Bulletin of the World Health Organization

Variations in catastrophic health expenditure estimates from household surveys in India

Magdalena Z Raban a, Rakhi Dandona a & Lalit Dandona a

a. Public Health Foundation of India, ISID Campus, 4 Institutional Area, Vasant Kunj, New Delhi – 110070, India.

Correspondence to Magdalena Z Raban (e-mail: rabanm@gmail.com)

(Submitted: 18 September 2012 – Revised version received: 29 April 2013 – Accepted: 30 May 2013 – Published online: 12 July 2013.)

Bulletin of the World Health Organization 2013;91:726-735. doi: http://dx.doi.org/10.2471/BLT.12.113100

Introduction

Out-of-pocket (OOP) payments are the primary source of health-care financing in many countries.1 In 2004–05, OOP payments in India were estimated to account for approximately two thirds of total health expenditure2 and fewer than 10% of households had health insurance for at least one member.3 OOP payments are considered “catastrophic” when they drive households into having to reduce expenditure on basic necessities.4 The proportion of households that incur catastrophic health expenditure (CHE) in a country is widely used as an indicator of the extent to which the health system protects households needing health care against financial hardship. Offering such protection is a major goal of health systems and is the purpose behind universal health coverage.410

In many countries, household surveys – some focused on consumer expenditure and others on health – are the main sources of data on households’ OOP payments for health care.1 The estimates of OOP payments vary substantially between surveys depending on survey type, type of respondents and the survey methods used, such as the length of the recall period or the number of items included in the survey questionnaire.1,1117 In India, data on household expenditure are routinely available from National Sample Survey Organisation surveys on consumer expenditure and from special survey rounds on health.18,19 All of these surveys exert an important influence on health policy because they are the sources of data for programme and policy assessment9,10,20,21 and for the preparation of the national health accounts.2,22 Other health-focused household surveys have also recently collected information on household expenditure.23,24 Although these surveys have all been used to estimate CHE and OOP payments in India, no one has ever assessed whether the estimates obtained from them are comparable.

For this paper, we generated household OOP payments and CHE estimates using data from five national and multi-state household surveys conducted in India since the year 2000 and we compared the results. We also examined and compared the number and type of household expenditure items included in each survey questionnaire to try to explain the variability in OOP payment and CHE estimates across surveys. This exercise may prove useful in standardizing survey methods to obtain CHE estimates that are valid and consistent.

Methods

Data sources

Table 1 shows the characteristics of the five surveys that have collected data on health expenditure and other expenditure in India since the year 2000. The surveys are of two types: consumer expenditure surveys and health-focused surveys.

Consumer expenditure surveys

We obtained data from the National Sample Survey on Household Consumer Expenditure, which was conducted in all Indian states in 2004–05 (NSS 2004–05)25 and 2009–1018 (NSS 2009–10). These surveys collected data on expenditure for any health service, whether or not the household paid for the service. The expenditure data thus collected is considered an approximation of OOP payments, since most private payments for health care in India are made out of pocket. NSS 2009–10 was conducted in two parts – Type I and Type II – with a different questionnaire for each one. The Type I survey used the same questionnaire as NSS 2004–05 and hence was used for all analyses; in the Type II survey, the recall period for food expenditure differed from the one that was used in the Type I survey.

Health-focused surveys

We analysed data on OOP payments from the World Health Survey conducted in 2003 (WHS 2003);23 the National Sample Survey on Morbidity, Health Care and the Condition of the Aged conducted in 2004 (NSS 2004),19 and the Study on Global Ageing and Adult Health conducted in 2007–08 (SAGE 2007–08).24 The WHS 2003 and SAGE 2007–08 were conducted in six states that were selected to be representative of India geographically and in level of development;23 the NSS 2004 was conducted in all Indian states. In WHS 2003 and SAGE 2007–08, data on OOP payments were collected from a household informant; in the NSS 2004, such data were collected from the individual treated for each episode of illness.

Expenditure variables

Table 2 shows the number of survey items or questions used to collect household expenditure data in each survey; Table 3 presents the health items recorded. NSS 2004 was the only survey that used a single question to investigate total household expenditure. As a result, it did not collect data on food expenditure separately. Since NSS 2004–05 and NSS 2009–10 were consumer expenditure surveys, they collected expenditure data on a wider variety of household items than WHS 2003 and SAGE 2007–08. The items included in the outpatient and inpatient expenditure categories varied across surveys (Table 3).

