Investigating the affordability of key health services in South Africa

Social Science & Medicine 80 (2013) 37e46

Contents lists available at SciVerse ScienceDirect

Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed

Investigating the affordability of key health services in South Africa Susan Cleary a, *, Steve Birch b, Natsayi Chimbindi c, Sheetal Silal d, Di McIntyre a a

Health Economics Unit, School of Public Health and Family Medicine, University of Cape Town, South Africa Centre for Health Economics and Policy Analysis, McMaster University, Canada c Africa Centre for Population and Health Studies, University of KwaZulu-Natal, South Africa d Department of Statistical Sciences, University of Cape Town, South Africa b

a r t i c l e i n f o

a b s t r a c t

Article history: Available online 12 December 2012

This paper considers the affordability of using public sector health services for three tracer conditions (obstetric care, tuberculosis treatment and antiretroviral treatment for HIV-positive people), based on research undertaken in two urban and two rural sites in South Africa. We understand affordability as the ‘degree of fit’ between the costs of seeking health care and a household’s ability-to-pay. Exit interviews were conducted with over 300 patients for each of the three tracer conditions in each of the four sites (i.e. a total sample of over 3600). Total direct costs for the service used at the time of the interview, as well as other health related costs incurred during the preceding month either for self-care or the use of plural providers were assessed, as were a range of indicators of ability-to-pay. The percentage of households incurring direct costs exceeding 10% of household consumption expenditure and those borrowing money or selling assets as a mechanism for coping with the burden of direct costs were calculated. Logistic regressions were also conducted to identify factors that were significantly associated with these indicators of affordability. There were significant differences in affordability between rural and urban sites; costs were higher, ability-to-pay was lower and there was a greater proportion of households selling assets or borrowing money in rural areas. There were also significant differences across tracers, with a higher percentage of households receiving tuberculosis and antiretroviral treatment borrowing money or selling assets than those using obstetric services. As these conditions require expenses to be incurred on an ongoing basis, the sustainability of such coping strategies is questionable. Policy makers need to explore how to reduce direct costs for users of these key health services in the context of the particular characteristics of different treatment types. Affordability needs to be considered in relation to the dynamic aspects of the costs of treating different conditions and the timing of treatment in relation to diagnosis. The frequently high transport costs associated with treatments involving multiple consultations can be addressed by initiatives that provide close-to-client services and subsidised patient transport for referrals. Ó 2013 Elsevier Ltd. All rights reserved.

Keywords: Affordability Access Obstetric services Tuberculosis Anti-retroviral treatment Service costs Ability-to-pay

Introduction The issue of affordability of health services has received increasing attention over the past two decades. Initially the focus was on what are termed ‘cost of illness’ studies. These studies quantified the direct, and sometimes also the indirect, costs related to health service use, usually focussing on specific diseases such as malaria, lymphatic filariasis and HIV/AIDS. These studies were often used for advocacy purposes to highlight the sometimes considerable cost burden placed on households by certain diseases,

* Corresponding author. E-mail address: [email protected] (S. Cleary). 0277-9536/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.socscimed.2012.11.035

and were used to motivate for increased public and global spending on preventing and treating these diseases. More recently, there has been a growing literature comparing the direct costs of health care to households’ ability-to-pay through assessing catastrophic levels of health care spending. These studies use a reference point of either 10% of household income consumed by health care expenditure (Prescott, 1999; Ranson, 2002) or health care costs exceeding 40% of non-food household expenditure (Xu et al., 2003) as being considered catastrophic. A related set of studies (that assess direct costs relative to ability-to-pay) has focused on estimating the number of households that have been impoverished (or pushed below the poverty line) due to incurring the direct costs of health care (Van Doorslaer et al., 2006; Wagstaff & van Doorslaer, 2003). The studies on catastrophic spending levels and impoverishment impact have been extensively used in arguing

38

S. Cleary et al. / Social Science & Medicine 80 (2013) 37e46

for moving away from out-of-pocket payments for health care towards pre-payment funding mechanisms (Xu, Evans, Carrin, & Aguilar-Rivera, 2005; Xu et al., 2006). A key limitation with many of these studies is that they have either not evaluated the direct costs of health care relative to households’ ability-to-pay in the case of the ‘cost of illness’ studies or have used what are essentially arbitrary reference points for determining whether or not health care expenditure is catastrophic for a household. More recent longitudinal studies have highlighted that some households that incur health care expenditure that is a relatively high proportion of total household income (as high as 28%) manage to cope with this cost burden, including some households with relatively low socio-economic status, while others with direct health care costs as low as 4% of household consumption expenditure become impoverished (Goudge et al., 2009). There is a growing awareness that it is important to be more explicit about the conceptualisation of ability-to-pay, catastrophic spending levels and affordability (Russell, 1996, 2004). Russell has particularly contributed in this regard, highlighting that in order to mobilise resources to meet the direct costs of health care, households “... may sacrifice other basic needs such as food and education with serious consequences for the household or individuals within it. The opportunity costs of payment make the payment ‘unaffordable’ because other basic needs are sacrificed” (Russell, 1996). This highlights the need to consider a household’s ability-to-pay in the context of not only the level of a household’s income but also other demands on that household’s budget. Linked to this understanding of ability-to-pay, the term catastrophic has been defined as expenditure levels that are “likely to force households to cut their consumption of other minimum needs, trigger productive asset sales or high levels of debt, and lead to impoverishment”(Russell, 2004). These conceptual observations stimulated research on how households mobilise resources to cover health care costs. While some households are able to use savings, others face the need to reduce consumption, particularly food (Rugalema, 1998; Tibaijuka, 1997). Other frequent coping strategies are the sale of assets (Kabir, Rahman, Salway, & Pryer, 2000; Sauerborn, Adams, & Hien, 1996; Wilkes, Hao, Bloom, & Xingyuan, 1997) and borrowing, either from family and friends or from a money lender (McPake, Hanson, & Mills, 1993; Nahar & Costello, 1998). One could regard the two different approaches to assessing affordability of health care costs respectively as being:

