in 2014 [2] which is lower than over 90% of countries within the European region [3].

Catastrophic health expenditure occurs when out of pocket health payments surpass a predefined threshold of a household resource such as household expenditure, resulting in a reduction of the household’s ability to spend on other essential items and may push the household into poverty [4, 5]. Previous studies on the incidence of catastrophic healthcare expenditure in South Africa – which have not focused on diabetes healthcare costs - report an incidence of catastrophic health expenditure ranging from 5 to 66% [6‐13]. The differences in the estimated incidence is a result of differences in the study design, the data used, the different populations studied and the disease under analysis. It is reported that households with individuals with diabetes are at a high risk of experiencing financial difficulties [14] and more likely to experience catastrophic health expenditure [15]. Frequent visits to healthcare facilities for glucose monitoring, foot care and associated complications increase the cost of healthcare and risk being confronted with catastrophic healthcare expenditure.

The International diabetes Federation projections indicate that within the African region, diabetes health expenditure will double to 12.3 billion international dollars (ID) by 2045 [16]. In 2017 South Africa was reported to spend ID 1884 per person with diabetes [16]. This makes South Africa the country with the second largest mean healthcare expenditure on diabetes in the African region and this burden is projected to grow [16]. Despite this, there is a paucity of research on the economic impact of diabetes for patients and households within South Africa and Africa in general [17], in particular studies that estimate the incidence of catastrophic health expenditure due to diabetes.

To the best of our knowledge, only a single study has investigated the incidence of catastrophic health expenditure among diabetic patients in an African country setting. Okoronkwo et al. find an incidence of catastrophic health expenditure among diabetic patients attending a tertiary healthcare institution in Nigeria of 45% when using a catastrophic expenditure threshold of 30% [18]. Although the authors find that all socio-economic groups suffered catastrophic healthcare expenditure, the lowest socio-economic group had the highest incidence, indicating an unequal distribution of catastrophic health expenditure across socio-economic groups [18]. For diabetic patients this is very disastrous as the poor may be forced to forego other vital needs such as dietary diversity in order to attain health services. To date, in South Africa, studies that have assessed the costs associated with diabetes have done so from a health system, societal or government perspective [19‐22]. No diabetes patient costing study has been conducted and no study has assessed the incidence and socio-economic inequalities of catastrophic health expenditure for diabetic patients seeking healthcare in public hospitals that provide subsidised healthcare.

Subsidised healthcare in South African public hospitals is provided according to a uniform payment fee schedule (UPFS), which groups patients into full paying patients and subsidised patients [23, 24]. Patients are classified as full paying if they are being treated within a public hospital by a private practitioner, are externally funded patients (for example funded by the road accident fund or medical scheme), or are non-South African citizens. Subsidised patients may either be fully or partially subsidised, depending on their ability to pay for healthcare services [23]. Patients are classified as fully subsidised if they provide proof of being social pensioners or formally unemployed. Partially subsidised patients pay according to their income level (commonly referred to as the sliding scale means test) [23]. Most healthcare charges are based on a grouping of services, meaning fees payable by patients include all costs such as consumables, overhead costs and salaries rather than itemised billing for each service [23]. The role of such OOP payments in healthcare financing, even in settings with modest healthcare bills, is of great concern given that this is a very regressive healthcare financing mechanism [25]. This creates the need to investigate financial protection measures in South African public hospitals.

Our study aims to (1) assess the health expenditure patterns of diabetic patients seeking healthcare in public hospitals that provide subsidised healthcare; (2) determine the incidence of poverty, catastrophic healthcare expenditure and subsequent impoverishment due to diabetes; (3) assess the socio economic inequalities in catastrophic health expenditure and impoverishment due to diabetes; and (4) explore the determinants of catastrophic health expenditure and impoverishment due to diabetes. A study of this nature is important for various reasons. The prevalence of diabetes is increasing rapidly and has a significant economic impact on individuals and households. The incidence of catastrophic health expenditure and impoverishment are important indicators of the extent of financial risk protection offered by the health system, which is particularly important within the context of universal health coverage.

