Examining the factors contributing to a reduction in hardship financing among inpatient households in India

In India, the rising double burden of diseases and the low fiscal capacity of the government forces people to resort to hardship financing. This study aimed to examine the factors contributing to the reduction in hardship financing among inpatient households in India. The study relies on two rounds of National Sample Surveys with a sample of 34,478 households from the 71st round (2014) and 56,681 households from the 75th round (2018). We employed multivariable logistic regression and multivariate decomposition analyses to explore the factors associated with hardship financing in Indian households with hospitalized member(s) and assess the contributing factors to the reduction in hardship financing between 2014 and 2018. Notably, though hardship financing for inpatient households has decreased between 2014 and 2018, households with catastrophic health expenditure (CHE) had higher odds of hardship financing than those without CHE. While factors such as CHE, prolonged hospitalization, and private hospitals had impoverishing effects on hardship financing in 2014 and 2018, the decomposition model showed the potential of CHE (32%), length of hospitalization (32%), and private hospitals (24%) to slow down this negative impact over time. The findings showed the potential for further improvements in financial health protection for inpatient care over time, and underscore the need for continuing efforts to strengthen the implementation of public programs and schemes in India such as Ayushman Bharat Pradhan Mantri Jan Arogya Yojana (PMJAY).


Improvements in the health system for inpatient care
In Asia, between 2015 and 2017, there was a reduction in the proportion of people who fell into relative poverty due to OOPE0 3 .In India, the OOPE as a percentage of total health expenditure has decreased to 48.2% in the financial year 2019 from 64.2% in the financial year 2014 11 .The literature points out an increase in the utilization of public healthcare facilities from 2014 to 2017-18 in India and a consequent decline in OOPE and CHE 31 .Nonetheless, people are pushed into financial catastrophe owing to inpatient expenditures 32 .The poor and near-poor people face financial hardship even when the OOPE is less than 10% of household income 3 .When encountering income losses, poor households often turn to selling off their productive assets or incurring debt.As documented, using hardship financing to investigate impoverishment can help overlook some CHE limitations 28 .
Most studies that looked into hardship financing in India in the past have either been disease-specific, statespecific or cross-sectional studies 18,29,[33][34][35][36][37] .In India, the majority of government schemes focus on providing financial protection for inpatient care, and government expenditure on health for inpatient care is always higher than outpatient care 38 .Evidence from a recent study indicates a decrease in hardship financing for inpatient care from 2014 to 2017-2018 13 .However, a decline in hardship financing does not mean that hardship financing has ceased.The reduction in hardship financing is an ongoing process, and at every stage, it is important to assess how each underlying factor has contributed to the decline.Therefore, an in-depth investigation to know the contributing factors to this decline is critical.It is required to understand the implications of existing government interventions and discover a further action plan to eliminate hardship financing.The present study is a novel attempt to identify the contribution of various socioeconomic and demographic factors to the change in hardship financing from 2014 to 2018 for inpatient households.This would help us determine the impact of various underlying factors on the change in hardship financing over time and what must be done to combat healthrelated expenditures further.Thus, the objectives of the present study are (1) to explore the factors associated with hardship financing in Indian households with hospitalized member(s) and (2) to assess the contributing factors to the reduction in hardship financing from 2014 to 2018.
The current study is divided into four sections.The methods section discusses the details of the data used in the study and the statistical methods used to decompose the impact of underlying factors on the reduction in hardship financing in India.The results section presents the interpretation of the results, followed by a discussion section that elaborates on our significant findings on what and how various factors contribute to the reduction in hardship financing and compares between existing studies and suggests policy implications of the current findings.The final part of the paper reports concluding remarks from the current findings on factors influencing the reduction in hardship financing among inpatient households in India.