Data analysis

We measured CHE using two definitions commonly used in the literature.4,7,9,2629 Under the first definition, OOP payments were estimated as a proportion of household capacity to pay; under the second, they were estimated as a proportion of total household expenditure. Household capacity to pay was calculated as the total household expenditure less subsistence expenditure, in accordance with the method described by Xu et al.4 Subsistence expenditure – defined as the mean food expenditure of households falling between the 45th and 55th percentiles of the total sample in terms of the share of total household expenditure spent on food – was estimated for each survey separately.4 We classified a household as having incurred CHE if it had spent out of pocket on health 40% or more of its capacity to pay or 10% or more of its total household expenditure.7,9,26,28,30 We applied both definitions to estimate CHE from all the surveys except NSS 2004, where we used only the second definition because the survey did not collect food expenditure.

Since WHS 2003 and SAGE 2007–08 sampled only six states in India and the other surveys sampled all states, we examined the possibility that any differences in CHE estimates were due to this difference in sample coverage. We did this by comparing the CHE estimates from NSS 2004–05, NSS 2009–10 and NSS 2004 for all states with CHE estimates from these same surveys for the six states sampled in WHS 2003 and SAGE 2007–08. Our premise was that if the estimates for all states turned out to be similar to those for the six states, this would indicate that CHE estimates were not affected by the difference in sample coverage.

Because the differences between surveys in CHE estimates could be due to differences in OOP payment and total household expenditure estimates, these estimates were compared. The OOP payments reported in the surveys were divided into outpatient and inpatient expenditure. Expenditure on food and “other” expenditure were also investigated. “Other” expenditure comprised all household expenditure other than out-of-pocket health-care payments and food expenditure; it included specific items under prepaid health expenditure, such as health insurance, in WHS 2003 and SAGE 2007–08, and durable items in NSS 2004–05, NSS 2009–10 and SAGE 2007–08.

The mean, median and first and third quartiles of outpatient and inpatient OOP payments, food expenditure, other expenditure and total household expenditure, documented in Indian rupees (INR), were converted to 2009–10 prices using gross domestic product deflators and then to United States dollars (US$; exchange rate: US$ 1 = 46.7 INR).31,32 The interquartile range was defined as the interval between the third and first quartiles. Since outpatient and inpatient OOP payments can be affected by the proportion of households reporting this expenditure, we also compared the proportions of households that reported such payments in the different surveys and the mean and median outpatient and inpatient OOP payments of the reporting households. Since different recall periods were used in the surveys for different items of expenditure, we prorated the reported expenditures to correspond to the same recall period to facilitate direct comparisons between surveys. Thus, for inpatient OOP payments we used a recall period of one year for all surveys. Outpatient OOP was reported for the most recent month in all surveys except NSS 2004, which used a 15-day recall period. For food expenditures, “other” expenditures and total expenditures we used a one-year recall period to allow comparison between surveys. Because the two parts of NSS 2009–10 used a different recall period for food expenditure, we assessed estimates of food expenditure from both parts.

We conducted all analyses at the household level and applied survey sampling weights. To calculate the 95% confidence intervals (CIs) of the proportions, we took into account survey design features such as stratification and clustering in estimating the variance with Taylor linearization.33 Data were analysed using SAS version 9.2 (SAS Inc., Cary, United States of America).

Results

Catastrophic health expenditure

Fig. 1 shows the estimated proportion of households that had CHE according to each of the two definitions of CHE used. The estimates of CHE for NSS 2004–05 and NSS 2009–10 were the same and, since these surveys used the same questionnaire to document expenditure, there was no change in CHE from 2004–05 to 2009–10. When defined as the proportion of a household’s capacity to pay, CHE was most frequently found in WHS 2003 (33.9% of households; 95% CI: 31.6–36.2) and SAGE 2007–08 (20.0%; 95% CI: 18.8–21.3). These figures were markedly higher than for NSS 2004–05 (3.8%; 95% CI: 3.6–3.9) and NSS 2009–10 (3.5%; 95% CI: 3.3–3.7). When defined as the proportion of a household’s total expenditure, CHE was, again, most frequently found in WHS 2003 (43.5% of households; 95% CI: 41.3–45.8) and SAGE 2007–08 (31.9%; 95% CI: 30.2–33.7). In NSS 2004, 20.2% of households (95% CI: 19.7–20.6) were found to have incurred CHE. This was a higher rate than the rates found in NSS 2004–05 (14.0%; 95% CI: 13.4–14.3) and NSS 2009–10 (13.9%; 95% CI: 13.4–14.3).