The characteristics of our study sample are shown in Table 1. The mean age for our sample was 52 years. The majority of the respondents were female (61%). Our sample was predominantly African (76%). About 59% of respondents reported being single, 86% reported having children, 66% reported having secondary education, and 63% were unemployed. The majority of respondents had a household size of 1–4 persons (64%).

Table 2 demonstrates that the mean total household expenditure for our sample was R4213 (1 US$ = 13.30 Rands) (median value R2810). Household food expenditure averaged R1687 (median R1500). Subsistence expenditure averaged R1003 (median R1016) and capacity to pay averaged R3325 (median value R1730).

Our analysis suggests that the mean OOP health expenditure per out-patient visit related to diabetes care was R53 (median value R40). The mean round trip transport cost was R79 (median R40). When we combine transport costs and health costs, we find an average of R132. The OOP health expenditure was statistically significantly different across the wealth index quintiles. Households in quintile 5 pay more for health services relative to other quintiles (see Fig. 1). Other direct non-medical costs incurred by patients included expenses on food whilst waiting at the hospital. The mean food costs was R30 (median R20).

Our estimation of productivity losses is restricted to those who reported being employed and reported their income (n = 153). Our findings show that out of a sample of 100 patients who reported missing work over the past 30 days due to diabetes, the average time lost was 10.6 h. No respondent reported taking more than 10 days off work due to diabetes related illness. The average time for a hospital visit was 7 h. The mean estimated indirect costs for missing work over the past 30 days and for a hospital visit was R 429 and R 277 respectively. When we combined the average time for a hospital visit and the time taken off work due to diabetes, the total time lost was 14.5 h. The mean estimated indirect costs for both time spent seeking healthcare and time taken off work was R 553 (see Table 2).

In estimating catastrophic health expenditure and impoverishment due to healthcare costs, we employed both the WHO standard method and the Ataguba method (see methods section). As mentioned earlier, we considered catastrophic expenditure due to direct OOP payments for diabetes care (approach 1) and also catastrophic expenditure due to both direct OOP payments for diabetes care plus direct non-medical costs of transport (approach 2). The sensitivity of the analysis to different thresholds was tested and results presented in Table 3. Due to the differences in the computational steps used in the construction of statistical variables (see methods), the sample sizes differ for each analytical method (WHO method and the Ataguba method). In each method observations with missing data in variables used for the construction of catastrophic health expenditure or impoverishment were excluded from the analysis.

Results from the WHO method show that approximately 6% of the study sample was poor. Catastrophic health expenditure for diabetes care was measured using four thresholds (10, 20, 30 and 40%) and decreased as we moved from the lower to higher thresholds. As expected, when we included transport costs in estimating the ratio of healthcare costs to capacity to pay, we find that catastrophic health expenditure measured at all four thresholds was much higher than when we exclude transport costs. This finding signifies the importance of transport costs. Using approach 1 (approach 2), catastrophic health expenditure for diabetes care varies from 2 to 9% (6 to 25%). Amongst those who were not poor and when we exclude transport costs, none of our study participants are impoverished due to diabetes care but when we include transport costs, 2% become impoverished.

For the Ataguba method, we make use of an initial threshold of 10% to estimate catastrophic health expenditure for diabetes care. We present results using a parameter of aversion to inequality of 0.8 (a varying threshold) and of 1 (a constant threshold of 10%). As shown in Table 3, the increase of this parameter from 0.8 to 1 decreases the head count from 4 to 2% (approach 1) and 13 to 10% (approach 2). Using the South African poverty line of R758, we find that the poverty head count increased from 21 to 23% (approach 1) and 21 to 25% (approach 2). This translates into a 2 and 4% rise in poverty using approach 1 and approach 2, respectively.

The proportion of diabetic patients who were poor, was highest in wealth quintile 1. The proportion of catastrophic health expenditure for diabetes care was highest in the first wealth quintile when measured using both the WHO standard method and the Ataguba method. This shows that diabetic patients who incur catastrophic health expenditure are in fact the poor. From the Ataguba method we find that higher levels of catastrophic expenditure exist within quintile 1 when measured using a variable threshold (y = 0.8) compared to when measured using a fixed threshold (y = 1). For example, in approach 2 we find an incidence of catastrophic expenditure of 20.27% within quintile 1 when measured using a variable threshold and we find an incidence of catastrophic expenditure of 16.44% within quintile 1 when measured using a fixed threshold. Using the Ataguba method, the highest increases in the poverty head count ratio is recorded for diabetic patients in quintile 1. For example, we see that OOP health expenditure for diabetic care leads to a 6.85% increase in the poverty head count ratio when using approach 1, i.e. without accounting for transportation costs. The figure for approach 2 is higher at 9.59% (see Additional file 2).