This paper considers the issue of affordability in relation to the use of public sector health services for three tracer conditions (obstetric care, tuberculosis (TB) treatment and antiretroviral treatment (ART) for HIV-positive people), based on research undertaken in two urban and two rural sites in South Africa. We understand affordability to relate to the interaction, or ‘degree of fit’, between the costs of seeking health care and household’s ability to pay (Penchansky, 1977). These tracer conditions are particularly problematic for South African policy makers in their attempts to achieve the Millennium Development Goals. In addition, although all of these services should be provided with no service fees at public facilities in South Africa, the nature of the conditions differ in ways that may involve different impacts on affordability. An important focus of this paper is the comparison of three services that have different characteristics, particularly in terms of the period over which costs will be incurred. Obstetric care represents once-off, or at least very infrequent, costs with the major costs occurring well after the time of diagnosis e hence providing an opportunity to plan for meeting the costs of the delivery. TB requires frequent service use, but this is generally over a limited period of time (usually 6e9 months, or longer in the case of drug resistant TB) with the expectation of recovery. ART not only requires frequent service use, but is also lifelong. Hence the tracer conditions present different types of affordability challenges to individual and households and provide a basis for gaining insights into how the ‘fit’ between health care costs and households’ abilityto-pay varies according to the type of health service needed. We seek to explore whether these differences result in different impacts on individuals’ and households’ affordability of using care. If so, policies aimed at improving affordability may need to be customised to the nature of the particular conditions. We are particularly interested in identifying the characteristics of households that experience difficulties in affording health care. As highlighted in the literature on household coping strategies, the clearest indicator of affordability difficulties is borrowing and/or sale of assets to cover the direct costs of using health services. We also explore the factors that make health service use affordable for some households, i.e. able to avoid borrowing or asset sales. We recognise that this is at the extreme end of the spectrum of affordability, but our data do not allow consideration of the impact of health care spending on other household budget items. Another important feature of our analysis is the comparison of affordability across rural and urban sites. Methods


Normative in the case of the thresholds for determining catastrophic expenditure, as they imply that household ought to be able to find the costs affordable if they are below the percentage of household expenditure threshold (or unaffordable if above the threshold); and
Positive in the case of households that have to borrow or sell assets in order to cope with service costs and hence, face serious affordability constraints. Another constraint of many of the earlier studies is that they have often focused on a single disease category. From a review of studies focussing on individual diseases, Russell (2004) concluded that different diseases impose different direct and indirect cost burdens and have different risks for household livelihood sustainability. However, there are very few instances where affordability issues for different diseases have been evaluated within a single study (Perera & Gunatilleke, 2004). A recent longitudinal study indicated that persistent conditions requiring frequent health service visits (such as TB and HIV) impose particularly high cost burdens (Goudge et al., 2009).

Sampling Four health sub-districts in different provinces were selected as sites for this research, two in urban areas (Mitchells Plain in the Western Cape province and Soweto in Gauteng) and two in rural areas (Bushbuckridge in Mpumalanga and Hlabisa in KwaZuluNatal). The sampling of these sites was designed to reflect different geographic locations (ruraleurban mix) and to allow for differences in governance contexts, given that provinces in South Africa have considerable decision-making autonomy in the provision of health services. Sampled sub-districts also needed to have at least one hospital providing comprehensive essential obstetric care (CEOC), including caesarean sections, at least two facilities providing ART and at least two facilities providing TB treatment. Key officials in the national and provincial health departments were consulted in finalising the selection of sub-districts. A two-stage sampling approach was used in each sub-district, first selecting a representative sample of health facilities, then within these facilities, a representative sample of users. All facilities

S. Cleary et al. / Social Science & Medicine 80 (2013) 37e46

providing CEOC services in the sub-district were included in the sample (one in Hlabisa, two each in Bushbuckridge and Mitchells Plain and three in Soweto). Proportional sampling methods and routine facility data on the number of deliveries per facility at the time of data collection were used to determine the number of interviewees per facility. As most public health facilities provide TB services, a minimum of five facilities were selected in each subdistrict and probability proportional to size (PPS) methods were used to select facilities. For ART, all accredited facilities were included where possible, and where multiple facilities existed, selfweighting stratified, proportional or probability proportional to size methods were used to select facilities, again using routine data on the total number of users in each facility at the time of the research. Within each chosen facility, a random sample of patients was interviewed until the proposed facility sample size was reached. In total, a minimum of 300 patients were interviewed per tracer and per sub-district; the planned sample size was therefore 3600 respondents. Data collection and capture Patient exit interview questionnaires were developed to collect demographic and socio-economic data as well as information on health service use, direct costs associated with health care and aspects of access to health care. While many questions were identical across tracers, some questions were specific to individual tracer services. The questionnaire was administered by trained interviewers in the language of the respondent’s choice. Completed questionnaires were checked for accuracy by a data collection coordinator and double entered into a data entry platform specifically designed for this purpose in Epidata. Data analysis Data were analysed using Stata/SE 10.0. The focus of this analysis was on quantifying direct costs related to the tracer condition or service, assessing ability-to-pay and comparing direct costs to households’ ability to pay. A number of different types of direct costs were assessed. Firstly, we assessed the costs associated with a TB or ART service visit and with a delivery. To reduce recall biases, these costs were collected either at the time of the visit (in the case of TB and ART costs) or directly after the delivery (in the case of CEOC). We probed specifically into the money spent on transport to the service, food, phone calls or accommodation during the visit or delivery, service user fees or medicine payments, and payments made to others to take over tasks (including childcare) during the visit or delivery. For CEOC services, we also asked about money spent on sanitary towels, nappies, or other supplies that mothers are required to bring with them when they give birth. Given that some CEOC patients would have sought care at more than one service point while in labour (for example they might have been referred from the Midwife Obstetric Unit to the district hospital), additional questions were asked about any transport costs incurred during referral. Because our aim was ultimately to assess overall monthly spending on health care, we asked respondents to indicate the frequency of their visits to TB or ART services so that we could calculate monthly equivalent visit costs (see more below). In addition to the costs associated with the visit or delivery, we asked respondents about any other health related expenditure during the preceding one-month period. Here, we specifically asked about spending on self care items (which we defined as nonprescription medicines purchased from a pharmacy, traditional medicines or special foods). We also probed into the use of other providers, asking respondents about any visits to private pharmacies, private general practitioners, traditional healers, public clinics

