Assessing the Pattern of Utilisation and Out of Pocket Expenditure in Public and Private Hospitals in India – Lessons for Universal Health Coverage from the Healthcare Rounds of National Sample Survey 1996 to 2017


 Introduction: Out of Pocket Expenditure (OOPE) contributes to impoverishment and Catastrophic Health Expenditure (CHE) and restricts equitable access to healthcare in many Low and Medium Income Countries (LMICs), including India. Indian government has implemented different strategies to expand access and to reduce OOPE in public and private hospitals in its mixed healthcare system. The study aims to assess the long-term pattern of utilization, OOPE and CHE in public and private hospitals and to draw policy lessons for Universal Health Coverage in India. Methods: Indian government conducts periodic household surveys called National Sample Survey (NSS). Unit data from the last four rounds of NSS (1996 to 2017) on healthcare utilization were analysed. Multivariate analysis was used to find out determinants of utilization, choice of provider, OOPE and CHE. Propensity Score Matching was applied to find effect of specific variables on OOPE and CHE. Results: The share of public-sector in hospital-utilisation fell from 1996 to 2004 but grew consistently after 2004, reaching 51% of utilization in 2017. Socio-economically vulnerable sections were more likely to utilize public-sector. Mean OOPE per hospitalization in public-sector registered a decline from 2004 to 2017, while it increased substantially in private-sector. OOPE in private sector was around six times greater than public sector in 2017 and incidence of CHE was nine times. Utilising private-sector was an important determinant of incurring CHE. Coverage under publicly funded insurance was ineffective in reducing OOPE or CHE. Discussion: Public sector provided effective protection to the poor from financial risk. While, the structural-adjustment policies of 1990s had resulted in reduction in public-sector utilisation, the supply-side strengthening of public sector after 2005 was more effective in improving access and financial-protection. For achieving UHC, Indian health-system needs increased public funding for strengthening public-sector, especially to provide services for NCDs and injuries. Persistently high OOPE in private-sector raises doubts whether public-funding or contracting can align provider incentives with goals of UHC. The debate on public-private provider mix and financing policies continues to hold relevance for health-systems performance across LMICs.

NSS followed a two-stage strati ed sampling. Detailed information on the sample design is available in NSS documents [24][25][26][27][28][29][30]. The sampling weights were taken into account in the analyses as applicable. For a detectable difference of 5% at 95% con dence and a design effect of 1.5, a minimum of 574 hospitalizations were required in each round. The actual number of hospitalizations covered, including child birth and maternal care in the four rounds were 26526, 35566, 57456 and 93925 respectively. Data Analysis: All hospitalizations, including those for maternal care were included in the analysis for all the four rounds. Out of Pocket Expenditure (OOPE) was calculated for each episode by adding medical expenses and expenses on transportation and deducting any cash-reimbursements received by the patient. OOPE amounts for 1996, 2004 and 2014 were adjusted at 2017 prices for valid comparison, as done by recent studies [21,22,34]. For the above adjustment, price de ators for rural (agricultural labour) and urban areas (industrial workers) were used [21,22,34,35]. The survey collected data on usual monthly consumption expenditure and it was multiplied by twelve to calculate the Usual Annual Consumption Expenditure. Recent studies analyzing the NSS datasets have used the same procedure for calculating Annual Household Consumption Expenditure [20][21][22].
Financial Protection was measured in terms of Catastrophic Health Expenditure (CHE) as proposed by Wagstaff and Doorslaer [36]. A threshold of 25% of concerned household's Annual Consumption Expenditure were taken for CHE and named CHE25 [20,21]. The analysis was repeated for 10% and 40% thresholds.
The survey data was analysed using STATA-15. The list of variables in the study is given in Additional File -S1. Cross-tabulations were carried out for descriptive analysis and the indicators were reported with 95% con dence intervals (CI). Multivariate analysis was used to nd out the determinants of utilization, choice of provider, OOPE and CHE25. Ordinary Least Squares (OLS) regression was applied for OOPE. Logistic model was used for binary outcome variables (Utilisation, CHE25). Signi cance was taken at 95% (p<0.05). Propensity Score Matching (PSM) was applied to nd out the effect of utilising public sector (as compared to utilisation in private sector) on OOPE and on CHE25.