Table 4 shows the concentration indices for catastrophic health expenditure and impoverishment related to diabetes care. All statistically significant concentration indices are negative, indicating that catastrophic health expenditure for diabetes care is more concentrated amongst the poor diabetic patients. In the WHO standard method, the value of the catastrophic health expenditure index decreases as we move to higher thresholds, but also declines in statistical significance. The Ataguba method (ƴ = 0.8), which uses a varying threshold, shows higher inequalities in catastrophic health expenditure when compared to the one which uses a constant threshold of 10% (ƴ = 1). However, the CI for the Ataguba method (ƴ = 1) are statistically insignificant. All the values for the concentration for impoverishment due to diabetes were negative, indicating that the proportion of impoverishment is concentrated amongst the poor. Because none of our study participants are impoverished due to diabetes care when estimated using the WHO method for approach one, we do not estimate this inequality.

This section analyses the socio-demographic factors associated with catastrophic health expenditure and impoverishment for diabetes care based on results from multivariate logistic regressions. For brevity and due to the loss of statistical power, we present results from the multivariate logistic regression analysis for four out of the 12 estimates of catastrophic healthcare expenditure, represented here by binary variables. Table 5 shows the results for the WHO standard method using a threshold set at 10% for both approach 1 (medical costs only) and approach 2 (medical and transportation costs). For the Ataguba method, we present results when catastrophic health expenditure for diabetes care is measured using approach 2 (health costs plus transport costs) when the parameter of aversion to inequality is ƴ = 0.8 and ƴ = 1. Given the low estimates of catastrophic health expenditure for Ataguba’s method when applied using approach 1, regression models were estimated only for approach 2.

Using the WHO method, for approach 1 we see that the factors that increase the odds of catastrophic health expenditure among diabetic patients are being female (Odds ratio [OR] 3.72; Standard error [SE] 2.06), not having children (OR = 4.01; SE = 2.71) and a household size of five or more people (OR = 2.74; SE = 1.32). Patients who were non-African had reduced odds of experiencing catastrophic health expenditure due to diabetes care (OR = 0.16; SE = 0.30). Applying the WHO method to approach 2, we find that female diabetic patient had increased odds of experiencing catastrophic health expenditure (OR = 1.73; SE = 0.52). Falling within the third wealth quintile (OR = 0.49; SE = 0.20), fourth wealth quintile (OR = 0.31; SE = 0.15) and fifth wealth quintile (OR = 0.30; SE = 0.17) were associated with reduced odds of experiencing catastrophic health expenditure. In both the regressions based on the estimates of catastrophic health expenditure calculated with Ataguba’s method, diabetic patients who reported not having any children had increased odds of experiencing catastrophic health expenditure for diabetes care (OR = 3.36; SE = 1.83 when ƴ = 0.8 and OR = 3.54; SE = 2.17 when ƴ = 1).

Due to the low incidence of impoverishment when calculated using the WHO method and the loss of statistical power, we present results for the multivariate logistic regressions of impoverishment calculated using the Ataguba method. Using approach 1 (excluding transport costs) patients who were female (OR = 12.83; SE = 15.94) and had no children (OR = 20.38; SE = 34.50) had increased odds of experiencing impoverishment due to diabetes care. Patients who were single (OR = 0.01; SE = 0.01), within the second (OR = 0.07; SE = 0.11) and third (OR = 0.04; SE = 0.06) wealth quintiles had reduced odds of experiencing impoverishment due to diabetes healthcare costs. Using approach 2 (health costs plus transport costs), diabetic patients who reported being single (OR = 0.30; SE = 0.19) had lower odds of impoverishment due to diabetes healthcare.