39

(for services other than those needed for the tracer condition), public or private hospital emergency or outpatient department visits, and public or private hospital inpatient stays. Should the patient report the utilisation of any of the abovementioned services, we then probed further to find out how much money had been spent. For TB and ART respondents, we asked about the use of these providers during the preceding one-month period while for CEOC users we asked for details of service use during their pregnancy. As suggested above, costs were collected for different time periods e self care costs were reported for the preceding month, plural provider costs were reported for the preceding month for TB and ART but for the duration of the pregnancy for CEOC, and the costs for TB, ART and CEOC services were reported for the current visit or for the birth. We calculated monthly visit costs for TB and ART by multiplying the reported visit costs against the reported frequency of visits and we calculated monthly costs for CEOC by assuming that the costs incurred in giving birth were spread over a month while the costs of plural providers were distributed over nine months (the assumed duration of the pregnancy). The distribution between different categories of direct costs was compared across tracers and sites. To assess direct costs relative to ability-to-pay, we contrasted these monthly direct costs with monthly household consumption expenditure. Because household expenditure data are difficult to collect in low and middle-income countries, we used a number of strategies to increase the response rate. Firstly, we asked the question about expenditure at the end of the interview. Secondly, we asked an open-ended question e “What is your overall monthly household expenditure” e but if this question proved challenging for the respondent, we asked whether the respondent could estimate their monthly household expenditure from within a range of categories. These were based on commonly used expenditure categories in other South African surveys. In analysis, we recoded all the responses to the first open-ended question into expenditure categories. Then, using a regression technique, we predicted the values for a quantitative monthly household expenditure variable where such expenditure was a function of the employment status of the individual and the household head, the sex of the household head, the sub-district where the household was located, the asset index, years of education of the respondent, and the value of grants in the household. Alternative ways of doing this (for example assuming that the quantitative variable was the arithmetic or geometric mean value between the lower and upper bands in the household expenditure category) were also explored. We found a very high level of correlation (>90%) between these different approaches, and in the final analysis chose to use the measure based on the regression technique. In contrasting monthly household expenditure to direct monthly costs, two measures of ability to pay were computed. The first expressed direct monthly costs as a percentage of monthly household consumption expenditure, while the second was a binary measure indicating whether households incurred direct monthly costs in excess of 10% of monthly household consumption expenditure following the practice in the literature (Ranson, 2002). We recognise that this is a crude measure of catastrophic spending using what could be described as an arbitrary cut-point. Nevertheless, as this indicator is widely used, it allows comparison with other studies. In addition to the above analysis of costs relative to overall consumption expenditure, we also collected and analysed a range of measures of socio-economic status to gain additional insights into ability-to-pay, including employment status of the respondent and the household head, whether or not the household received social grants and the value of those grants, and a composite asset index. We specifically probed into the receipt of the following social

40

S. Cleary et al. / Social Science & Medicine 80 (2013) 37e46

grants: unemployment insurance, worker’s compensation, old age pensions, disability grants, child support grants, care dependency grants and foster care grants. The asset index was created using multiple correspondence analysis (MCA) on several household level variables including type of house, material of walls, type of toilet, primary source of energy for cooking and ownership of assets such as a vehicle, fridge and livestock etc. Finally, in order to explore affordability issues from a different perspective, we focused on households that had to borrow money or sell assets as a mechanism for coping with the burden of direct costs. In analysis, p-values were computed using the KruskaleWallis comparison of means test for quantitative data and the Pearson’s chi-squared test for categorical data. Logistic regressions were conducted to identify factors that were significantly associated with incurring catastrophic spending and with having to borrow or sell assets. Direct costs, household expenditure and the value of grants are expressed in South African Rands. The average exchange rate over the period of data collection was 1 US dollar to 9.0294 South African Rands. Ethical issues Ethical approval for the study was granted by committees at the University of Cape Town, the University of the Witwatersrand and the University of KwaZulu-Natal. Permission from health department officials and individual facility managers was obtained to conduct the study in the selected facilities. Written informed consent to participate in the study was obtained from each participant; participants were only interviewed if they were over 18 years of age. Results A total of 3727 patients were interviewed, 51% in the urban sites and 49% in the rural sites. There were 1231 respondents for the CEOC tracer, 1229 for TB treatment and 1267 for ART. Costs associated with health service use Very few respondents indicated that they had incurred expenses directly related to the health service itself (e.g. service fees, paying for medicines). Only 2.3% of users reported such expenditure (mean expenditure of R25), and this was restricted to obstetric services in one of the rural sites. However, this does not imply that respondents did not incur any direct costs. The mean direct costs associated with delivery were R266 (95% CI: 251e281), compared with R16 (95% CI: 12e20) for the most recent TB visit and R28 (95% CI: 26e31) for the most recent ART visit. In order to provide greater comparability between the three tracer conditions, given that in some sites TB patients have more frequent visits to health care providers than ART patients, we calculated monthly equivalent costs for TB and ART. Monthly costs, related to the use of other health services (i.e. not their primary provider for ART and TB care) and on self-treatment, were also included to provide a more comprehensive overview of health care related costs for patients with TB and on ART. In the case of TB patients, the mean total monthly costs were R100 (95% CI: 89e111) and R81 (95% CI: 72e90) for patients on ART. The mean total monthly cost for pregnant women was R321 (95% CI: 303e339) (i.e. for the delivery and their average monthly spending on other health services and self-treatment) (see Fig. 1). Fig. 1 highlights that there are differences in the composition of direct monthly costs across tracers and sites. For patients receiving TB treatment, higher transport costs are incurred for patients in Mitchells Plain, while self care costs are higher in the rural sites. On

Fig. 1. Mean monthly spending on health care by tracer and site. p-Values computed using KruskaleWallis test. Other costs at facility: food, phone, accommodation etc. Self care: over the counter medicines, special foods, and traditional medicines. Supplies: sanitary towels, nappies etc

average, ART patients incur higher expenditure on self-care than on other categories of direct costs, but these findings are driven by particularly high spending on this category in the rural site of Hlabisa. The pattern is quite different in the case of CEOC, where supplies (nappies, sanitary towels etc) are by far the largest share of total direct costs. Transport and self-care costs are also important contributors to direct costs for those using delivery services; indeed, they are greater in absolute terms than for either ART or TB treatment. Within each tracer condition, we found significant differences in the mean overall cost incurred between the different sites (all p-values <0.001). In general, there were higher transport costs for patients in rural areas, and in Mitchells Plain there were high transport costs for patients receiving TB treatment, which is influenced by the higher visit frequency of TB patients in Mitchells Plain relative to other sites. Indicators of ability-to-pay Table 1 highlights that there were high levels of unemployment amongst respondents and amongst household heads, with the highest levels for TB service users, followed by those on ART and finally pregnant women. As shown in the table, there were

S. Cleary et al. / Social Science & Medicine 80 (2013) 37e46

41

Table 1 Indicators of ability to pay, by site and tracer. Asset index p-Value Respondent p-Value Household head p-Value Median household p-Value % Households p-Value Median value p-Value (% in richest unemployed unemployed (%) expenditure receiving of household half) (%) (Rands) grants grants (Rands) TB Bushbuckridge Hlabisa Soweto Mitchells Plain

