The data for the paper are derived from the unit records of the “Social Consumption: Health” survey (71st round) conducted by the NSS Office, which provides information at national and state (provincial) levels related to morbidity, hospitalization, reported causes for illness and hospitalization, the cost of care, the type of service provider, and care in pregnancy and for the elderly [15].
In India, the NSS Office, a government institution, organizes annual surveys on different areas of consumption, poverty, and employment, and once every 10 years, it undertakes a survey called Social Consumption: Health [15].
This survey uses a two-stage stratified sampling approach, with the first sampling units composed of village and urban blocks and the second stage composed of households. Data collection was performed from January 2014 to June 2014 in two segments of 3 months each. A total of 65,932 households (rural: 36480, urban: 29452) were surveyed for the entire Indian Union, which included a total of 333,104 individuals (rural: 189573, urban: 143531; male: 168697 female: 164407).
Morbidity as reported by the household is captured in response to three questions: 1) Were you ill in the past 15 days? 2) Were you hospitalized in the past 365 days? 3) If so, what was the cause? The reported cause is then attributed to the nearest fit of one of 60 diagnostic categories on the basis of a medical diagnosis conveyed to the study team by the respondent, or failing this approach, the main symptom was used. This survey also probed whether each household and individual had any insurance coverage, including the type of insurance coverage. For each illness or hospitalization, the study team noted the choice of provider and reported costs of care, disaggregated into medical and non-medical costs by component. The sample consisted of 57,456 hospitalization episodes (55,026 episodes excluding death), of which approximately 10,168 episodes had insurance coverage.
Those with insurance coverage were asked about the type of scheme, and their responses were categorized into four groups. One is the PFHIs, which include both the social insurance schemes that began in the 1950s and the post-2005 government-sponsored schemes that target the poorer socio-economic groups for coverage. The second category is composed of individual households that voluntarily purchase their own insurance coverage from private insurance firms or “private insurance”. The third category is employer-provided insurance coverage for employees, and a fourth category of “others” is associated with special schemes such as Yashwasini, which is privately initiated and linked to cooperatives. Because the last is very small, we have not shown it in the calculations (only 0.1%). Of these four types of insurance, only PFHI and the post-2005 types of schemes focus on providing financial protection for the poor and are being considered for scaling up as the national strategy. Therefore, we have limited most of our analysis to only this first category, and within that, we separately analysed the lower three quintiles since those covered by the two social insurance schemes would largely or entirely belong to the upper two quintiles.
This study examines various dimensions of equity, defined as unequal insurance coverage, hospital rates, and/or financial protection based on sex, social group, economic quintile, or rural/urban residence. This study measures financial protection in terms of seven indicators. These are 1) the mean OOPE per hospitalization episode; 2) the median OOPE per hospitalization episode; 3) the proportion of hospitalization episodes where OOPE was less than Rs 500; 4) the proportion of hospitalization episodes where OOPE was less than Rs 1000; 5) the incidence of households experiencing catastrophic health expenditures, which is the proportion of households whose costs of hospitalization were beyond a threshold, defined as 10% of annual consumption expenditure (CHE-10); 6) the incidence of catastrophic health expenditure, which is the proportion of households whose costs of hospitalization exceeded a threshold, defined as 25% of annual consumption expenditure (CHE-25); and 7) the impoverishment related to hospitalization costs.
OOPE for hospitalization was calculated per episode, including transportation, and reimbursements were subtracted. Only 2% of those hospitalized received reimbursements.
The two indicators of incidence of hospitalization for which OOPE was below Rs 500 or Rs 1000 are introduced since the stated objective of most government-funded insurance schemes is to provide cashless services for hospitalization care, not merely a reduction in OOPE. Since some incidental expenses may be counted, thresholds of Rs 500 and Rs 1000 for OOPE enabled better measurement of the proportion of hospitalizations in which this objective of cashless services was achieved.
The methodology discussed by Wagstaff and Doorslaer [16] to assess catastrophic health expenditure (CHE) for healthcare was applied in this study. A household OOPE on hospitalization in the preceding year is considered an incidence of CHE when the payment exceeds the 10% (CHE-10) and 25% (CHE-25) thresholds of the household annual total household consumption expenditure. This household annual total consumption expenditure represented 12 times the UMPCE (usual monthly per capita consumption expenditure). UMPCE is measured by a consumption survey using a methodology standardized by NSSO and accepted widely in India as a reliable proxy for income levels. We chose to study OOPE and CHE only in regard to hospitalization because this is what approximately all government insurance covers.