PSM has been used widely for evaluating PFHI, including in India [20,22,[37][38][39]. For nding out the effect of PFHI enrolment on OOPE and CHE25, Average Treatment Effect on the Treated (ATET) under PSM modeling was used. For PSM, the area of common support and bias balance after matching was checked, using STATA-15. Since PFHI in India was focused on covering the poorer sections of population, PSM was repeated for the bottom two quintiles of the dataset [20].
Difference in Difference (DID) is known to be a suitable method for addressing selection issues in insurance (37)(38)(39). DID was applied for utilisation in private sector to see whether PFHI made any impact on OOPE in private sector. DID requires a 'treated' and a 'control' group, that are compared before and after an intervention. For this purpose, Andhra Pradesh and Telangana -the two states with highest rate of enrolment under PFHI (67% population covered in 2017) were taken as 'treated'. Bihar and Jharkhand -two states with nearly zero rate of enrolment under PFHI (0.1% population covered in 2017) were taken as 'control'. Since PFHI was started in India between 2004 and 2017, these years were taken as 'before' and 'after' respectively.

Sample Pro le
The sample pro le for the four rounds of survey is given in Additional File -S2.

Hospital Utilisation
The proportion of individuals who utilised hospital-care is given in Table 1. It includes the hospitalisations for child-birth and maternal care. Multi-variate logistic regression showed that women, married persons, rural inhabitants, persons belonging to smaller households and those enrolled under PFHI were more likely to utilize hospital-care than the other corresponding categories (Additional File -S3). The poorer quintiles, the Scheduled Tribes, informal workers and the illiterate individuals were less likely to utilize hospital care than the other corresponding categories.
Pattern of Out of Pocket Expenditure: The mean OOPE per hospitalization is given in Table 2.   The proportion of episodes involving catastrophic health expenditure at 25% threshold of annual household expenditure (CHE25) is given in Table 4.  Table 4 shows that the CHE25 for hospitalization increased sharply from 1996 to 2004. CHE25 for hospitalization in public-sector also increased from 1996 to 2004 but declined sharply thereafter. CHE25 in private sector increased over the years. The occurrence of CHE25 was twice as common for private sector utilisation as compared to using public sector in 1996. The difference grew over the years. In 2017, CHE25 was almost nine times as common for private sector utilisation in comparison to public sector.
Logistic regression for nding out determinants of CHE25 showed that its occurrence was more likely for hospitalizations of men, persons from bigger households, married persons, rural inhabitants, informal workers and the elderly (Additional File -S5). CHE25 also increased with education. The Scheduled Tribes were less likely to incur CHE in comparison to other social groups. Higher levels of poverty were associated with increased chances of CHE25.
Hospitalisations for injuries and NCDs and were likely to involve greater incidence CHE25 than the communicable diseases. CHE25 increased with length of hospitalization. Those utilising private-sector hospitals were likely to have signi cantly greater incidence of CHE25 than those hospitalized in public facilities.
PSM model showed that utilisation in public hospitals was likely to have signi cantly less incidence of CHE25 than in private hospitals ( Table 3).

Effect of Publicly Funded Health Insurance (PFHI) on OOPE and CHE:
The 2014 round was the rst NSS round conducted after the PFHI schemes got introduced in India. Table 5 gives the mean OOPE per episode according to PFHI-enrolment status of individuals utilising hospital care.  Table 5 shows that the mean OOPE was marginally lower for those enrolled under PFHI. Public sector was several times cheaper than private sector for the PFHI-enrolled as well as the non-enrolled.
PSM model for effect of PFHI on OOPE showed that insurance was likely to have a small effect on OOPE (Table 6). When the PSM was repeated for the bottom two quintiles, there was no signi cant effect of PFHI on OOPE (Table 6).  Table 7 gives the CHE25 occurrence for hospitalizations according to insurance-enrolment status of patients.  Table 7 shows that CHE25 was nearly equal for the PFHI-enrolled and the non-enrolled.