46.70 28.81 13.85 88.22 55.09

ART Bushbuckridge Hlabisa Soweto Mitchells Plain

44.44 22.12 9.33 80.06 62.04

CEOC Bushbuckridge Hlabisa Soweto Mitchells Plain

58.98 41.47 17.00 87.59 86.84

All tracers Bushbuckridge Hlabisa Soweto Mitchells Plain

49.99 30.67 13.39 85.08 68.20

0.000

0.000

0.000

0.000

83.47 89.74 90.51 75.08 79.04 77.67 82.69 88.00 71.95 69.04 75.14 87.63 87.33 65.86 61.40 78.75 86.64 88.60 71.04 69.77

0.000

0.000

0.000

0.000

71.05 77.74 81.69 67.57 58.61 68.13 72.90 86.44 60.12 54.83 42.02 47.49 63.64 32.07 26.90 60.44 66.15 77.23 53.62 46.48

0.000

0.000

0.000

0.000

903.48 926.35 920.28 710.76 933.47 972.85 940.11 923.33 978.61 1055.38 1072.20 957.22 973.40 1072.10 1691.67 983.42 939.12 935.82 989.15 1151.18

0.0087

0.0001

0.0001

0.0001

73.72 79.47 88.51 65.99 62.28

0.000

77.98 85.58 92.33 65.56 70.06

0.000

66.53 75.59 94.00 49.66 48.83

0.000

72.79 80.28 91.63 60.68 60.20

0.000

570.00 570.00 1200.00 380.00 190.00 820.00 820.00 1390.00 380.00 380.00 380.00 380.00 1200.00 0.00 0.00 380.00 570.00 1200.00 190.00 190.00

0.0001

0.0001

0.0001

0.0001

p-Values computed using KruskaleWallis test for quantitative data; chi-squared goodness of fit test for binary data.

significant differences in these employment levels between tracers overall, and, within the tracers, between the different sites (all pvalues <0.001). Although not shown in the table, the CEOC respondents and their household heads were significantly more likely to be employed than their counterparts using TB and ART services (p-values <0.001). There was a similar pattern in the median monthly household consumption expenditure, which was highest in households using CEOC services, followed by those using ART services and finally TB service users. There are quite large significant differences in these indicators between rural and urban areas and also within these areas. Mitchells Plain has indicators reflecting higher socio-economic status than Soweto (the two urban areas), with the exception of the asset index. In the two rural areas, Bushbuckridge has higher socio-economic status than Hlabisa. Within-tracer differences in socio-economic status across the sites were significant for all indicators. Given the differences in the employment status of household heads, it may have been expected that there would have been greater differences in consumption expenditure across the tracer conditions and sites. This may partly be explained by differences in the receipt of social grants. Table 1 indicates that a higher percentage of households with a member receiving ART and TB treatment were receiving a social grant than households with a pregnant woman and the median value of the grants follows this pattern. The differences between rural and urban areas, as well as between the individual sites, are even more striking, with rural areas having a significantly higher coverage by social grants and grant values. It is interesting to note that the value of social grants is at times higher than the reported household expenditure. While it is possible that this may reflect household savings, it is more likely to be symptomatic of the difficulty of collecting household expenditure data, particularly in rural settings where households are very large. In collecting the grant data, we simply asked respondents how many of the various grants were received by their household members. We subsequently calculated the monthly value of these grants using data on the monthly payout per grant type from the South African Social Security Agency. The grant data

is therefore likely to be more accurate than the household expenditure data. Direct costs relative to ability-to-pay To crudely explore the ‘degree of fit’ between health care related expenditure and households’ ability-to-pay, Table 2 demonstrates that spending on health services accounts for a relatively large percentage of household expenditure, particularly in the case of Table 2 Direct costs relative to ability-to-pay. Health care as % of household spending TB Bushbuckridge Hlabisa Soweto Mitchells Plain

13.06 17.31 15.34 4.02 14.02

ART Bushbuckridge Hlabisa Soweto Mitchells Plain

9.41 10.49 19.00 3.41 5.08

CEOC Bushbuckridge Hlabisa Soweto Mitchells Plain

33.21 51.43 49.71 25.16 16.65

All tracers Bushbuckridge Hlabisa Soweto Mitchells Plain

17.87 24.70 25.45 10.89 11.92

p-Value

0.0001

0.0001

0.0001

0.0001

Incurred catastrophic expenditure (%) 32.99 35.25 50.38 10.76 32.21 22.71 25.60 50.18 8.18 8.20 66.09 87.28 80.54 55.42 50.64 39.35 46.80 57.95 24.83 30.30

p-Value

0.000

0.000

0.000

0.000

p-Values computed using KruskaleWallis test for quantitative data; Pearson’s chisquared test for categorical data.

42

S. Cleary et al. / Social Science & Medicine 80 (2013) 37e46

delivery services. The table also reports on ‘catastrophic’ health care expenditure, measured as expenses that exceed 10% of overall household expenditure (as a proxy of household resources). Twothirds of households using CEOC services and a third of those using TB treatment services face monthly direct costs exceeding 10% of monthly household expenditure. Table 3 explores the relationship between ‘catastrophic’ health care spending, and other variables. The variables that are significantly associated with catastrophic health care spending (at the 5% level) are the study sites (with the two urban sites being significantly less likely to experience catastrophic health care expenditure than the reference site of Bushbuckridge, and Hlabisa being the most likely to experience catastrophic expenditure) and the tracers (with obstetric care patients being the most likely to face health care costs exceeding 10% of household expenditure while ART are least likely to face such costs). Although patients with higher levels of education are more likely to incur catastrophic expenditure, the odds ratio is close to 1. Strategies for coping with health care costs The majority of respondents (72% for TB patients, 65% for those receiving ART and 93% for those using obstetric services) indicated that they received support, largely from family (spouse, partner, parents or other relatives), to meet the costs of health services (see Table 4). Such support was specified separately from questions about borrowing money to pay for health care costs. Only 7% of pregnant women reported receiving no support, while 29% of those receiving TB treatment and 35% of those on ART received no support. These differences were significant (p < 0.001). With the exception of CEOC, there were significant differences in the level of support across sites. A higher percentage of users of TB and ART services (almost 20% in both cases) borrowed or sold assets to cover the costs of health services in comparison to obstetric care (10%) (p < 0.001) (see Table 4). Borrowing was the primary coping strategy, with asset sales to cover health care costs being very low (2.7% for those on ART, 2.5% for TB service users and 1.1% for those using obstetric care). There were significantly higher rates of borrowing in rural than urban areas, but for those who did borrow, the median amount borrowed was higher in Soweto than the other sites in the Table 3 Regression exploring relationship between ‘catastrophic’ health care related expenditure (>10% of household expenditure) and key socio-economic and related factors. Variable

