Social class related inequalities in household health expenditure and economic burden: evidence from Kerala, south India

This paper has the following two broad objectives:

In a context where health insurance is almost non-existent and the population depends on private health care providers to a large extent, households' out-of-pocket health expenditure can be a good reflection of their health care consumption and their effective access to quality health care. By analysing households' per capita health expenditure while taking into consideration their burden of health care need, we were able to identify the influence of social stratification at equal levels of need. To do this, we used a multivariate linear regression framework with the following specification:

In Kerala's rural society, households' education and income levels are largely determined by their socio-economic status, and social class (caste) and landholdings together can capture the socio-economic status of the household to a great extent. In informal rural economies, it is not easy to get reliable income data at the household level. A household's per capita consumption expenditure or asset holding (including landholding) is often used as a second-best proxy to capture its economic status [31]. We analysed households' economic burden of health care and variations across caste groups by identifying households that spend relatively large amounts on health care and investigating their sources of funds for out-of-pocket health care expenditures. The analysis was mostly done at the household level, except when we estimated per capita health expenditure by type of illness episode.

The context of our study was Kottathara Panchayat (panchayat being the lowest layer of the decentralised government) in the Wayanad District of India's southern state of Kerala. Wayanad has a large tribal population [14, 32]. Kottathara, with a population of 16,613, is a multi-religious, multi-caste/tribe setting with 3% SC and 28% ST, predominantly Paniya and Kurichiar tribes [14]. Kottathara Panchayat has one primary health centre and four sub-centres. The district hospital in Mananthavady is 30 kilometres away, but private health care facilities, including clinics, nursing homes and hospitals, are located in Kalpetta, about 10 kilometres away.

The major source of data used for the analysis was a 12-month long panel survey conducted between October 2003 and September 2004. The panel survey was preceded by a baseline survey in 2003 that covered the entire population of Kottathara Panchayat. The panel survey of 2,925 individuals from 543 sample households was designed to collect detailed information on health care consumption for all reported episodes. In comparison to single-visit cross sectional surveys with long recall periods, data from a year-long panel survey provide more accurate information on the incidence of illness episodes and on out-of-pocket health expenditure with a minimum level of recall bias. Details were recorded on consumption of out-patient (OP) visits, in-patient hospitalisations and expenses incurred. The total number of uncensored episodes (i.e., episodes for which we have full information) reported by the sample population was 4,949, consisting of 4,408 acute, 430 chronic and 111 pre-natal, delivery and preventive care episodes.

The sample households' socio-economic, demographic and health status characteristics are presented in Table 1. The table shows that caste is a good stratification variable for classifying households by social, economic and demographic characteristics. Compared to the FC, the socially backward castes (viz. Paniya, Other ST/SC, and OBC) have a higher share of female-headed households, a characteristic that generally adds to a household's vulnerability. Head-of-household's education clearly shows a caste gradient. The distribution of land, the most crucial economic asset, shows that the majority of the landless households are Paniya. The occupation profile also supports the Paniya households' status as the poorest. The households' annual per capita consumption expenditure (collected in the baseline survey) further confirms the caste gradient.

Like per capita consumption expenditure, per capita health expenditure across caste groups reaffirms the caste hierarchy (Table 2). The FC households show the highest per capita health expenditure, followed by the OBC, Other ST/SC and Paniya households, which show the lowest figure. Our need-based classification of households is validated by the per capita health expenditure of households with different levels of health care need. The per capita health expenditure is highest for very-high-need households, followed by high-need households and then low-need households.

Since it is to be expected that households' health care need, health care utilisation rate and choice of health care providers will vary, not all differences in per capita health expenditure across caste groups necessarily reflect inequality. For example, not all high-need households have a high health care utilisation rate, as they may face barriers in accessing health care. Further, a household's choice of health care providers may determine its health care expenditure, as OP visits to government and private informal health care providers cost less than those to formal private providers. If caste differences in per capita health expenditure were nothing unusual and did not necessarily reflect inequality faced by different castes in accessing necessary health care, one would expect to see similar patterns of caste differences among low-need, high-need and very-high-need households. Similarly, one would expect a similar pattern of caste differences for expenditure on acute and chronic illnesses. Figure 1 presents the per capita health expenditures for low-, high- and very-high-need households by caste. Although the gradual increase in per capita health expenditure is clearly evident, within each caste, for each group of households with comparable levels of health care need, the per capita health expenditure for each caste category does not show a similar rise when moving from low-need to high-need and from high-need to very-high-need. In the latter case, a steady rise in per capita health expenditure is clearly evident for FC households, but not for other caste groups, especially the Paniya. Irrespective of the level of health care need, a Paniya household, on average, spends less than 8% of what a need-comparable FC household spends per person. Further, it can be observed that a very-high-need household belonging to an Other ST/SC or OBC group spends a much lower percentage of what a need-comparable FC household spends on health care. This indicates that very-high-need households belonging to Other ST/SC and OBC households are not able to spend as much as they should on health care.