To estimate impoverishment due to hospitalization costs, this study used the threshold recommended by the Government of India Planning Commission Report of June 2016 for measuring the poverty line (per capita) as Rs 972 in rural and Rs 1407 in urban areas [17].. Households whose MPCE was initially above this threshold but later fell below it after incurring hospital expenditures were considered to have experienced ‘impoverishment due to hospitalization costs’. Households whose MPCE was initially itself below the poverty line were considered to have experienced ‘deepening of poverty due to hospitalization costs’.
The levels of financial protection are measured and presented for four situations: a) care is from the private sector, and there is no insurance coverage; b) care is from the private sector, and there is PFHI coverage; c) care is from a tax-funded public provider, which affords financial protection through subsidies, but there is no insurance coverage; and d) care is from the tax-funded public provider and is complemented by PFHI coverage. This allows us to compare the reduction in OOPE and the protection against CHE that are provided in each of these contexts.
Other than the above, another indicator of financial hardship available from this survey is the source of financing. Households were asked for the source of financing for hospitalization, and responses were categorized into following options: 1) household income or savings; 2) borrowings; 3) sale of physical asset; 4) contributions from friends and relatives and 5) other. There could be multiple responses. Having to borrow or sell physical assets as one of the sources of financing indicates financial hardship.
We know that enrolment in PFHI is not random, and it is correlated with the OOPE and CHE; those who were enrolled in PFHI may differ from those who were not in some systematic way. In this situation, reduction in OOPE and CHE could be under-estimated because of confounding factors, such as economic quintile, type of provider, social group, education level, sex and disease category for which treatment was sought.
We used propensity score matching (PSM) to match for these characteristics across the households with PFHI coverage to those without any insurance coverage to estimate the contribution that PFHI makes to reducing the incidence of CHE due to hospitalization expenses. In cross-sectional data, PSM establishes that an intervention of interest (in this case PFHI) contributes to an outcome of interest (in this case household CHE incidence). This method ensures that other observed background characteristics or variables are matched in intervention and non-intervention groups so that their influence can be controlled. Using a counterfactual model, we have estimated the average outcome of the treated households (which is the incidence of CHE in households with insurance coverage in this study) and the average outcome that the treated households would have obtained in the absence of PFHI, which is unobserved. The average treatment on treated (ATT), which measures the average difference in CHE incidence that PFHI affords to households with PFHI coverage, is a measure of the effectiveness of the PFHI [18].
We matched for the following six variables: sex, social group (caste), education of head of household, economic quintile, choice of public or private provider and disease category. The main causes of hospitalization were categorized into 12 groups: infections, cancers, metabolic and blood diseases, mental and neurological illness, eye, cardiovascular, respiratory, gastro-intestinal, musculo-skeletal and genito-urinary, obstetric (including childbirth), injuries and others. The nearest neighbour matching method with replacement was used in combination with a logit model. To satisfy the balancing property on all of the background characteristics, a ‘hit’ or ‘miss’ approach was used.
The category of PFHI includes a variety of schemes (see Additional file 1), including those that provide insurance coverage for government employees. To study the contribution of the subset of PFHIs that are designed to prevent CHE among the poorer socio-economic groups of the society, we have conducted the PSM in two stages. First, PSM was applied to the total population (all five income quintiles), and later, it was applied only for the bottom three quintiles. We conducted PSM for all hospitalizations, irrespective of quintile groups, then conducted PSM again separately for the bottom three quintiles.
This study also examined whether PFHI coverage is associated with a change in the choice of provider since PFHIs were expected to overcome financial barriers to access of care in the private sector. Furthermore, to also study whether PFHI coverage led to increased utilization of hospitalization services, a multiple logit model was constructed, with the likelihood of hospitalization as the dependent variable and with publicly funded insurance coverage, education, age, economic quintile, sex, urban residence and social status as independent variables. Analysis was conducted using STATA software (version 13).