Incurring catastrophic expenditure OR (95% CI)

p-Value

Asset index (1 ¼ richer) Respondent employment (1 ¼ employed) Household head employment (1 ¼ employed) Grant value Education of respondent (years) Received no financial support (1 ¼ yes) Tracer ¼ TB Tracer ¼ ART Tracer ¼ CEOC Site ¼ Bushbuckridge Site ¼ Hlabisa Site ¼ Soweto Site ¼ Mitchells Plain Pseudo R2

0.87 (0.71e1.07) 1.06 (0.86e1.32)

0.202 0.586

1.06 (0.87e1.29)

0.575

1.00 (1.00e1.00) 1.03 (0.71e1.06)

0.397 0.015

0.87 (1.01e1.06)

0.168

Referent 0.56 (0.46e0.68) 4.72 (3.80e5.87) Referent 1.89 (1.49e2.40) 0.27 (0.20e0.35) 0.37 (0.30e0.47) 0.1826

OR ¼ odds ratio; 95% CI ¼ 95 percent confidence interval.

0.000 0.000 0.000 0.000 0.000

case of TB treatment and CEOC service users. The median amount borrowed was higher for obstetric care (R100) than for ART and TB services (both R50) (p ¼ 0.0053). Table 4 also indicates that the majority of respondents did not face serious affordability difficulties, in the sense of not having to resort to either borrowing or asset sales during the past one month period. This was particularly the case in relation to obstetric care and for those living in urban areas. However, in Hlabisa, up to 40% of TB and ART patients had to borrow to cover costs related to their health care. A logistic regression enabled us to explore the factors that contributed to the likelihood of a household having to borrow and/ or sell assets to cover the costs of health care, i.e. the likelihood of a household facing serious affordability difficulties. As shown in Table 5, the most significant factors were community location, tracer condition and asset ownership. As shown in Table 4, the proportion of respondents borrowing and/or selling assets was significantly greater in Hlabisa than in other districts, followed by the other rural district (Bushbuckridge), with Mitchells Plain and finally Soweto (the two urban districts) having the lowest likelihood of encountering severe affordability problems. These findings are verified within the logistic regression summarized in Table 5. The likelihood of borrowing and/or selling assets was significantly lower for obstetric care than TB and ART services, with no significant difference between TB and ART. As expected, higher socioeconomic status (represented by ownership of assets) translates into a lower likelihood of encountering severe affordability difficulties. Discussion This paper focuses on an analysis of the affordability of health services for only those using services and hence we are not able to explore affordability issues at a population level. Instead, we have analysed affordability among service users from an equity perspective in terms of whether the affordability of these services is distributed equitably among users. We may, however, be excluding some who face sizeable affordability barriers, as there is considerable evidence that many of the poorest households do not seek care when ill to avoid incurring health care costs (Chuma, Thiede, & Molyneux, 2006; Russell & Abdella, 2002). For example, a census of four villages in Lao PDR found that 40% of those who reported illness episodes adopted the cost prevention strategy of not seeking treatment (Patcharanarumol, Mills, & Tangcharoensathien, 2009), while a household survey in a rural community in South Africa found that 48% of those reporting a chronic illness (an illness lasting more than one month) did not seek treatment (Goudge, Gilson, Russell, Gumede, & Mills, 2009a). We are also focussing on the most extreme end of the affordability spectrum in the form of borrowing and asset sales. While we recognise that it is also important to consider affordability in terms of the impact of health care expenditure on household spending on other basic items, our data did not permit such analysis. Borrowing and asset sales are particularly important in the case of diseases such as TB and HIV where health care costs are incurred on an ongoing basis. Meeting the direct costs of any health service use may crowd out household spending on other basic needs in the short term, but this type of coping strategy may not be sustainable in the longer term because of limitations on the capacity to borrow or deplete assets. Hence affordability issues need to be considered not only in terms of whether or not individuals use care, but also whether or not individuals are able to comply or adhere to treatment protocols when they use care. This would be particularly important for non-acute conditions such as TB and HIV where service use extends over a period of time (lifetime in the case of

S. Cleary et al. / Social Science & Medicine 80 (2013) 37e46

43

Table 4 Key strategies for coping with direct health care costs. Received financial support to cover expenses (%) TB Bushbuckridge Hlabisa Soweto Mitchells Plain

71.50 69.54 70.67 66.33 78.61

ART Bushbuckridge Hlabisa Soweto Mitchells Plain

65.07 60.53 68.23 61.82 69.78

CEOC Bushbuckridge Hlabisa Soweto Mitchells Plain

92.60 91.97 91.33 93.10 93.84

All tracers Bushbuckridge Hlabisa Soweto Mitchells Plain

76.34 73.92 76.87 73.17 80.99

p-Value

Borrowed or sold assets to pay for health care (%) 19.69 23.84 40.88 0.34 14.37

0.005

19.49 27.56 38.67 2.11 11.73

0.033

9.91 13.71 14.67 0.69 10.23

0.624

16.39 21.80 31.36 1.09 12.10

0.000

ART) and is an important area for future research. Nevertheless, exploring the extent to which households have to become indebted or sell assets in order to use health services even in the short term is of policy concern. Our results indicate that the majority of users of the three tracer services did not have to borrow or sell assets to cover the costs associated with such use, although rates of borrowing and asset sales were considerably higher in the rural sites. They were able to cover these costs out of household income, including income from social grants and financial assistance from family, or savings. A number of other studies have found that a household’s first response to covering the costs of using health services is to use the available household budget and to mobilise savings (Chuma, Gilson,

Table 5 Logistic regression results for households that borrow and/or sell assets to cover health care costs. Variable

Borrowing and/or selling assets to pay for health care OR (95% CI)

p-Value

Asset index (1 ¼ richer) Respondent employment (1 ¼ employed) Household head employment (1 ¼ employed) Grant value Household expenditure Education of respondent (years) Received no financial support (1 ¼ yes) Tracer ¼ TB Tracer ¼ ART Tracer ¼ CEOC Site ¼ Bushbuckridge Site ¼ Soweto Site ¼ Hlabisa Site ¼ Mitchells Plain Pseudo R2

0.63 (0.49e0.82) 0.80 (0.55e1.04)

0.001 0.087

0.91 (0.71e1.18)

0.482

1.00 (1.00e1.00) 1.00 (1.00e1.00) 1.00 (0.97e1.02)

0.616 0.941 0.806

0.96 (0.76e1.21)

0.722

Referent 1.02 (0 0.82e1.28) 0.532 (0.40e0.71) Referent 0.05 (0.03e0.11) 1.79 (1.40e2.29) 0.65 (0.49e0.86) 0.1474

OR ¼ odds ratio; 95% CI ¼ 95 percent confidence interval.