Data
The study relied on two rounds of repeated cross-section National Sample Surveys (NSS), namely Survey on Social Consumption (71st round; 2014) and Social Consumption in India: Health (75th round; 2017-18) by the www.nature.com/scientificreports/National Sample Survey Organization (NSSO), Ministry of Statistics & Programme Implementation, Government of India.NSS is a large-scale, nationally representative data that provides unit-level information.A stratified multi-stage sampling method for data collection was used to ensure that the sample data will represent the whole nation.The data includes household and individual-level information on their socioeconomic status, morbidity indicators, government-sponsored health insurance schemes, utilization of health services, and health expenditures associated with inpatient and outpatient services.Additional information regarding the survey design and data collection is available in the survey reports 39,40 .
For the present study, we used the inpatient NSSO data, which provides detailed information on hospitalization cases 365 days prior to the interview.The sample survey entailed individual-level information on 57,546 hospitalization cases in 2014 and 93,925 hospitalization cases in 2018, and the entire survey covered 65,932 households in 2014 and 113,823 households in 2018.The current analysis was conducted at the household level, and we limit the study to households with at least one ill individual who has sought inpatient hospitalization services.We focus on studying the impact of unforeseen health shocks on hardship financing and, therefore, exclude hospitalization cases for childbirth from the study.Thus, in our analyses, we used a sample of 34,478 households with at least one inpatient from the 71st round (2014) and 56,681 households with at least one inpatient from the 75th round (2018).

Measures Outcome variable
The outcome variable of our study was hardship financing of the inpatient households.NSSO 71st and 75th round questionnaires asked the respondents to report the source of funds for household health expenditure (excluding any reimbursement) during hospitalization.The possible responses were coded as 1 if the money was funded through household income or savings, 2 if they had borrowed money, 3 if they sourced the fund by selling physical assets such as cattle or jewellery, 4 if the money came as contributions from friends or relatives, 5 if there are any other sources of financing 39,40 .An inpatient household in this study was identified as having hardship financing if it resorts to any financial coping strategies of borrowing, selling of physical assets, contribution from friends or relatives or any other source to meet the health care expenditure of at least one of its members 18,28 .We have coded it as 0 (if the inpatient household does not resort to hardship financing) and 1 (if the inpatient household resorts to hardship financing).

Explanatory variables
Various socioeconomic, demographic and health characteristics that have been shown to impact household hardship financing have been used in this study 13,29,41 .

Household-related variables
The age group to which the household head belongs to was recoded as 0 = working-age adult (18-59 years) and 1 = older adult (60 + years), sex of the household head was recoded as 0 = male head 1 = female head, the household composition was recoded into categories 0 = household with children and adult only, 1 = household with elderly and adults only, 2 = household with only adults, 3 = household with children, adults and elderly, 4 = household without adults, household sex composition was recoded as 0 = households with both women and men or only men and 1 = household with only women.
The employment status of the household head was categorized as 0 = self-employed, 1 = regular wage, 2 = casual labour, 3 = no income, living condition index was coded as 0 = low living condition index, and 1 = high living condition index, household expenditure was used to categorize households based on their economic status.It was recoded as 0 = quintile one, 1 = quintile two, 2 = quintile three, 3 = quintile four and 4 = quintile five, CHE of the household was coded as 0 = household without CHE, 1 = household with CHE.

Health-related variables
Insurance status was recoded as a household with no member having insurance = 0 and at least one member having insurance = 1, death of a household member was recoded as 0 = all members are alive and 1 = household with at least one deceased member, chronic disease status was recoded as 0 = no member of household suffered from chronic disease and 1 = at least one member suffered from chronic illness, type of medical facility used was recoded as 0 = only public facilities and 1 = at least one patient member used private facilities, NCD status of household was recoded as 0 = household with one inpatient member suffering from NCD only, 1 = household with more than one inpatient member suffering from NCD, 2 = household with one inpatient suffering from non-NCD alone, 3 = household with more than one inpatient suffering from non-NCD and 4 = households with inpatient(s) suffering from both NCD and non-NCD, number of days hospitalized was categorized as 0 = up to 4 days, 1 = 5-10 days, 2 = more than 10 days up to a year.An exploratory analysis based on the number of patients in each group was used to categorize the length of hospitalization 42,43 .
The diseases were categorized into NCD, communicable and other diseases based on categorization followed by previous literature 44,45 .We have re-categorized them into NCD and non-NCDs, where non-NCDs comprise communicable and other diseases.Chronic ailments reported in NSSO 75th round were severe ailments affecting any of the organs in the person's body and have significant symptoms lasting for more than a month.Factor analysis using principal component analysis for factor extraction was conducted based on household-level information on the kind of energy used in households, the types of drainage systems, latrine systems, and drinking water, and arrangements for garbage disposal to measure the living condition index 13,46,47 .This was following the Indian Demographic and Health Surveys (DHS) that have employed principal component analysis to construct wealth indices with a set of binary variables on various asset ownerships 48 .Before conducting the factor analysis, we conducted the Kaiser-Meyer-Olkin Measure of Sampling Adequacy (within acceptable range) and Bartlett's test for sphericity (significant at 1% level of significance).The eigenvalues and proportion of variations explained by each factor are given in the appendix (see Appendix Table 1).Households' OOPE on health is catastrophic when the health expenditure exceeds an arbitrarily set threshold of the total household expenditure 27,[49][50][51] .The different thresholds and the percentage of people suffering CHE at all these thresholds are given in Table 2 of the Appendix.The official CHE threshold for measuring universal health coverage financial protection are 10% and 25%, of which, the 10% threshold is the most commonly used 52 .In India, around 30% of inpatient households sacrificed more than 10% of total household expenditure to meet their health expenditures 27 .Based on evidence from these previous studies, we kept the threshold for CHE as 10% of the total household expenditure 27,53 .Since we are using the information on household expenditure in two periods, to adjust for inflation, we have adjusted the 2018 prices using the annual average consumer price index 54 .