Figure 2 presents per capita expenditure on acute and chronic episodes by caste. For the population as a whole, per capita expenditure on acute episodes is slightly higher than expenditure on chronic episodes. The higher confidence interval of the latter compared to the former is an indication that households vary considerably from one to another in their per capita expenditure on chronic episodes compared to their expenditure on acute episodes (Table 2). Although the magnitude of difference varies, all castes show higher per capita expenditure on acute episodes than chronic episodes, but the difference is striking for the Paniya and Other ST/SC. Such a striking difference is not observed in the OBC class, which is socio-economically closer to the FC. This clearly points to a huge unmet need for chronic health care in the Paniya and Other ST/SC groups.

A closer look at Table 1 and Figures 1 and 2 therefore suggests that the observed overall caste gradient in health expenditure (Rs 43, Rs 251, Rs 501, Rs 801) is a product of compositional effect (composition of low-, high- and very-high-need households in each caste group) and a pure caste effect. Caste groups differ from each other in terms of their composition of low-, high- and very-high-need households (Table 1). Sorting household health expenditure by level and type of health care need shows that the caste gradient is stronger for high-need and very-high-need households (Figure 1) and for chronic health care need (Figure 2).

The greater inequality observed among high-need and very-high-need households probably indicates a violation of the vertical equity principle. We examined this issue more closely using a multivariate linear regression, as outlined in the Framework sub-section. The independent variables were household's caste, household's landholding, household's health care need, and other confounders such as sex of the household head, number of government and private OP visits, and number of hospitalisations. Household's landholding was included to capture the effects of a household's economic status on its out-of-pocket health expenditure that are not captured by the caste. Sex of the household head was included to examine whether having a female head places a household in an unfavourable position to allocate money to health care. It has been observed in the Indian context that female-headed households are more vulnerable to poverty due to gender bias in employment opportunities [35]. The numbers of OP visits and hospitalisations were included as confounders because per capita health expenditure is expected to be higher for households with a greater number of private OP visits and hospitalisations. Because a household's utilisation of health care is also expected to be dependent on its health care need, including both need and utilisation may put the model at risk of multicollinearity. However, we included the utilisation variables that capture both the number of OP visits and hospitalisation as well as the type of institution utilised (government/private) in order to capture the price effects of health care utilisation that have a strong bearing on the level of per capita health expenditure. To highlight the extent to which inclusion of utilisation variables made a difference in capturing caste-based inequality in health expenditure, we present both models: one without the utilisation variables (Model 1) and the other with them (Model 2). Because per capita health expenditure exhibited a positively skewed distribution, we made a log transformation to remove the skewness after adding Re 1 to the value of each household's per capita health expenditure.

Table 3 presents the results of the multivariate regression. Using FC as a reference caste, Paniya and Other ST/SC show negative coefficients that are significant in both models, indicating lower per capita health expenditure. It is important to observe that Other ST/SC, which is closer to OBC and FC in a number of characteristics, behaves like the Paniya. In both models, with reference to low-need households, both high-need and very-high-need households show significant increase in per capita expenditure. Neither the household's landholdings nor the sex of the household head is significant in the models, which is an indication that they do not play a role in lowering health expenditure. In Model 2, more OP visits to government or informal care providers do not significantly increase per capita health expenditure, but OP visits to private facilities result in a significant increase. A household's episodes of hospitalisation, whether public or private, are associated with a significant increase in its per capita health expenditure. In a nutshell, with reference to FC, the Paniya and Other ST/SC households spend less on health care per capita, even when we take into account the effects of the household's health care need and its volume of government and private health care utilisation.

To shed further light on caste-related inequality in per capita health expenditure, we used the estimated regression equation of Model 2 to predict average values of per capita health expenditure incurred by the four caste groups for each level of health care need (viz. low, high and very high). Because we used household's landholdings, sex of the household head, number of OP visits and hospitalisations as confounders in the regression, we estimated the predicted values of health expenditure at the mean values of the confounding variables. The predicted values, presented in Figure 3, highlight certain aspects of inequality that were uncovered in Figure 1. First, moving from low-need to very-high-need households, both FC and Other ST/SC caste groups are able to systematically increase per capita monetary allocation on health care - a pattern which is not observed for the Paniya and the OBC caste groups. Second, there is no difference between Paniya low-need and high-need households in per capita health expenditure, while the very-high-need Paniya households' expenditure is marginally higher.