This study illustrates the burden of diabetes healthcare amongst patients attending diabetes clinics at subsidised tertiary public healthcare hospitals in Tshwane, South Africa. To the best of our knowledge it is the first study that assesses the incidence of catastrophic health expenditure amongst diabetic patients and assesses inequalities in catastrophic health expenditure. Our study shows that although considerable progress has been made in health service delivery through for example the provision of free healthcare when accessing primary healthcare [23], failures still exist in making healthcare affordable for chronic patients within public hospitals. Our findings show that despite the applied Uniform Payment Fee Schedule (UPFS) in public sector hospitals, there is a high incidence of catastrophic healthcare expenditure amongst diabetic patients. To the best of our knowledge there are no previous studies that estimated the incidence of catastrophic healthcare expenditure amongst diabetic patients in South African public hospitals, we are therefore not able to determine if these inequalities are changing.

We find that the existing UPFS has been ineffective in protecting diabetic patients in low socio-economic groups from financial hardship resulting from diabetic healthcare costs. This finding is consistent with a study that examined health expenditure patterns in rural South Africa and also found high healthcare costs burdens within poorer wealth quintiles [13]. This is of concern given that diabetic patients make multiple visits to healthcare centres [50] and have to contend with diabetes related healthcare costs throughout their life. Although we only focus on indirect costs due to time lost seeking care and time taken off work due to diabetes our finding that indirect costs of diabetes are much higher than the direct costs is consistent with other studies that used national diabetes prevalence estimates to measure indirect costs due to disability, premature mortality and loss of income [17]. Our results also show that transport costs contribute over 50% of the direct diabetes healthcare costs, which results in much greater levels of catastrophic health expenditure and impoverishment when these costs are considered healthcare costs. Transport costs have also been found to be high in other healthcare studies in South Africa [6, 7, 13]. For example, using data from the Agincourt Health and Demographic Surveillance site in South Africa, Goudge et al. finds that transport costs contribute 42% to health expenditure [13]. These high costs point to the urgent need for interventionist programmes to improve access to healthcare for chronic patients who make multiple visits to healthcare centres via provision of free or subsidised patient transport services.

Consistent with other studies we find that the incidence of catastrophic health expenditure is sensitive to the method used [29, 42, 49]. Given the arbitrary nature of threshold selection, the high poverty levels amongst Africans in Tshwane [26], the high transport costs and the high reliance on social grants (30.3% of households) [51], our discussion focuses on the catastrophic health expenditure measured using two methods: (1) healthcare costs plus transport costs at a 10% threshold of capacity to pay; (2) healthcare costs plus transport costs at a variable threshold of total household expenditure. We argue that in this setting, expenditure as low as 10% of capacity to pay or any expenditure at all is catastrophic.

The incidence of catastrophic health expenditure due to diabetes has previously been investigated in other countries [52, 53]. Using the WHO method and a threshold of 10%, the incidence of catastrophic health expenditure due to diabetes care was 25%. On the other hand, the Ataguba method yields a lower incidence of catastrophic health expenditure due to diabetes of 13%. This finding is similar to other studies that find that the incidence of catastrophic health expenditure is higher when measured as a proportion of capacity to pay (non-food expenditure) compared to when measured as a proportion of total household expenditure [42, 54, 55]. The capacity to pay method displays a greater ethical concern for the poor by recognising the higher spending on essential items by poor households when compared to richer households [5, 42].

Consistent with results from a systematic review by Njagi et al., an assessment of the patterns and socioeconomic inequalities in catastrophic health expenditure and impoverishment due to diabetes in our study, showed that the incidence is highest within the lowest socio-economic groups [49]. Using a cross sectional sample of 308 type 2 diabetic patients attending a tertiary healthcare institution in Nigeria, Okoronkwo et al. also find that the incidence of catastrophic health expenditure is highest within the lowest socio-economic group [18]. Although we focused on diabetes only, our finding is similar to other studies that find negative CIs for catastrophic health expenditure [56‐58] and impoverishment [57, 58]. This means that catastrophic expenditure due to diabetes healthcare is concentrated amongst the poor as indicated by the negative CI and odds ratios for wealth. To the best of our knowledge there is no study that has investigated inequalities in catastrophic health expenditure due to diabetes in South African public hospitals. We are thus unable to make comparisons with earlier inequality estimates. The results from our study also showed that depending on the method used, up to 4% experience impoverishment due to diabetes healthcare. This finding is comparable to other African studies that assessed the impoverishing effects of healthcare using national datasets [40, 41, 59]. This means that up to 4% of households are negatively affected and forced into poverty as a result of diabetes healthcare costs. This is a huge problem particularly in households that do not only incur diabetes related costs but other health related costs.