Statistical analysis
The statistical analysis of the data was conducted on Stata version 15.1, and the multivariate decomposition analysis was conducted using the command mvdcmp.Sampling weight was used for reporting weighted percentages in descriptive statistics, logistic regression, and decomposition analysis.De-normalized weights were used in the logit decomposition analysis using the pooled data from the period 2014 and 2018 47,55 .
Descriptive statistics were reported to understand the general distribution of samples.The prevalence of hardship financing was calculated for both periods.A proportion test was conducted, and p values were reported to show the level of significance of the change in hardship financing between 2014 and 2018.Multivariable logistic regression analyses were carried out to find the significant factors associated with hardship financing in households in 2014 and 2018.
The logistic regression can be summarized in the following Eq.(1): where p i = Probability of success or Hardship financing (coded as1), 1 − p i = Probability of failure or no Hardship financing (coded as 0), X ij = Household disease composition, categorized as: 1 = Household with one member having only NCD, 2 = Household with more than one member having only NCD, 3 = Households with one member have only non-NCD, 4 = Household with more than one member having only non-NCD, 5 = Households with both NCD and non NCD, Y ij = Vector of individual characteristics, Z ij = Vector of household characteristics, ǫ 0 = Error term.Furthermore, a decomposition analysis identified the crucial factors that led to the change in hardship financing between 2014 and 2018.The Blinder and Oaxaca models of decomposition developed in the early 1970s have been used for decomposing linear regression models 56,57 .Multivariate decomposition analysis is extended to non-linear regression decomposition models and can be used for linear, count or logit models 58 .Throughout literature studies have used multivariate decomposition analysis to study the contribution of underlying factors in non-linear models with a time component [59][60][61][62] .Such a decomposition analysis will help us to quantify the contributions of underlying factors to change in hardship financing over time.Our model is summarised as follows: In Eq. ( 2) mentioned above, time t1 (2014) is the comparison group, and t2 (2018) is the reference group.X represents the vector of dependent variables and β vector of regression coefficients.The mean difference in Y between 2014 and 2018 can be decomposed into a characteristics effect and a coefficient effect.The E component captures the differential in two time periods due to differences in the endowments or characteristics.This is the explained component of the model, also called the characteristics effect.The C refers to the differential in the two time periods due to differences in coefficients or the effects.This is the unexplained component of the model, also called the coefficient effect.A positive E coefficient would indicate the reduction in hardship financing for households with hospitalization cases between the two time periods if the distribution of X between the two time periods were equal.At the same time, a negative C coefficient indicates an increase in hardship financing between the two time periods if both periods have the same behavioral responses 58 .
The mean-variance inflation factor of the logistic regression models was less than two and shows no evidence of multicollinearity.We checked for the robustness of the logistic models of t 1 (2014) and t 2 (2018) using the pooled cross-section regression model combining the information from both periods.The outcome of pooled regression was comparable to the results of t 1 and t 2 models, thereby indicating the relevance of each independent variable used in the analyses. (1)