Table 4 presents the cut-off level of per capita health expenditure for identifying high-spending households, the percentage of high-spending households and the distribution of high-spending households across low-, high- and very-high-need categories. Looking at the cut-off levels of health expenditure for the four caste groups at three different values of k, two points are clear. First, at any given value of k, the health expenditure cut-off levels follow the caste hierarchy, that is, the cut-off is lowest for the Paniya and highest for the FC. Second, for any given caste group, the health expenditure cut-off level is higher for higher values of k.

Although the Paniya caste group spends much less on health care per capita compared to households from other caste groups, the distribution of per capita health expenditure within the Paniya caste shows that a higher percentage of Paniya households incur large expenditure on health care. However, Other ST/SC shows the lowest percentage of high-spending households (except at k = 0.5, where its expenditure is marginally higher than that of the FC). Furthermore, most of the high-spending households belong to either the high-need or very-high-need categories. At all three values of k, more than 85% of the households belong to the high-need or very-high-need categories. Though this pattern is observed in the Other ST/SC, OBC and FC caste groups, the Paniya group seems to be different, as 32%-37% of high-spending Paniya households have low health care need. In other words, a Paniya household may spend relatively more on health care irrespective of its level of health care need - a pattern not observed in other caste groups.

The choice of health care provider seems to affect a household's likelihood of impoverishment due to out-of-pocket health expenditure. Whereas the number of government (or informal) OP visits has no significant effect on the level of a household's out-of-pocket health expenditure, private OP visits and hospitalisations (government or private) have significant positive effects on the level of out-of-pocket health expenditure (Table 3). The higher values of the coefficient of hospitalisation at government or private facilities clearly indicate that a household is most likely to fall into the high-spending category if it experiences an episode of hospitalisation, irrespective of the type of hospital - government or private - utilised. This hypothesis is also corroborated by the distribution of OP visits and hospitalisations by type of health care provider for households that fall under the category of high-spending and for those that do not. Among the households that did not incur high out-of-pocket health care expenditure, 42% of OP visits and 69% of hospitalisations were in private facilities. However, high-spending households showed much higher percentages of OP visits (55%) and hospitalisations (92%) at private facilities. In other words, households with higher numbers of OP visits to private providers and hospitalisations at private facilities are more likely to be high-spending.

The distribution of households' out-of-pocket health care expenditure by sources of financing highlights an important dimension of the hardship faced by different caste groups in mobilising monetary resources to meet health care expenditure (Table 5). Available cash or current income contributes 65% of households' total out-of-pocket health expenditure, followed by loans from friends, self-help groups or money lenders (24%), savings including selling of food stocks (8%), and donations from friends or relatives (3%). Two questions seem important in this context: (a) Is the distribution of total out-of-pocket health care expenditure by sources of financing similar for all caste groups? (b) Is the pattern of financing health care expenditure similar for OP visits and for hospitalisations? Table 5 shows that Paniya and OBC households depend more on loans and donations for meeting total health expenditure than do FC households. It is worth noting that in spite of Paniya households' having very low per capita health expenditure, 31% of their out-of-pocket health expenditure is financed by loans and donations. Though Other ST/SC households finance more than three-quarters (77%) of their total out-of-pocket health expenditure from available cash and savings, this probably entails compromise in their consumption of other basic necessities. Despite spending only one-third of total out-of-pocket expenditure on hospitalisation, the households depend on loans and donations more for hospitalisation than for OP visits. Whereas only 21% of households' out-of-pocket expenditure on OP visits comes from loans (19%) and donations (2%), 43% of their expenditure on hospitalisation comes from loans (40%) and donations (3%). Patterns of financing hospitalisation expenditure reveal greater vulnerability in all backward castes as compared to FC. The percentages of hospitalisation expenditure financed by loans and donations from friends and/or relatives for the Paniya, Other ST/SC, OBC and FC households are 48, 45, 45 and 42 respectively. There is no evidence to suggest that backward caste groups depend more on loans to finance expenditure on OP visits compared to the FC group. Despite Paniya households' having a very low expenditure level, one-quarter of their total household out-of-pocket expenditure on OP visits is financed by loans (23%) and donations (2%). This is a clear indication of their vulnerability in meeting out-of-pocket health expenditure. It is interesting to observe that Other ST/SC households mobilise a large portion of their health expenditure (especially hospitalisation expenditure) from donations. This is an indication of a much higher level of social networking among these tribe and caste groups.