Measuring catastrophic health expenditure and impoverishment
The measurement of catastrophic health expenditure and impoverishment has been discussed extensively in the literature [
4,
32‐
36]. Regardless of the method, however, a choice has to be made regarding the threshold to use in determining catastrophic health expenditure and a choice in defining the household resources used to pay for healthcare [
5]. Whilst the choice of threshold is arbitrary and has typically varied between 10 and 40%, there have been two commonly used methods employed in defining household resources and measuring catastrophic health expenditure in the literature.
The first method by Wagstaff and van Doorslaer [
32] defines health expenditure as catastrophic when it exceeds a certain threshold of total expenditure or household income [
32,
33]. Critics of this method have argued that they underestimate the financial impact of health costs among poorer households due to the use of uniform thresholds [
34,
36,
37]. The second method by the WHO (further referred to as the ‘WHO standard method’) defines health expenditure as catastrophic when it exceeds a certain threshold of capacity to pay [
4,
38]. There however are some reservations with this method, which are related to how exactly subsistence expenditure is measured [
32,
35,
36] and how relevant the initial estimate of the equivalence scale is [
39]. More recently, some authors have argued that in order to ensure fair and ethical measures of catastrophic health expenditure, the threshold applied in measuring it should be a function of the income distribution [
34,
37]. Ataguba proposes a method that uses a threshold that varies with income when estimating catastrophic health expenditure [
34]. This method (further referred to as the ‘Ataguba method’) has recently been applied in measuring catastrophic health expenditure in Swaziland and Uganda, and is useful in countries with high inequalities [
40,
41]. To check the robustness of our results, we employ both the ‘WHO standard method’ and the ‘Ataguba method’ in estimating catastrophic health expenditure. In both methods the incidence of catastrophic health expenditure is defined as the proportion of patients attending the diabetes clinics, whose healthcare expenditure due to diabetes is catastrophic. The steps followed in calculating catastrophic health expenditure and the extent of impoverishment are provided below, with each of the two methods being discussed in turn.
Construction of statistical variables
The computational steps used to generate the variables used in the method by WHO [
4] are shown below. A detailed description of the steps followed in constructing these variables is provided elsewhere [
4].
Step 1: Generate the food expenditure share (FES)
$$ {FES}_h={FEH}_h/{THE}_h $$
Step 2: Generate the household equivalent size (HES) as follows:
$$ {HES}_h= hh\_{size}^{\beta } $$
Where hh_size is the household size, and the coefficient β is the value of an equivalence scale. Our study makes use of β = 0.56 which was estimated from a regression equation based on 59 countries [
38]. A study by Koch has queried the applicability of the estimate because most of the data used to calculate it were more than 2 decades old [
39]. The use of a range of scales is therefore recommended [
39]. However in the case for South Africa, Koch finds that although the scale has changed over the years, the choice of scale does not really affect the average incidence of catastrophic health expenditure [
39]. For the purposes of this study and consistent with recent studies [
29,
42], we make use of the commonly applied household scale multiplier of 0.56
Step 3: The equivalent food expenditure is obtained as follows:
$$ {EFE}_h={FEH}_h/{HES}_h $$
Step 4: Identify the FES at the 45th and 55th percentile across the entire sample and name them FES45 and FES55.
Step 5: Calculate the average of food expenditures of the households that lie within FES45 and FES55 to obtain the poverty line (PL).
Step 6: Subsistence expenditure for each household is then calculated as follows.
$$ {SE}_h= PL\ast {HES}_h $$
Step 7: Generate the household’s capacity to pay (CTP
h) which is defined as the household’s non-subsistence expenditure (SE
h) as follows:
$$ {CTP}_h={THE}_h-{SE}_h\kern0.5em \mathrm{if}\kern0.5em {FEH}_h>={SE}_h $$
$$ {CTP}_h={THE}_h-{FEH}_h\kern0.5em \mathrm{if}\kern0.5em {FEH}_h< SE $$
Step 8: Health expenditure is defined as catastrophic if OOP health expenditure exceeded a certain threshold (e.g. 10%) of the household’s CTP
h.
$$ {cata}_h=1\kern0.5em \mathrm{if}\kern0.5em {OOPHE}_h/{CTP}_h>=10\% $$
$$ {cata}_h=0\kern0.5em \mathrm{if}\kern0.5em {OOPHE}_h/{CTP}_h<10\% $$
There is a lack of consensus on the appropriate threshold to use when measuring catastrophic health expenditure. Lower thresholds are typically used in the total expenditure method and higher thresholds in the capacity to pay method [
5]. Consistent with other studies the sensitivity of the analysis to various thresholds was tested [
29,
43,
44] . Since the selection of threshold is a normative and somewhat arbitrary choice we present results using thresholds set at 10, 20, 30 and 40% and leave it to the reader to determine their selection.