Results
Table 1 shows the profile of the sample population in 2014 and 2018.Out of the total households, 25.74% and 23.59% were headed by older adults in 2014 and 2018, respectively.Most of the households in both survey rounds were male headed.In 2018, 3.54% of the sample households had only older adults, whereas 23.71% had only adults.Similarly, 2.21% of the sample households had only female members.Further, around 27.1% of the households were headed by members without formal education, whereas 9.71% of household heads had a higher education (graduation and above).Around 23.37% of the households had at least one member covered by health insurance, and 24.5% had a chronic disease patient in 2018, which was even higher (32.21%) in 2014.
Figure 1 presents the change in the prevalence of hardship financing from 2014 to 2018.Hardship financing was significantly reduced from 29.10% in 2014 to 18.55% in 2018 (Difference: − 10.59%, p < 0.001).Table 2 presents the prevalence of hardship financing by the background characteristics in 2014 and 2018, the differences, and their statistical significance.
Table 3 provides the adjusted estimates from the logistic regression of factors associated with hardship financing, separately for 2014, 2018 and the pooled data.In the pooled data, households with a head in the age group of 60 and above were less likely to have hardship financing [AOR 0.88, CI 0.76-1.02]than those with a head in the age group of less than 60.In comparison to those households with children and adults only, households with only adults [AOR 0.90, CI 0.81-1.00]or adults and older adults [AOR 0.82, CI 0.69-0.96]or those with children, adults and older adults [AOR 0.83, CI 0.72-0.96]had lower odds of hardship financing in the pooled data.Further, households with only female members had higher odds of hardship financing [AOR 1.35, CI 1.01-1.80]than other households.Higher levels of education reduced the odds of hardship financing.Households with a head with higher education had lower odds of hardship financing [AOR 0.33, CI 0.28-0.39]than those with no formal education.On the other hand, households with health insurance had higher odds of hardship financing [AOR 1.35, CI 1.23-1.49]than those without insurance.
Notably, households with a member utilizing private health facilities had higher odds of hardship financing [AOR 1.41, CI 1.28-1.55]than those with members using public health facilities.Households with more than one inpatient member suffering from NCDs had higher odds of hardship financing [AOR 1.49, CI 1.16-1.90]than non-NCDs.On the other hand, households with a high living condition index value were less likely to have hardship financing [AOR 0.68, CI 0.62-0.76]than those with a low living condition.Also, households of members with hospital admissions of more than 10 days had higher odds of hardship financing [AOR 2.07, CI 1.83-2.34]compared to those with less than 5 days.Households with a CHE had higher odds of hardship financing [AOR 2.85, CI 2.59-3.14]than those without CHE.Finally, households with at least one member suffering from a chronic disease were more likely to have hardship financing [AOR 1.26, CI 1.15-1.38]than those without a chronic disease patient.
The estimates from the decomposition analysis (Table 4) suggest that the differences in effects (due to coefficients, C) account for 77.07% of the observed differences (decrease) in the prevalence of hardship financing.However, only 17% of the reduction in hardship financing was explained by differences in compositional characteristics (due to characteristics, E).The E reflects the counterfactual comparison of the difference in outcomes from 2014, i.e., the expected difference if households in 2014 were given a distribution of covariates similar to those of households in 2018.Similarly, C reflects a counterfactual comparison of outcomes from the perspective of households in 2018, i.e., the expected difference if households in 2018 experienced behavioural responses to covariates of those of households in 2014.
Differences due to characteristics suggest that equalizing the distribution between 2014 and 2018 for education [Secondary or higher (1.92%), and graduation and above (1.7%)],equalizing the living conditions in the households (2.22%), CHE (3.43%) is expected to contribute significantly to the decrease in hardship financing.Differences due to coefficients suggest that equalizing the behavioural responses of two time periods for female headship (− 7.57%), household composition (− 6.47%), education (− 18.43%), presence of insurance coverage (− 10.5%), social groups (− 2.16%), households with non-NCD and NCD patients (− 2.48%), household living conditions (− 14.75%) is expected to contribute to increase in hardship financing.Additionally, differences due to coefficients also suggest that equalizing behavioural responses of two time periods for factors such as CHE (31.39), private sector hospitalization (24.2%) and geographical regions (39.02%) are primary contributors expected to contribute significantly to the decrease in hardship financing in this study.