0.856 0.000 0.000 0.000 0.002

p-Value

0.000

0.000

0.000

0.000

Median amount borrowed 50.00 60.00 90.00 150.00 22.00 50.00 50.00 100.00 30.00 20.00 100.00 100.00 100.00 1125.00 50.00 53.00 60.00 100.00 50.00 30.00

p-Value

0.0019

0.0001

0.0642

0.0001

& Molyneux, 2007; McIntyre, Thiede, Dahlgren, & Whitehead, 2006; Wyss, Hutton, & N’Diekhor, 2004). Sometimes this requires making a special effort to generate additional income. For example, a study in Tanzania found that the main way of funding treatment costs was to sell crops or undertake petty trading (Save the Children, 2005). As indicated earlier, it may also require a reduction in household spending on other items. The need to borrow or sell assets is a particularly severe problem for users of TB and ART services in rural areas. It is particularly concerning that about 40% of TB and ART service users in the one rural site (Hlabisa) had to borrow or sell assets to cope with health care costs. Given the long-term nature of both these illnesses, it is to be expected that there would be affordability difficulties in relation to the costs of TB and ART services. A study that compared the costs to users of HIV services with users of other health services found that borrowing and the sale of assets was greatest amongst users of HIV services (Wyss et al., 2004). The high incidence of borrowing and sale of assets to cover health care costs has been found with other long-term illnesses, such as visceral leishmaniasis (Adhikari, Maskay, & Sharma, 2009; Sharma et al., 2006) and dengue (Van Damme, Van Leemput, Por, Hardeman, & Meessen, 2004). There is also strong evidence that the need to borrow and sell assets in order to cover the costs of using health services is far greater for households with a lower socio-economic status than higher income households (Levie & Xu, 2008). This was confirmed in our analysis by the significant relationship between the asset index and the incidence of borrowing or selling assets. The incidence of borrowing and asset sales is lowest for CEOC users, despite the absolute value of direct costs associated with obstetric care being higher than for TB and ART. This is partly due to costs of obstetric care being incurred on an infrequent basis. It is also possibly the case that individuals and households plan for these costs during the course of the pregnancy. Also, households using obstetric care services on average tend to have higher socioeconomic status, particularly in relation to a greater proportion of household heads being employed and higher median per capita household expenditure. This may be explained by pregnancy being experienced across the socio-economic spectrum, while TB and HIV

44

S. Cleary et al. / Social Science & Medicine 80 (2013) 37e46

are more heavily concentrated among poorer groups (Ataguba, Akazili, & McIntyre, 2011). Moreover, the high cost of obstetric care in the private sector may mean that more prosperous individuals and households may be forced to use the public sector for deliveries. In relation to the ability-to-pay side of the affordability equation, the results indicate that social grants are playing an important role in promoting the affordability of health service use. Our findings also suggest that social grants are reasonably well targeted, with a higher rate of grant receipts and higher median grant values in households where the household head and/or respondent is unemployed. Goudge et al. (2009: 244) noted that social grants in South Africa “. contributed considerably to regular and higher incomes, assisting protection against illness-related risks in three ways: health care was more affordable; grant status allowed automatic qualification for hospital [fee] exemption and cash transfers strengthened social networks. A regular grant also made a household member more creditworthy and more likely to give help and so request help in return (although grants also reduced the need to make claims on networks)”. Ability to cover direct costs of health care without borrowing or asset sales is also influenced by financial support from social networks. There is a striking difference in the support households receive from social networks, particularly close family and other relatives, across the tracer conditions. TB and ART service users were far less likely than those using obstetric services to receive financial support from family members. The birth of a baby may lead to financial support from close family members enthused by the prospect of an addition to the family, whereas TB and HIV are heavily stigmatised illnesses that may attract less support from social networks. This was found to be the case of TB and mental illness in Sri Lanka (Perera & Gunatilleke, 2004). Differences across tracers in the support received from social networks may also be related to the fact that the birth of a baby is an infrequent event whereas TB and HIV patients require long-term financial support. From a health system policy perspective, efforts to minimise direct costs facing service users is of particular importance. A key finding from our study is that the South African ‘free care policy’ is not being fully implemented in some areas. In 1994, President Mandela announced that all health services provided in any public sector facility would be provided free of any charge to pregnant women except those that had private health insurance coverage (McCoy, 1996). The policy was introduced to reduce financial barriers to health service use for vulnerable groups and incomplete implementation of it contributes to affordability barriers to institutional delivery for pregnant women. Ours is not the first study to find that the ‘free care policy’ (which includes free services for children under 6 years, pregnant women and recipients of social grants, and free services at all public sector primary care facilities) is not being fully implemented in South Africa (Goudge, Gilson, Russell, Gumede, & Mills, 2009b). Even if no service-related fees are charged, users still face direct costs. Transport costs can be substantial, and were the largest direct costs for TB and ART service users (along with self-care and seeking care from multiple providers) and the second largest (after the supplies that mothers in labour are required to bring to the delivery such as sanitary towels and nappies) for obstetric care users. A recent study in urban and rural areas of South Africa found that 91% of patients on ART incur transport costs, and that these can be substantial (Rosen, Ketlhapile, Sanne, & Bachman DeSilva, 2007). Transport costs have been found to be a high proportion of direct costs in many other studies, and are particularly so for lower income groups. For example, a recent study in rural South Africa found that transport costs accounted for 42% of direct costs across all respondents, but were 51% of direct costs in the case of the