PSM for effect of PFHI on CHE25 showed that the PFHI-enrolled had 2 percentage points less likelihood of CHE25 for hospitalization than the non-enrolled. When the PSM was repeated for the bottom two quintiles, there was no signi cant effect of PFHI on CHE25.
DID for highest enrolment states (Andhra Pradesh-Telangana) and zero-enrolment states (Bihar-Jharkhand) for 2004 and 2017 showed that PFHI-enrolment had no signi cant effect on OOPE or CHE25 for utilizing private hospitals (Additional File S6).
The above analyses were repeated for CHE at 10% and 40% thresholds but the results remained similar.
Pattern for Choice of Provider: The share of public providers in utilisation of hospital care is given in Table 8.   Table 9 shows that the share of public sector in utilisation by most of the categories of population registered a decline from 1996 to 2004. From 2004 to 2017, there was a rise in share of public sector in utilisation by most categories and it was more pronounced for the women, rural population, informally employed, the marginalized social groups (ST, SC) and the poorest (bottom two quintiles). The increase in share of public sector utilization was less impressive in case of socio-economically better-off categories. The multivariate logistic regression showed that among the hospitalized, utilisation of public-sector care was more likely for persons belonging to smaller households, men, unmarried persons and those enrolled under PFHI (Additional File -S7). The socio-economically disadvantaged sections of the poor, the Scheduled Tribes and Scheduled Castes and informal workers were signi cantly more likely to utilize the public sector. Maternal conditions were strongly associated with use of public sector. The utilization of public sector decreased with education status.

Discussion
The current study found that many vulnerable sections of the population -the poor, the Scheduled Tribes and informal workers were less likely to utilize hospital care. This shows that considerable inequity has persisted in India in terms of access to hospital care in India. This issue has been highlighted by many studies from India as well as from other LMICs [10,[40][41][42][43][44][45].
Similar to the ndings reported in earlier studies, hospitalisations for injuries and NCDs and were likely to involve greater OOPE and CHE than other diseasecategories in the current study [13,33,34,46]. Studies in other parts of the world have also found that OOPE increased with NCDs [47][48][49][50][51].
involvement in public services has been one of the common themes in existing literature [100][101][102][103]. Studies have suggested that engaging private-sector can be useful for some speci c and restricted needs but was unlikely to work for a system wide scale for purposes of UHC [102,104].
In terms of the determinants of choice of provider, the current study found that the socio-economically disadvantaged sections of -the poor, the Scheduled Tribes and Scheduled Castes and informal workers were signi cantly more likely to utilize the public hospitals. Earlier studies have also pointed out a similar pattern [20,32,40,41,43]. A similar reliance of the poor on the public sector has been reported from many LMICs [49].
The current study found that incidence of CHE increased in public sector from year 1996 to 2004 and the share of public sector in hospital-care declined in the period. There is evidence that the above decline in public sector had started before 1996. Studies comparing the rst Health round of NSS (1986) with the 1996 round found that the share of public sector in hospitalizations declined sharply. The period around 1990 was when the Structural Adjustment Programme (SAP) was implemented in India at the advice of international nancial institutions [105,106]. This was part of widespread changes happening globally, including in the developing world. The key components of the reforms in India under the SAP were -scal discipline and downsizing public sector, introducing user fees for public services and promoting expansion of private sector healthcare [107,108]. World Bank played a key role in advancing SAP reforms in health sector in India. Citing the de ciencies in public sector, it advocated privatization of healthcare. It was advocated that patients were likely to get better services if they paid for them. The ability of the people to pay was also expected to rise with fast growth of economy. The episode based healthcare was seen as a private good and e ciency was expected to be achieved through competition and the choice exercised by patients as paying consumers [108][109][110].