Discussion
The study explored the impact of various socioeconomic, demographic and health characteristics on hardship financing among inpatient households in India.We observed an overall reduction in hardship financing from 2014 to 2018 for households with at least one inpatient in India.The prevalence of hardship financing was 29.10% in 2014 and decreased to 18.55% in 2018.This is on par with a previous study that points out a reduction in hardship financing over time for inpatient cases 13 .Further, a previous study showed that large household OOPE forces people to resort to informal coping strategies 63 .Consistent with previous findings, we found that hardship financing is highly prevalent in households with CHE, and it was more than two times higher in both the time periods (2014 and 2018) compared to households without CHE.Besides, the findings from the decomposition analysis indicate that there is a potential to slow down this increase in hardship financing by about 32% by reducing CHE.
Many studies in India suggest an unchanged or even increased financial hardship after being covered by government-sponsored insurance schemes 13,37,[64][65][66] .This could be due to the flaws in the implementation of the schemes and inefficient allocation of healthcare resources to the insurance beneficiaries.We found similar evidence among inpatient households with insurance having a higher prevalence of hardship financing.These households were more likely to resort to hardship financing than those without insurance in both the time Vol:.(1234567890 www.nature.com/scientificreports/periods.This could be explained by the low awareness among the enrolled population.It could also indicate the increased utilization of hospital services by the inpatient household members with insurance coverage due to their greater affordability and accessibility than those that are not covered by any health insurance.These require further research to find out the reasons considering more qualitative and quantitative aspects and the nature of health insurance, public or private and state or central schemes.Findings from decomposition analysis, however, revealed the expected reduction in hardship financing if inpatient households in 2018 had the same insurance coverage status as in 2014.This shows that the hardship financing from being insured reduced between 2014 and 2018 and has implications for policy.The finding on the impact of health insurance coverage on hardship financing is particularly relevant for debate in the context of achieving UHC in the country.Importantly, hardship financing is higher in households with only women than in households with men and women.This underlines the reality of large mortality rates among women [67][68][69] .However, this result contradicts a previous study from an individual-level analysis, which points out the lower hardship financing among women than men 13 .This variation in outcome could be attributed to multiple reasons.One can be the discrimination women face in mixed-sex households.Evidence shows that in a household where women and men are suffering from illnesses, the healthcare needs of women are given lesser attention than that of their male counterparts 70,71 .Secondly, Kochar 72 suggests that the non-disabled male member is essential for households to smooth shocks.The lack of a male member or a male head could have also contributed to the higher hardship financing in households with only women.However, from decomposition analysis, the positive coefficient of female headship indicates the expected reduction in hardship financing between 2014 and 2018 if women in 2018 had the same protective effects of female headship as in 2014.This highlights the need for developing gender-specific economic policies in order to reduce the levels of hardship financing, especially in female headed households.
The age of the household head and their education status had a significant impact on hardship financing.The prevalence of hardship financing increased with the increase in the age of household head and reduced with the increase in education levels.The findings also indicate that households with elderly heads compared to working-age heads are more likely to face hardship financing, and compared to heads with no formal education, those with higher levels of education are less likely to encounter hardship financing.In China, older adults with chronic diseases who are household heads had higher intensity and incidence of CHE 73 .In India, elderly member households had significantly higher health spending than other households 74 .Literature further showed evidence of a decline in health expenditure with higher education of the household head 24,75 .The positive coefficient of education levels from the decomposition analysis also showed the protective effects of education on reducing hardship financing between 2014 and 2018.
Furthermore, there is strong evidence of high OOPE among households with members suffering from NCD in India 76 .However, a dual burden of disease in the case of India worsens the situation and burdens the public health system 77 .In our study, we found evidence of a dual burden of diseases.In 2014, households with more than one NCD and those with both NCD and non-NCD had similar levels of higher prevalence of hardship financing.In both rounds, households with more than one NCD had the highest likelihood of hardship financing, followed by those households where NCD and non-NCD co-exist.Additionally, studies show that an extended period of hospitalization leads to impoverishment in Indian households 78 .Literature suggests that the financial hardship from illnesses worsens with the severity of diseases 13,20,24 .Evidence from the USA shows that diseases lasting more than 100 days negatively impact consumption levels 79 .In the present study, the prevalence and likelihood of hardship financing were higher among households with extended hospitalization periods and chronic diseases.The prevalence and likelihood of hardship financing were higher in households with patients who visited the private health facilities compared to the public health facilities for hospital care in 2014 and 2018.This could be attributed to the exorbitant charges of private hospitals and the inadequacy of public sector hospitals 75 .The negative coefficient of the utilization of private health facilities from decomposition analysis shows that hardship financing would have increased between 2014 and 2018 if households in 2018 had the same behavioural responses to utilization of private health facilities as in 2014.However, this increase is slowed by an expected potential for hardship financing to reduce by about 24% from 2014 to 2018.Even when complete financial protection against health costs is far-fetched, government-sponsored schemes like RSBY have contributed to some increased access to hospital care, especially through empanelled private hospitals 80 .Since 2018, PMJAY, the world's largest health assurance scheme caters to the needs of the poorest 40% of the population for hospitalization expenditures in secondary and tertiary care through empanelled hospitals 81 .In 2020, 44% of these empanelled hospitals belonged to the private sector 82 .The rising increased access to private hospitals must contribute to the slowing down of increased hardship financing.
This study has several limitations.Firstly, it is limited to households with at least one case of hospitalization.Secondly, we assume that health shocks are severe when they are idiosyncratic and unexpected.Thus, we do not include hospitalization for childbirth in our study.Thirdly, the dependent variable of the study, hardship financing, is binary.Therefore, we fail to capture the extent and depth of each inpatient household's hardship financing.Fourth, the recall period for inpatient cases is 365 days, and for outpatients, 15 days 39,40 .Therefore, in this study, we could not combine the health expenditure on inpatients and outpatients to calculate the total CHE.Additionally, since the NSSO data we use is pooled cross-sectional, the absence of a time series component does not allow us to test for a causal relationship that could exist between the independent and dependent variables.