poorest quintile (Goudge et al., 2009b). In relation to obstetric care specifically, a study in Nepal found that the costs of paying a traditional birth attendant to assist with a home delivery were very similar to the average fee for a normal delivery in a health facility. However, the average cost of transport to a health facility was more than four times the delivery fee, which was a key deterrent to delivering in a health facility (Borghi, Ensor, Neupane, & Tiwari, 2006). In our study, the costs of transport to a health facility were higher in absolute terms for CEOC than for TB and ART services. This reflects the fact that far fewer health facilities provide CEOC services than TB or ART services. It may be feasible to decentralise delivery services for low-risk pregnant women to reduce transport costs. The cost of supplies for facility-based deliveries is also very high, which is another area where health services could potentially reduce direct costs to patients. The finding that the monthly equivalent transport costs for TB services are far greater in Mitchells Plain than in other sites is also striking. This is largely attributable to the policy in the Western Cape that patients on TB treatment are required to attend a health facility on a daily basis to receive directly observed treatment (DOT). Other sites provide TB patients with a month’s supply of TB medicines, which patients are either trusted to take regularly without supervision or under community-based DOT supervision. The DOT supervision policy in the Western Cape requires urgent reconsideration. The approach adopted in the ART program, where patients undergo rigorous treatment compliance counselling before initiating therapy, and are encouraged to inform family and friends of their HIV status and get their support for treatment adherence, could also be applied for TB treatment. However, in the context of increasing incidence of multi-drug resistant TB in the Western Cape, there is likely to be resistance to such an approach. The recently initiated ‘PHC reengineering’ policy, which envisages community-health workers (CHW) in every district (with each CHW serving 250 households) (Department of Health, 2011), will make community-based DOT supervision feasible in all provinces, which would substantially reduce transport cost burdens for those receiving TB treatment. The finding that a large proportion of direct costs for ART and TB service users are related to spending on self-care and the use of multiple providers requires further investigation. In particular, it is necessary to determine whether this is due to deficiencies in the services provided at public facilities or whether it is simply a matter of choice (e.g. a preference for traditional healers). Our study has a number of limitations. Firstly, costs were collected in different ways for TB and ART services in comparison to CEOC. These differences predominantly related to the period over which cost data were collected e with some costs collected for the current visit or delivery and some costs collected for the preceding monthly period. To facilitate comparison, we calculated monthly costs (for TB and ART by multiplying the most recent visit costs against the reported frequencies of visits per month; for CEOC by assuming that delivery costs were incurred over a month). This was done purposely to reduce recall biases and to capture the different nature of these services, but it does introduce some potential limitations. For example, our use of short recall periods, while important for minimizing recall bias, may also mean that costs that are incurred on an infrequent basis might have been missed (such as inpatient care). However, our large sample size might mitigate against this error. There are also limitations to the way in which we collected household consumption expenditure and assessed ability to pay. Firstly, asking a single respondent about his or her household expenditure may be problematic, particularly for larger rural households. Secondly, our approach falls short of gold standard approaches where household members keep expenditure diaries. Thirdly, our catastrophic expenditure indicator relates the

S. Cleary et al. / Social Science & Medicine 80 (2013) 37e46

respondent’s direct costs to household expenditure, which would underestimate affordability barriers for households where more than one member incurred health care expenditure. Finally, traditional subjective approaches to defining affordability that relate point in time estimates of the percentage of household income spent on health care fail to incorporate dynamic aspects of the costs of treatment and are inadequate as definitions of catastrophic expenditures. Conclusions The most important findings of the study relate to the variation between urban and rural areas and between tracer conditions. The differences between the urban and rural areas are consistently significant in the analyses presented. Costs, particularly transport costs, are higher in the rural than urban areas while socio-economic status is lower. This translates, not unexpectedly, into a greater proportion of households in rural areas selling assets and borrowing money, and the amount borrowed is generally greater in rural areas. With respect to the tracers, although costs for obstetric care were higher in absolute terms, they are infrequent, generally their incidence is known in advance and households seeking such care had a higher socio-economic status on average than for the other tracers. As a result, fewer households seeking obstetric care borrowed money, although when they did they borrowed a larger amount on average, than households with ART and TB patients. It is of considerable concern that TB and ART patients are incurring monthly out-of-pocket expenses of an average R100 and R81 respectively, given that these expenses continue to be incurred over an extended period of time. It is also of concern that in one rural site almost 40% of households with members receiving ART and TB treatment had to borrow or sell assets to cover direct costs. Taken together, our findings therefore suggest that affordability is impacted differently across different tracers indicating that policies aimed at reducing or removing barriers to affordability will need to be designed around the particular characteristics of the condition. The problems are unlikely to be solved across tracers by a single uniform policy on affordability. The government policy on social grants has undoubtedly improved affordability of health services, as have the free health care policies. However, it is important that the free care policy is fully implemented in all provinces. In addition, it is feasible for health policy makers to reduce direct costs even further, particularly through providing services that are close to communities to minimise transport costs, especially in rural areas. Most importantly, the policy of directly observed daily TB treatment at public health facilities requires review. The Minister of Health recently announced a ‘PHC re-engineering’ initiative which could assist in this regard. Nevertheless, consistent efforts will be required to gradually reduce affordability barriers for South Africans. Acknowledgements This study is part of the REACH (Researching Equitable Access to Health care) project. This work was carried out with support from the Global Health Research Initiative (GHRI), a collaborative research funding partnership of the Canadian Institutes of Health Research, the Canadian International Development Agency, Health Canada, the International Development Research Centre, and the Public Health Agency of Canada. DM is supported by the South African Research Chairs Initiative of the Department of Science and Technology and National Research Foundation. The usual disclaimers apply. The funders had no involvement in study design; in the collection, analysis and interpretation of data; in the