Fig. 1. Percentage of households with catastrophic health expenditure (CHE), defined two different ways, as estimated from data obtained from five major household surveys conducted in India since 2000
Fig. 1. Percentage of households with catastrophic health expenditure (CHE), defined two different ways, as estimated from data obtained from five major household surveys conducted in India since 2000
NSS 2004, National Sample Survey on Morbidity, Health Care and the Condition of the Aged 2004; NSS 2004–05, National Sample Survey on Household Consumer Expenditure 2004–05; NSS 2009–10, National Sample Survey on Household Consumer Expenditure 2009–10; SAGE 2007–08, Study on Global Ageing and Adult Health 2007–08; WHS 2003, World Health Survey 2003.
a Out-of-pocket payments equalling or exceeding 40% of a household’s capacity to pay.
b Out-of-pocket payments equalling or exceeding 10% of a household’s total expenditure.

When we assessed the relative contribution of inpatient and outpatient OOP payments to CHE, we found outpatient OOP payments to be responsible for a large proportion of the households with CHE: 73.1% in NSS 2009–10 to 84.6% in SAGE 2007–08 when CHE was defined as the proportion of a household’s capacity to pay; 73.2% in NSS 2004 to 78.6% in NSS 2004–05 when CHE was defined as the proportion of a household’s total expenditure.

Table 4 shows the proportion of households that incurred CHE in NSS 2004–05, NSS 2009–10 and NSS 2004 in all states of India and in only the six states that were sampled in WHS 2003 and SAGE 2007–08. CHE estimates were slightly higher for the six states than for all states in NSS 2004–05 and NSS 2004, but not in NSS 2009–10. However, these small differences do not explain why the estimates from WHS 2003 and SAGE 2007–08 were much higher than those from the other surveys.

Outpatient care

Table 5 shows the mean and median OOP payments for outpatient care reported by all households over the most recent month. Such payments were approximately 2.6 to 3.8 times higher in WHS 2003 and SAGE 2007–08 than in NSS 2004–05 and NSS 2009–10. The interquartile ranges for WHS 2003 and SAGE 2007–08 were 2.5 to 3.8 times higher than for NSS 2004–05 and NSS 2009–10. In NSS 2004, the mean OOP payment for outpatient care in the most recent 15 days was US$ 3.1 and the median was zero.

The proportion of households that reported OOP payments for outpatient care in the most recent month varied by only 13 percentage points between the surveys; the highest proportion was found in SAGE 2007–08 (75.2%) and the lowest in NSS 2004–05 (62%) (Table 6). When OOP payments for outpatient care were considered only for the households that reported them, WHS 2003 and SAGE 2007–08 again showed substantially higher estimates (2.3 to 2.8 times higher) than NSS 2004–05 and NSS 2009–10. This suggests that the use of different items to assess how much households spent out of pocket on outpatient care had a significant impact on estimates. In NSS 2004, with a recall period of 15 days, 32.9% (95% CI: 32.3–33.4) of the households reported OOP payments for outpatient care, and the mean and median amounts paid out of pocket for such care by these households were US$ 9.4 and US$ 4.2, respectively.

Inpatient care

The OOP payments for inpatient care in the most recent year were 1.6 to 2.8 times higher in WHS 2003 and SAGE 2007–08, respectively, than in NSS 2004–05 and NSS 2009–10 (Table 5). The OOP payment for inpatient care in NSS 2004 was 0.5 and 1.1 times as high as the payment in NSS 2004–05 and NSS 2009–10, respectively.

The proportion of households that reported paying out of pocket for inpatient care varied substantially between surveys. This proportion was much higher in WHS 2003 (25.9%) and SAGE 2007–08 (25%) than in NSS 2004–05 (9.2%), NSS 2004 (12.8%) and NSS 2009–10 (13.2%) (Table 6). Interestingly, when expenditure for inpatient care was examined for the households that reported it, the lowest median (US$ 68.4, NSS 2009–10) was only 26% lower than the highest median (US$ 92.8, WHS 2003) (Table 6). In contrast, the highest median expenditure for all sampled households was three times larger than the smallest median (Table 5), which suggests that the methods used in each survey had a greater effect on the frequency with which households reported having paid out of pocket for inpatient care than on the amount reported.