Implications for program and policy
Our findings showed that, even when complete financial protection for health care is yet to be achieved, there are still improvements in health protection for inpatient care over time.The plausible contributors to the decline in hardship financing could be the continuous efforts of the government to improve public-funded health insurance in India.The government's budgetary allocation and focus on inpatient care is always higher than outpatient care 38 .From RSBY to PMJAY, there has been a significant increase in the target population covered under the health insurance scheme.Moreover, access to private hospitals has increased through empanelled hospitals under these schemes over time.The paper reveals the need to improve the public health system further and make the private sector accessible to people with limited resources.The study also underscores the impact of the rising dual burden of NCDs and non-NCDs on worsening hardship financing.As many NCDs are lifestyle-based diseases, the study urges the government to empower and incentivize people to adopt preventive health measures to improve their quality of life.Approximately 77% of the inpatient households under study are uninsured, and the study's findings stress the importance of sound health insurance to reduce hardship financing.Government schemes and policies should aim to develop a health system that does not financially drain society's poor and vulnerable section.

Conclusions
The findings suggest an overall reduction in hardship financing for inpatient households from 2014 to 2018.In a developing country like India, the government plays a significant role in expanding the provision of health care.The findings can help draw a picture of the government's efforts to combat financial hardship for health.Even though the CHE has impoverishing effects on hardship financing, the study provides evidence of the potential to slow down this negative impact.Similarly, factors such as the length of hospitalization private sector utilization, despite its impoverishing effects on households in both 2014 and 2018, have the potential to decrease its negative impact on hardship financing over time.Our findings urge the policymakers to address the pressing need for designing and regulating a functional health insurance program to maximize the benefit of risk pooling and safeguard the health and welfare of poor households.

Table 1 .
Descriptive statistics of household characteristics in 2014 and 2018.N Number of observations in the sample.

Table 2 .
Prevalence of hardship financing by background characteristics of hospitalized households.The p value reported is based on the proportion test.

Table 3 .
Estimates from logistic regression of inpatient household characteristics on hardship financing.Odds ratios are adjusted for all the selected covariates.Ref., Reference category; 95% CI, Robust confidence interval in parentheses.

Table 4 .
Estimates from multivariate decomposition analysis of reduction in hardship financing.Coeff., Coefficient; Ref., Reference category; 95% CI, Robust confidence interval in parentheses.