Public expenditure on healthcare in India was stagnant in the 1996-2004 period at less than 1% of Gross Domestic Product (GDP) [111,112]. Although Indian economy grew at a rapid pace during the 1990s, it did not translate into an increase in public funding for healthcare [106]. From 1995 to 2006, public expenditure on health remained stagnant both at the national and state levels in India [113]. The above period was characterized by chronic under-funding of public sector [106]. The reforms included reducing investment in public hospitals, especially at tertiary level [108]. The lack in health service provision resulted in inaccessibility of health services for the general population [42]. The inability of the public sector to meet the healthcare needs resulted in dependency on private providers. User fees were imposed on utilisation of public sector in India, starting from 1992 [106,107,115]. Cost recovery in public sector was a key component of World Bank's advice for LMICs [106,108]. Studies in India and globally have reported the adverse effects of user fees on access and expenditure of the poor for healthcare [11,42,112,[116][117][118]. User fees could have contributed to the reduction in access to public sector in India. The pharmaceutical sector was liberalized in India in 1990s and the price controls were reduced considerably [109]. The rise in medicine costs has also been cited as a key contributor to rising OOPE in India, including in the public sector [106,115].
The current study found that hospital utilisation grew in the period 1996 to 2004 and so did the share of private sector in hospital-care. Other studies have also reported similar ndings [42,112,115]. In order to meet the increasing demand including from an expanding middle-class, Indian governments promoted private hospitals by giving them free land and other subsidies [42,107,109,112]. Encouragement was offered by government for increasing private investment in hospitals, including through foreign capital. The number of private hospitals rose sharply in the period [109]. The shortage of services in public sector further boosted the growth of private sector [112,115]. The reforms agenda of the period was focused on increasing the role of private sector [40,42].
In terms of equity, the access of the poor to hospital care registered a decline and OOPE increased substantially from 1990s to 2004 [42,106,112]. Private hospitals were always accessed less by the poorer sections due to affordability barrier. But in the above period, even public sector was utilized more by the rich [112]. Researchers have concluded that SAP policies pursued in India from 1990s to 2004 had a bad effect on the public sector and resulted in rise in CHE, impoverishment and inequity [40,42,74,112,115].
The current study shows that many of the above patterns got reversed after 2004. The share of public sector in hospital utilization grew substantially from 2004 to 2017. In the above period, the share of public sector in out-patient care also increased [32]. The OOPE per episode of hospitalization in public sector declined from 2004 to 2017, while the OOPE for utilizing private hospitals increased in this period. The period after 2004 has seen two major policy reforms in Indian healthcare -a) Launch of National Rural Health Mission (NRHM) in 2005 b) Publically Funded Health Insurance (PFHI) schemes [74].
In response to the high proportion of OOPE and to make private-sector affordable for the poor, PFHI schemes got introduced in many parts of the world including India [37,67,119,120]. While a national insurance scheme began in India in 2008, some states had started such schemes a few years earlier. These PFHI schemes were focused almost exclusively on in-patient or hospital-care for the poorer sections. The current study found that the schemes were not effective in avoiding CHE for the insured. Increase in insurance cover did not reduce nancial risk in India [20,21,33]. There was negligible decrease in OOPE for utilization in private sector because of such schemes.
The ineffectiveness of the insurance-cover through PFHI in Indian situation has been attributed to "double-billing" i.e. private providers tend to appropriate the insurance-bene t while taking extra charges from the patients illegally [21,22,[70][71]. The continued poor regulation of private sector contributed to failure of PFHI programmes [69, 73. 121, 122]. Another assessment is that the development of PFHI in India was market oriented and it had a damaging effect on access and equity [106,123]. The problems seem to have persisted in the revised national PFHI scheme called Ayushman-Bharat Pradhan Mantri Jan Arogaya Yojana (AB-PMJAY) [22]. PFHIs have been promoted globally as a vehicle for UHC [38,124]. The evidence on effectiveness of PFHIs in ensuring nancial protection in LMICs has not been very convincing so far [37][38][39][125][126][127][128][129][130][131][132][133].
There has been a perception that most of the healthcare in India takes place in private sector [134]. Recent policy documents of central government have also stated that private sector accounts for around 80% of the healthcare in India [135]. Various studies have reported that the share of public-sector in hospital-care in India has been declining [49, [134][135]. Contrary to the above, the current study found that share of public sector in hospital-utilisation in India was rising from 2004 to 2017. In 2017, public sector crossed 50% share in hospital-care utilisation. The current study included hospitalizations for child-birth in the utilization gures whereas most of the other studies did not. Maternal care has been an area in which the expansion of public sector services was most impressive [23]. This may explain how some of the earlier studies have ended up with a different conclusion regarding the share of public sector in hospital care in India. The dominance of public sector in hospital care has been a phenomenon common to a majority of countries globally [49].