45

writing of the report; and in the decision to submit the paper for publication. We thank all our REACH colleagues who contributed in so many ways to the overall research projects, and particularly Duane Blaauw and Till Bärnighausen who made extremely valuable comments on an earlier version of this paper. References Adhikari, S., Maskay, N., & Sharma, B. (2009). Paying for hospital-based care of Kalaazar in Nepal: assessing catastrophic, impoverishment and economic consequences. Health Policy and Planning, 24, 129e139. Ataguba, J., Akazili, J., & McIntyre, D. (2011). Socioeconomic-related health inequality in South Africa: evidence from General Household Surveys. International Journal for Equity in Health. Borghi, J., Ensor, T., Neupane, B., & Tiwari, S. (2006). Financial implications of skilled attendance at delivery in Nepal. Tropical Medicine and International Health, 11(2), 228e237. Chuma, J., Gilson, L., & Molyneux, C. (2007). Treatment-seeking behaviour, cost burdens and coping strategies among rural and urban households in Coastal Kenya: an equity analysis. Tropical Medicine and International Health, 12(5), 1e14. Chuma, J., Thiede, M., & Molyneux, C. (2006). Rethinking the economic costs of malaria at the household level: evidence from applying a new analytical framework in rural Kenya. Malaria Journal, 5, 76. Department of Health. (2011). National Health Act (61/2003): Policy on national health insurance. Government Gazette 34523. Pretoria: Department of Health. Goudge, J., Gilson, L., Russell, S., Gumede, T., & Mills, A. (2009a). Affordability, availability and acceptability barriers to health care for the chronically ill: longitudinal case studies from South Africa. BMC Health Services Research, 9, 75. Goudge, J., Gilson, L., Russell, S., Gumede, T., & Mills, A. (2009b). The household costs of health care in rural South Africa with free public primary care and hospital exemptions for the poor. Tropical Medicine and International Health, 14(4), 458e467. Goudge, J., Russell, S., Gilson, L., Gumede, T., Tollman, S., & Mills, A. (2009). Illnessrelated impoverishment in rural South Africa: why does social protection work for some households but not others? Journal of International Development, 21, 231e251. Kabir, M. A., Rahman, A., Salway, S., & Pryer, J. (2000). Sickness among the urban poor: a barrier to livelihood security. Journal of International Development, 12, 707e722. Levie, A., & Xu, K. (2008). Coping with out-of-pocket health payments: empirical evidence from 15 African countries. Bulletin of the World Health Organization, 86, 849e856. McCoy, D. (1996). Free health care for pregnant women and children under six in South Africa: An impact assessment. Durban: Health Systems Trust. McIntyre, D., Thiede, M., Dahlgren, G., & Whitehead, M. (2006). What are the economic consequences for households of illness and of paying for health care in low- and middle-income country contexts? Social Science & Medicine, 62, 858e865. McPake, B., Hanson, K., & Mills, A. (1993). Community financing of health care in Africa: an evaluation of the Bamako initiative. Social Science & Medicine, 36(11), 1383e1395. Nahar, S., & Costello, A. (1998). The hidden cost of ‘free’ maternity care in Dhaka, Bangladesh. Health Policy and Planning, 13(4), 417e422. Patcharanarumol, W., Mills, A., & Tangcharoensathien, V. (2009). Dealing with the cost of illness: the experience of four villages in LAO PDR. Journal of International Development, 21, 212e230. Penchansky, R. (1977). The concept of access: A definition. Hyattsville: National Health Planning Information Centre, Department of Health, Education and Welfare. Perera, M., & Gunatilleke, G. (2004). ALPS Sri Lanka case study. Colombo: Marga Institute. Prescott, N. (1999). Coping with catastrophic health shocks. Conference on social protection and poverty. Washington, DC: Inter-American Development Bank. Ranson, M. K. (2002). Reduction of catastrophic health care expenditures by a community-based health insurance scheme in Gujarat, India: current experiences and challenges. Bulletin of the World Health Organization, 80(8), 613e621. Rosen, S., Ketlhapile, M., Sanne, I., & Bachman DeSilva, M. (2007). Cost to patients of obtaining treatment for HIV/AIDS in South Africa. South African Medical Journal, 97, 524e529. Rugalema, G. It is not only the loss of labour: HIV/AIDS, loss of household assets and household livelihood in Bukoba District, Tanzania. Paper presented at the East and Southern Africa Regional Conference on Responding to HIV/AIDS: Development Needs of African Smallholder Agriculture, Harare, 8e12 June 1998. Russell, S. (1996). Ability to pay for health care: concepts and evidence. Health Policy and Planning, 11(3), 219e237. Russell, S. (2004). The economic burden of illness for households in developing countries: a review of studies focusing on malaria, tuberculosis and human immunodeficiency virus/acquired immunodeficiency syndrome. American Journal of Tropical Medicine and Hygiene, 7(Suppl. 2), 147e155.

46

S. Cleary et al. / Social Science & Medicine 80 (2013) 37e46

Russell, S., & Abdella, K. (2002). Too poor to be sick: Coping with the costs of illness in East Hararghe, Ethiopia. London: Save the Children. Sauerborn, R., Adams, A., & Hien, M. (1996). Household strategies to cope with the economic costs of illness. Social Science & Medicine, 43(3), 291e301. Save the Children. (2005). The unbearable cost of illness: Poverty, ill-health and access to healthcare e Evidence from Lindi rural district, Tanzania. London: Save the Children. Sharma, D., Bern, C., Varghese, B., Chowdhury, R., Haque, R., Ali, M., et al. (2006). The economic impact of visceral leishmaniasis on households in Bangladesh. Tropical Medicine and International Health, 11(5), 757e764. Tibaijuka, A. K. (1997). AIDS and economic welfare in peasant agriculture: case studies from Kagabiro Village, Kagera Region, Tanzania. World Development, 25(6), 963e975. Van Damme, W., Van Leemput, L., Por, I., Hardeman, W., & Meessen, B. (2004). Outof-pocket health expenditure and debt in poor households: evidence from Cambodia. Tropical Medicine and International Health, 9(2), 273e280. Van Doorslaer, E., O’Donnell, O., Rannan-Eliya, R., Somanathan, A., Adhikari, S., Garg, C., et al. (2006). Effect of payments for health care on poverty estimates in

11 countries in Asia: an analysis of household survey data. Lancet, 368, 1357e1364. Wagstaff, A., & van Doorslaer, E. (2003). Catastrophe and impoverishment in paying for health care: with applications to Vietnam 1993e1998. Health Economics, 12(11), 921e934. Wilkes, A., Hao, Y., Bloom, G., & Xingyuan, G. (1997). Coping with the costs of severe illness in rural China. IDS Working Paper 58. Wyss, K., Hutton, G., & N’Diekhor, Y. (2004). Costs attributable to AIDS at household level in Chad. AIDS Care, 16(7), 808e816. Xu, K., Evans, D., Carrin, G., & Aguilar-Rivera, A. (2005). Designing health financing systems to reduce catastrophic health expenditure (Technical Briefs for PolicyMakers Number 2. WHO/EIP/HSF/PB/05.02). Geneva: World Health Organization. Xu, K., Evans, D., Kadama, P., Nabyonga, J., Ogwal, P., Nabukhonzo, P., et al. (2006). Understanding the impact of eliminating user fees: utilization and catastrophic health expenditures in Uganda. Social Science & Medicine, 62, 866e876. Xu, K., Evans, D., Kawabata, K., Zeramdini, R., Klavus, J., & Murray, C. (2003). Household catastrophic health expenditure: a multicountry analysis. Lancet, 362(362), 111e117.