Looking at various socio-economic indicators reveals a caste hierarchy with FC at the top of the ladder, followed by OBC and Other ST/SC, and Paniya at the bottom. The per capita health expenditures reported by these caste groups accord with this caste hierarchy, with FC and Paniya households' health care expenditure being very high and very low, respectively. In its last two health rounds, India's National Sample Survey Organisation observed people's high dependence on private providers, which has increased over the years, even among the poor and marginalised [17–19]. In a context where households depend on private health care providers to a great extent for meeting their health care needs, differences in per capita health expenditure indicate inequality in access to quality health care. Moreover, the fact that poor households spend less on health care does not mean their indirect costs of illness are low. Even if poor households spend significantly less on health care, they incur a higher proportion of health-related loss of income than do other non-poor groups .
Our analysis shows that caste differences in per capita health expenditure are not similar for households with different levels of health care need, and there is an indication that very-high-need households belonging to the Paniya, Other ST/SC and OBC caste groups do not have the means to cover what they are required to spend for health care. On average, households spend slightly more annually on acute episodes than on chronic episodes, but unlike their expenditure on acute episodes, households in each caste group vary considerably among themselves in per capita expenditure on chronic episodes. Compared to the FC and OBC caste groups, the acute-chronic difference in per capita expenditure is greater among the Other ST/SC and Paniya caste groups. From the per capita expenditure figures, there appears to be a huge unmet need for chronic health care in Paniya and Other ST/SC households, though Paniya households report less need for chronic health care. Most studies on health financing in the Indian context have recommended introducing or scaling up social health insurance as the only remedy for improving access to health care for the poor and marginalised [37, 38]. Our findings question the effectiveness of these remedial measures, as many of the existing health insurance packages pay very little attention to chronic health care need, especially in the elderly population. It has been pointed out that very little analysis was done before various social health insurance schemes were proposed as remedial measures to solve problems of health care access for the poor in the Indian context . Many of these recommendations either ignored or downplayed the immediate need to strengthen supply side factors, i.e., health care infrastructure and manpower, including in the government health sector. The lower need for chronic health care expressed by the Paniya caste groups also calls for further scrutiny.
The multivariate analysis shows that Paniya and Other ST/SC households spend significantly less on health per capita than do FC households, even taking into account the effects of levels of health care need, household's landholding, vulnerability of female-headed households, and volume and type of health care utilisation. The insignificance of the variable 'landholding' is an indication that the social variable 'caste' adequately captures the effect of socio-economic status on per capita household expenditure. Though higher utilisation of government and private informal OP services makes no significant positive impact on per capita health expenditure, private OP visits and hospitalisations (government or private) do produce a significant increase. This is in line with an earlier study in Kerala which found that, even in government hospitals, households spent significant amounts of money on buying services outside the hospital .
It has generally been observed that health care payments and financial burden (payment share) increase with an increase in ability to pay . Therefore, owing to the steep income gradient, FC households would be expected to have higher health care payments and a greater financial burden than other caste groups. Studies have shown that it is not only the better-off but also poor households that can be at the risk of large health care payments [40, 41]. Our analysis also points to the vulnerability of backward caste groups, especially the Paniya, when exposed to relatively high health expenditure. While the per capita health expenditure of the Paniya caste group is lowest, a higher proportion of Paniya households incur relatively large expenditure compared to the other caste groups. This is an indication that not all Paniya households are in a position to take advantage of free or nearly free public health care. The Paniya's low utilisation of government health care suggests that unless steps are taken to remove the social barriers to health care access faced by marginalised populations, bringing them under social insurance may not improve their access to health care. In contrast to the Other ST/SC, OBC and FC groups, in which households that spend relatively large amounts on health care have either high or very high needs, a large percentage of Paniya households with low health care need spend relatively large amounts on health care. This has an important policy implication, which is that while high-need and very-high-need households belonging to Other ST/SC, OBC and FC caste groups need financial protection, all Paniya households need universal protection.
In comparison to the FC group, the Paniya and OBC caste groups depend more on loans and donations for meeting total health expenditure. The households' patterns of financing expenditure for OP visits and hospitalisations were found to be different. It has been found in the Indian context that out-patient care is more impoverishing than in-patient care in urban and rural areas . Other empirical findings suggest that high health expenditure for a household is not usually the result of one single disastrous event such as hospitalisation, but rather a series of events [42, 43]. Our analysis in Kerala showed that, per year, a household's average expenditure on hospitalisation was less than one-third of what it spent on OP visits. This is significantly higher than what has been observed elsewhere in India [43, 44]. It has been found in other parts of India that low expenditure on hospitalisation is due to low utilisation of hospital care. The concentration of hospitals mostly in urban areas and district headquarters is a barrier to access for rural populations, resulting ultimately in their low utilisation of hospital care . However, a larger hospitalisation component in total health care expenditure in Kerala is not unusual, since Kerala also reports a much higher incidence of hospitalisation compared to other Indian states [17, 18]. Contrary to other studies [42, 43], we found that hospitalisation expenditure has more impoverishing effects on households, as evidenced by our analysis of the distribution of households' out-of-pocket expenditure by sources of finance. The households depend more on loans and donations to meet hospitalisation expenditure than they do to meet expenditure on OP visits. The patterns of financing hospitalisation expenditure clearly show the Paniya, Other ST/SC and OBC households to be more vulnerable than FC households. This is most likely due to the unpredictable nature and large amount of the hospitalisation expenses. Other studies have found that households' preference for private health care, economic status, utilisation of modern medical care, presence of ill elderly member(s), presence of member(s) with chronic illness and incidence of hospitalisation are key determinants of high health expenditure [40, 41].