The current study showed that the mean OOPE in real terms and the incidence of CHE for utilizing private sector increased substantially in India from 2004 to 2017. In this period, OOPE and CHE for hospitalisation in public sector decreased. Indian government had introduced NRHM in 2005. It has been called a watershed in health-sector reforms in India [136]. It was initiated to improve the availability of healthcare in rural areas, primarily by increasing funding for existing public sector [55]. It improved government spending on healthcare, both by the central government and the states [136]. Under NRHM, the funding was mainly for supply side strengthening of public sector. More human resources were added to the public sector. One key area of improvement in public services under NRHM was of maternal health [23]. It seems that the maternal care acted as the engine for revival of public hospitals in India. NRHM abolished user fees in maternal, neonatal and child care. In 2012, National Urban Health Mission (NUHM) was introduced to strengthen primary care for the urban poor through supply-side strengthening of public sector. Recently, an emphasis has emerged on services for NCDs including through additional funding by the states [32,138].
States that have improved availability of free medicines and diagnostics in public facilities have achieved remarkable reductions in catastrophic expenditure for the poor [137,139,140]. Many feel that as the public sector services expand, there will be fewer takers for private sector [32,109]. As found in the current study, the argument that the people do not prefer public hospitals does not seem to be true in Indian context.
However, the progress in supply-side reforms has been uneven across states in India [33,140]. Further supply-side reforms may be necessary in many states including -attracting and retaining more doctors and health human-resources, strengthening diagnostic services, removing all user fees especially for tests and providing free drugs in public hospitals. Improving quality of care in public facilities can help them in retaining and expanding their clientele.
The debate on health sector reforms needed in India has reached an important juncture now. Researchers have highlighted that there was a lack of consensus regarding the role of private sector in India and the optimal public-private mix [134]. In early 2000s, India had both the options -a) relying on public sector or b) shifting to public funded but largely privately provided services [141]. India has experience of around fteen years of trying both the options simultaneously for hospital-care. PFHI schemes seem to have largely failed, mainly due to their reliance on private providers in absence of effective regulation. The quali ed private providers tend to be concentrated in urban areas [109,142]. Supply side strengthening of public sector seems to have given better results for UHC in terms of covering the poor and saving them from catastrophic expenditure. The public sector seems to be more cost-effective than private provisioning [106].
Public spending was also found to be effective in improving the health status of the poor [143]. Public spending on health is still very low but with a growing economy India has the scal space to increase it [8,143,144]. The recent inclusion of NCDs in essential primary care by Government of India seems to be a step in the right direction [145]. The current study suggests that greater attention is needed in India to also expand the hospital-based services for NCDs and injuries in the public sector.
Global literature has indicated that health systems with greater utilization of public sector were likely to have lower levels of OOPE [146,147]. Some studies from LMICs have shown that only those utilizing public facilities could avoid OOPE [148]. Some South East Asian LMICs like Srilanka, Malaysia and Thailand achieved remarkably low levels of CHE by relying mainly on public-sector based healthcare delivery and this form of nancing has been found to be pro-poor [92,[149][150][151]. Recently, it has been suggested that keeping the provider incentives in mind, public spending through public agencies may be more effective in reducing CHE globally [5].
While the structural adjustment policies of 1990s caused damage to public sector and reduced its utilisation, the supply-side strengthening of public sector under the National Health Mission has been more successful in improving access and nancial protection in India. India certainly needs an increase in public funding for healthcare for moving towards UHC and it is likely to be more effective if policies focus on further strengthening of public sector, especially to expand services for NCDs and injuries.
Policies of publicly funded and privately provided care need to be evaluated in light of current evidence from LMICs. Out of pocket expenditure is a very important aspect in performance of health-systems and it depends upon national policies. The debate on public-private provider mix and its relationship with nancing policies has great relevance for health systems design across LMICs.