Step 9: A household is defined as poor if its THE
h was smaller than its SE
h and non-poor when THE
h was greater than or equal to SE
h.
$$ {poor}_h=1\kern0.5em \mathrm{if}\kern0.5em {THE}_h<{SE}_h $$
$$ {poor}_h=0\kern0.5em \mathrm{if}\kern0.5em {THE}_h>={SE}_h $$
Step 10: A non-poor household was considered impoverished by healthcare payments once it became poor after paying for healthcare
$$ {impov}_h=1\kern0.5em \mathrm{if}\kern0.5em {THE}_h-{OOPHE}_h<{SE}_h $$
$$ {impov}_h=0\kern0.5em \mathrm{if}\kern0.5em {THE}_h-{OOPHE}_h>{SE}_h $$
In order to check the robustness of our results and due to the limitations of the WHO standard method outlined above, we also estimated catastrophic health expenditure using the method proposed by Ataguba et al. [
34]. Computational steps for the method by Ataguba et al. [
34] are shown below.
Step 1: Estimate the rank dependent threshold Z’
cat$$ {Z}_{cat}^{\hbox{'}}=\gamma {\left(1-\rho \right)}^{\left(\gamma -1\right)}\ast {Z}_{cat} $$
where ρ is the household’s percentile generated when households are ordered according to income, Z
cat is the initial threshold (an initial threshold of 10% is used in our paper), ƴ is a parameter of aversion to inequality. Following Ataguba et al. [
34], we use a value of 0.8. However, for illustrative purposes we also present results when ƴ = 1. This is the case when Z
cat does not change across the income distribution and is similar to applying the method by Wagstaff and van Doorslaer [
32].
Step 2: Estimate the rank dependent overshoot which shows the extent to which health cost as a fraction of total household cost exceeds the threshold
$$ {OS}_h^{\hbox{'}}=\frac{OOPHE_h}{THE_h}-{Z}_{cat}^{\hbox{'}}\kern0.5em \mathrm{if}\kern0.5em \frac{OOPHE_h}{THE_h}>{Z}_{cat}^{\hbox{'}} $$
$$ {OS}_h^{\hbox{'}}=0\kern0.5em \mathrm{if}\kern0.5em \mathrm{otherwise} $$
Step 3: Estimate the rank dependent catastrophic health expenditure head count ratio which shows the proportion of households that incur catastrophic health expenditure. Where E = 1 when OS
h > 0 and 0 when otherwise.
$$ {HC}_h^{\hbox{'}}=\frac{1}{N}\left(\sum \limits_{h=1}^N{E}_h^{\hbox{'}}\right)={\mu}_h^{\hbox{'}} $$
-
Step 4: Using a poverty line, estimate the pre-health payment poverty head count ratio. Our study makes use of the 2017 lower bound poverty line of R758. This is a poverty line estimate generated by Statistics South Africa which takes into account both basic food and other basic needs [
45] and is the preferred threshold in policy making.
$$ {H}_{pov}^{pre}=\frac{1}{N}\left(\sum \limits_{h=1}^N{P}_h^{pre}\right)={\mu}_{p^{pre}} $$
Where \( {P}_h^{pre}=1 \) if adult equivalent household expenditure THEh < poverty − line and \( {P}_h^{pre}=0 \) if otherwise.
Step 5: Estimate the post-health payment poverty head count ratio.
$$ {H}_{pov}^{post}=\frac{1}{N}\left(\sum \limits_{h=1}^N{P}_h^{post}\right)={\mu}_{p^{post}} $$
Where \( {P}_h^{post}=1 \) if THEh − OOPHEh < poverty − line and \( {P}_h^{post}=0 \) if otherwise.