Our study also provides insight into the determinants of catastrophic health expenditure and impoverishment amongst diabetics visiting public tertiary healthcare facilities. When measuring catastrophic health expenditure as healthcare costs plus transport costs at a 10% threshold of capacity to pay and healthcare costs plus transport costs at a variable threshold of total household expenditure we found that gender, wealth and children were the variables associated with catastrophic health expenditure. Our finding that households within the higher wealth quintile have reduced odds of incurring catastrophic expenditure, is intuitive and consistent with a study by Babikir et al. that makes use of the National Income Dynamics Survey [12]. Patients who reported not having any children, had increased odds of incurring catastrophic health expenditure and impoverishment. Although the study did not distinguish between young and old or employed and unemployed children, our finding may point to the role that family or social networks play in assisting with healthcare costs. We found that being female was associated with an increased odds of catastrophic health expenditure and that being single was associated with a reduced odds of impoverishment. We find that when compared to the first quintile the odds of being impoverished for those in the fourth and fifth quintile is statistically insignificant. This finding is not consistent with previous literature [60] and calls for further investigations.

Our study has some limitations. Due to the recall periods, results may be subject to recall bias. Our study focused on those who used healthcare services, thus, excluding those with undiagnosed diabetes who still incur healthcare costs as a result of other illnesses that could be related to diabetes. Our study does not take into account other healthcare costs that may be incurred by patients and are not related to diabetes. The exclusion of observations with missing data in variables used for the construction of catastrophic health expenditure or impoverishment may have introduced bias in our study results. Also, our study is not nationally representative as it does not reflect the costs associated with diabetes care in private outpatient clinics. The resulting impact is that the findings from this study cannot be generalised as they are not applicable to diabetic patients who access private healthcare facilities where medical insurance is a common healthcare payment method. Furthermore as reflected by the recorded response rate of 81%, the risk of selection bias cannot be ignored. A comparison of our final analytic sample and refusals showed that our final analytic sample contained more Africans (76% versus 72%) and contained less females (61% versus 62%). Socio-demographic differences between the final sample and refusals could have introduced bias in the study. The methods used in our study have been criticised for ignoring the role of savings, assets, family and friends in healthcare financing [42, 61]. Although recent studies encourage the use of a range of equivalence scales when using the WHO method our study only applies one scale. Despite these shortcomings, these methods provide useful measures of catastrophic health expenditure [42] that inform policy makers on the need for financial risk protection amongst diabetics visiting public hospitals. Our study may form the basis for future investigations across the public healthcare facilities in South Africa. Further qualitative research is needed to complement the quantitative findings from this study, in particular to capture the complex nature of factors associated with patients’ diabetes management.

Our results are an important contribution to the literature on diabetes costs amongst patients visiting public hospitals in South Africa. We find that transport costs contribute a significant portion to direct healthcare costs. These high costs point to the urgent need for interventionist programmes to improve access to healthcare for chronic patients who make multiple visits to healthcare centres, such as through subsidised or free transport services. Our study shows that some diabetic patients do incur catastrophic healthcare expenditure and that inequalities in catastrophic health expenditure favour those within more affluent socio-economic groups. We find that financial protection of diabetes patients visiting South African public hospitals is limited due to the difficulty in identifying unemployed individuals to be exempted from paying for health services. These expenditures are quite high considering the high unemployment rates and multiple annual hospital visits by many of the diabetics. Our study also shows that being female and not having children significantly increases the odds of catastrophic health expenditure for diabetes care. Being more affluent reduced the odds of catastrophic health expenditure. These observations suggest health financing interventions amongst diabetic patients should further target the poor and poor women in particular. Thus, a recommendation of the study to government is for the elimination of hospital fees for diabetic patients, which must be accompanied by targeted interventions to reduce transport costs associated with accessing hospitals.