Food expenditure

Food expenditure in the most recent year is shown in Table 5. The surveys that used a one-month recall period (NSS 2004–05, NSS 2009–10 Type I and WHS 2003) had median food expenditure estimates that were from 14% to 36% lower than the surveys that used a one-week recall period, alone or in combination with a one-month recall period (NSS 2009–10 Type II and SAGE 2007–08). NSS 2009–10 Type II and SAGE 2007–08, both of which had a one-week recall period for some or all items, had similar median expenditure, even though SAGE 2007–08 used only 9 items to capture food expenditure and NSS 2009–10 Type II used 142. However, WHS 2003, which only used one item, had a higher median food expenditure estimate than the other surveys with the same recall period, namely NSS 2004–05 and NSS 2009–10 Type I. The interquartile range for food expenditure in WHS 2003 (US$ 604.6) and SAGE 2007–08 (US$ 576.6) was higher than in NSS 2004–05 (US$ 378.6), NSS 2009–10 Type I (US$ 408.2) and Type II (US$ 486.0).

“Other” and total household expenditure

“Other” expenditure was lowest in WHS 2003; it was about 1.5 to 2 times higher in SAGE 2007–08, NSS 2004–05 and NSS 2009–10 (Table 5). WHS 2003 used the least number of items to assess “other” expenditure; it also used a one-month recall period for all items and it used no items to specifically document expenditure on durables. Thus, “other” expenditure is higher in surveys with a higher number of items and a variety of recall periods. The low “other” expenditure estimate in WHS 2003 would have contributed to the fact that CHE estimates for WHS 2003 were higher than for the other surveys. The total household expenditure in the most recent year was lowest for NSS 2004 (median: US$ 829.6), a survey that did not collect disaggregated household expenditure data like the other surveys (Table 5).

Discussion

CHE is an important indicator of the financial protection offered to patients by a health system and has been estimated for health systems throughout the world using a variety of survey instruments. Although two publications in 2009 highlighted some of the difficulties of measuring OOP payments in household surveys,11,12 CHE continues to be estimated with survey methods that have not been validated. Our study demonstrates that CHE estimates can vary dramatically depending on the survey instrument used. This has major implications for health policy planning not only in India, but also in other low- and middle-income countries, especially if they are striving to offer universal health coverage.

The wide variation seen between surveys in the estimates of CHE was the result of differences in OOP payments for health care and in “other” household expenditure. In WHS 2003 and SAGE 2007–08, OOP payments for outpatient and inpatient care were two to three times higher than in the other surveys. Our results suggest that most of the variation in OOP payments for outpatient care resulted from the expenditure amount reported. On the other hand, much of the variation in OOP payments for inpatient care resulted from the proportion of households that reported having incurred such payments in the most recent year. This proportion was substantially higher in WHS 2003 and SAGE 2007–08 than in the other surveys. These findings suggest that survey design has a different effect on recall in the case of outpatient and inpatient OOP payments.

The types of items used to document outpatient OOP probably influenced their estimates. More items and more specific probing can improve respondent recall, particularly with respect to minor events.11,17,34 WHS 2003 and SAGE 2007–08 both had specific questions about dental care and care by traditional healers, whereas the consumer expenditure surveys did not. This may account for the higher outpatient OOP payments found in WHS 2003 and SAGE 2007–08. Additionally, in lengthy questionnaires respondents tend to invest less time in trying to recall events, and this may have been true for the consumer expenditure surveys.35 Conversely, it is possible that surveys focused on health, such as WHS 2003 and SAGE 2007–08, prime respondents to report events beyond the recall period, and this may lead them to overestimate OOP payments.1,35 However, studies have also shown that health care use is more commonly underreported than overreported.36 Such factors may have also contributed to the substantially higher proportion of households that reported OOP payments for inpatient care in WHS 2003 and SAGE 2007–08, by comparison with the other surveys. The health survey NSS 2004 documented every event involving inpatient care separately. The fact that it collected the data from the person who was treated might lead respondents to recall each event more accurately. Interestingly, however, in this survey, the proportion of households that reported inpatient OOP payments was practically the same as in NSS 2004–05 and NSS 2009–10. Since OOP payments for inpatient care in those households that reported such expenditure were similar across surveys, one might conclude that OOP payments for inpatient care are less sensitive to the number of items in the questionnaire than OOP payments for outpatient care. The obvious reason is that being an inpatient is a major event and hence any expenditure associated with this event is more accurately remembered by households.1,13,35 It should be noted that the indirect costs of health care, such as transportation and lost earnings, also contribute to the financial burden incurred by households, but we did not assess them because they were not consistently documented in the surveys.