Step 6: Estimate the impoverishing impact of OOP health expenditure, i.e. the difference between the pre-payment and post-payment indices.
$$ {PI}_{H=}{H}_{pov}^{post}-{H}_{pov}^{pre} $$
A detailed description of these steps is provided elsewhere [
34]. In our study, total monthly expenditure is used as a proxy for income.
Data were also collected on the time spent visiting the hospital for diabetes care and work days missed due to diabetes. Patients were asked to report in hours and minutes, how much time it took them travelling to the hospital, waiting to consult the doctor, during consultation and waiting for medication. For patients who reported being employed, the number of work days missed due to diabetes over the last 30 days, were collected using a categorical variable that took on a value of 1 when respondents took half a day, a value of 2 when respondents took 1 to 4 days, a value of 3 when respondents took 5 to 10 days and a value of 4 when respondents took more than 10 days. A continuous variable was then created by taking the mid-point estimate of each category.
The indirect costs due to productivity loss were estimated for patients who reported being employed by using the monetary value of time spent seeking care and the monetary value of days missed from work. Hourly wage rate was estimated by using respondent reported monthly income and assuming patients worked 20 days a month and 8 h a day. We then follow the method applied by Oloniniyi et al. to estimate productivity losses [
46]. The hourly wage rate was multiplied by the total hours spent seeking care and time taken off work due to diabetes over the last 30 days.
Inequalities in catastrophic health expenditure and impoverishment
Our study makes use of the concentration index (CI) to measure socio-economic inequalities in catastrophic health expenditure and impoverishment amongst the diabetes patients. The CI ranges between − 1 and + 1 and is measured as twice the covariance of the catastrophic health expenditure/impoverishment variables and the ranking of the living standards variable r all divided by the mean of the catastrophic health expenditure or impoverishment variables (
μ):
$$ CI=\frac{2}{\mu}\mathit{\operatorname{cov}}\left(h,r\right) $$
(1)
Our study makes use of multiple correspondence analysis (MCA) to generate the wealth index which is our living standards variable. Although there are various methods that can be used for the construction of the asset index MCA is chosen because it is the preferred technique for categorical variables [
47]. Based on items included in the questionnaire, a commonly used set of living conditions and ownership of household assets were included in constructing the wealth index. Ten household conditions and assets were considered in the analysis. The full list is as follows: housing type, water and sanitation services, ownership of a television, refrigerator, 4 plate stove, radio, cell phone, computer and car. The wealth index was later categorised into wealth quintiles. The wealth index was then applied in generating our CIs.
A negative CI means that catastrophic health expenditure or impoverishment is concentrated amongst the poor whilst a positive value means it is concentrated amongst the rich. The CI takes on a value of zero when there is no socio-economic inequality meaning catastrophic health expenditure or impoverishment variable is equally distributed across the sample. Since our variables are binary this study makes use of the Erreygers corrected CI.
$$ E(h)=\frac{4\mu }{b-a} CI $$
(2)
Where
μ is the mean of the catastrophic health expenditure or impoverishment variables, CI is the concentration index, b is the maximum value of the variable (in this case 1), a is the minimum value of the variable (in this case 0). We make use of STATA’s
conidex command [
48].
Determinants of catastrophic health expenditure and impoverishment
We used logistic regression to analyse the association between socio-demographic variables and catastrophic health expenditure and impoverishment. In assessing these associations, our selection of socio-demographic variables was guided by the literature [
4,
43,
49]. The individual and household variables included in our analysis are age, gender, race, and marital status, having children, education, employment status, household size and index quintile. Age was measured in years and was included as a continuous variable. Gender was included as a binary variable taking on the values 1 – male, 2 – female. Race was also included as a binary variable with 1 – African and 2 – non-African (white, coloured, Indian/Asian). Marital status was included as follows: 1 – married/living with a partner, 2 – single. Respondents were asked if they had any children, this was included as a binary variable taking on the value of 1 when the respondent had children and 0 when the respondent did not have any children. Education was categorised as 1 – primary, 2 – secondary and 3 – tertiary education. Employment status was included as 0 – unemployed and 1 – employed. The size of the household was included as a categorical variable that took on the following values; 0–1 to 4 household members, 1–5+ household members. An outline and description of these variables is provided in an additional file (see Additional file
1).
The statistical analysis was conducted using STATA version 13. In order to allow for the skewed distributions, all household income and expenditure related data are presented as means (standard deviations) and medians (percentiles). Our study reports proportions for categorical variables.