CHE estimates will be inaccurate if the estimated expenditure on “other” household items is not accurately captured. If estimates of this other expenditure are too low, CHE may be overestimated because the denominator will be small. WHS 2003, which had the least number of items, lacked specific items for durable goods and had only a one-month recall period, was the survey that yielded the lowest estimates of “other” expenditure. Although the evidence suggests that estimates of household expenditure increase as the number of items in the questionnaire increases,17 in a health-focused survey it is highly impractical to ask questions as detailed as those that are included in consumer expenditure surveys. Thus, it is useful to note that SAGE 2007–08, which had 13 items, including durable goods, and various recall periods, had a higher estimate of “other” expenditure than WHS 2003, which included only four items. A single question, as in NSS 2004, does not appear to be enough to capture total household expenditure or expenditure on food. However, the 9 items used in SAGE 2007–08 for food expenditure provided an estimate similar to the estimates yielded by the consumer expenditure surveys. A one-week recall period yielded higher estimates of food expenditure. Other studies also suggest that one week is a more appropriate recall period for food expenditure than one month.37

We cannot comment on the accuracy of the CHE estimates derived from the different surveys since none of the surveys we examined can serve as a gold standard for measuring CHE. This highlights the need for validation studies to determine what questions and methods can most accurately capture CHE. These validation studies should not only examine the accuracy of the data, but also how to best use the data on OOP payments for outpatient care based on a relatively short recall period. Although a short recall period reduces recall error, it does not provide information about OOP payments for outpatient care in the population over a time frame more relevant for policy decisions, such as 6 months or one year.36 Simply multiplying the reported expenditure by as many times as necessary to obtain an estimate for the longer period, as we have done in this study, is equivalent to assuming that the expenditure is a recurrent one within a household, which is seldom the case. Hence, it probably caused overestimation of OOP payments in a one-year period for those households that reported such expenditure for a short recall period, and underestimation of OOP payments in the remaining households. This approach, which was used by others before us as well,9,10,29 also leads to an overestimation of the contribution made to CHE by OOP payments for outpatient care. It might be possible to more accurately estimate how much OOP payments for outpatient care contribute to CHE, by performing longitudinal panel surveys that assess the distribution of outpatient care in households across the population over a one-year period, but studies of this kind are too costly to conduct on a regular basis. However, an occasional longitudinal study can provide validation data that would allow cross-sectional survey data for outpatient OOP payments based on a one-month recall period to be adjusted to a one-year period more accurately than simple multiplication.

Because CHE estimates and OOP payments for health care varied widely across surveys, only data from surveys with comparable methods should be used to make longitudinal comparisons. Policy-makers should consider this limitation when formulating policies and programmes that depend on data from household surveys. Survey methods for estimating OOP payments for health care must undergo standardization to allow effective tracking and monitoring of the impact of policies designed to improve financial risk protection. With universal health coverage and financial risk protection being recognized as goals for health systems in many low- and middle-income countries, comparisons of CHE estimates from different household surveys, like the ones in this study, should be the first step towards planning validation studies of OOP payment data in these countries. This is especially important in India, given the launch of government-subsidized health insurance programmes for poor households21,38 and the recommendations to reduce OOP payments made by the High Level Expert Group on Universal Health Coverage.39


Acknowledgements

MZR, RD and LD are affiliated with the Public Health Foundation of India, New Delhi, India and the Sydney School of Public Health, University of Sydney, Australia. In addition, LD is affiliated with the Institute for Health Metrics and Evaluation, University of Washington, Seattle, United States of America.

Funding:

This work was supported in part by a grant from the Indian Council of Medical Research. MZR was supported by an Endeavour Research Fellowship from the Department of Education, Employment and Workplace Relations, Government of Australia, and an Australian Postgraduate Award from the University of Sydney, Australia. The funding bodies were not involved in the design, analysis or interpretation of this research.

Competing interests:

None declared.

References

Share