The Economic Burden of Cancers on Indian Households Ajay Mahal1, Anup Karan2*, Victoria Y. Fan3, Michael Engelgau4 102485 1 School of Public Health and Preventive Medicine, Monash University, Melbourne, Australia, 2 Public Health Foundation of India, New Delhi, India, and University of Oxford, Oxford, United Kingdom, 3 Center for Global Development, Washington, District of Columbia, United States of America, 4 South Asia Human Development Unit, The World Bank, Washington, District of Columbia, United States of America Abstract We assessed the burden of cancer on households’ out-of-pocket health spending, non-medical consumption, workforce participation, and debt and asset sales using data from a nationally representative health and morbidity survey in India for 2004 of nearly 74 thousand households. Propensity scores were used to match households containing a member diagnosed with cancer (i.e. cancer-affected households) to households with similar socioeconomic and demographic characteristics (controls). Our estimates are based on data from 1,645 households chosen through matching. Cancer-affected households experienced higher levels of outpatient visits and hospital admissions and increased out-of-pocket health expenditures per member, relative to controls. Cancer-affected households spent between Indian Rupees (INR) 66 and INR 85 more per member on healthcare over a 15-day reference period, than controls and additional expenditures (per member) incurred on inpatient care by cancer-affected households annually is equivalent to 36% to 44% of annual household expenditures of matched controls. Members without cancer in cancer-affected households used less health-care and spent less on healthcare. Overall, adult workforce participation rates were lower by between 2.4 and 3.2 percentage points compared to controls; whereas workforce participation rates among adult members without cancer were higher than in control households. Cancer-affected households also had significantly higher rates of borrowing and asset sales for financing outpatient care that were 3.3% to 4.0% higher compared to control households; and even higher for inpatient care. Citation: Mahal A, Karan A, Fan VY, Engelgau M (2013) The Economic Burden of Cancers on Indian Households. PLoS ONE 8(8): e71853. doi:10.1371/ journal.pone.0071853 Editor: Salomon M. Stemmer, Davidoff Center, Israel Received March 4, 2013; Accepted July 5, 2013; Published August 12, 2013 Copyright: ß 2013 Mahal et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Funding: AK was supported by the Wellcome Trust Capacity Strengthening Strategic Award to the Public Health Foundation of India and a consortium of United Kingdom universities. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing Interests: The authors have declared that no competing interests exist. * E-mail: anup.karan@dph.ox.ac.uk Introduction conclude that the aggregate economic impacts of cancer range from US$ 290 billion to US$ 900 billion worldwide [7–9]. The growth story of India has attracted much attention from This study assesses the economic burden of cancer on economists, particularly over the last decade where annual average households in India using a large cross-sectional household dataset growth rates of income per capita have averaged 5.7%. On from India. This burden can be potentially significant in India, average, an Indian enjoyed an income of US$ 840 in 2011, nearly given that low public sector allocations to health (ranging from 5 times as high as his or her counterpart in 1960. Health gains between 0.9% to 1.2% of GDP over the last few decades) and have accompanied improved economic outcomes. In India, life limited insurance options have forced households to rely on out of expectancy at birth increased from 39 years in 1950 to 65 years in pocket spending to finance their healthcare [10]. We assess several 2010, an increase of almost 67%. Even as these health and dimensions of this burden, including health-care utilization, out- economic gains are being experienced, however, Indian health of-pocket expenditure on inpatient and outpatient care, aggregate policymakers are faced with the challenge of non-communicable household expenditures on items other than healthcare, reliance chronic diseases (NCD). According to the Global Burden of on borrowing and asset sales, and labor force participation rates of Disease (GBD) Study, ischemic heart disease, diabetes, stroke and household members. We also examine the implications for other asthma collectively accounted for almost 18% of years of life lost household members when an individual with cancer lives in a due to premature death in India in 2010, almost double their share household, including their labor force participation, healthcare use in 1990 [1]. and health spending. To our knowledge, this is the first systematic Although not immediately apparent from the GBD study, India study of the economic burden of cancers on households in a also faces a major non-communicable disease burden from cancer. developing country. In India, each year nearly 600,000 cancer deaths occur [2,3]; and 5-year cancer prevalence among individuals aged 15 years and Methodology over was estimated to be 1.7 million in 2008 [4]. Cancer incidence is increasing, owing to a mix of risk factors such as changes in diet 1. Statistical Methods and lifestyle, the legacy of high tobacco consumption along with Previous calculations of the economic impacts of cancer population aging with cancer being more common in older cannot adequately describe household burden of cancer, for populations [5,6]. Cancers carry high levels of mortality and three reasons. First, their assessments of the economic burden of disability and require expensive treatments. Recent studies cancer are confounded by competing risks and co-morbidities. PLOS ONE | www.plosone.org 1 August 2013 | Volume 8 | Issue 8 | e71853 The Economic Burden of Cancer Because there are common risk factors that increase one’s risk 3. Definition of Cancer-Affected Household of multiple diseases (e.g. tobacco use or obesity that could be A household was defined as ‘cancer-affected’ if at least one of its linked to cancer, heart disease or diabetes), unadjusted estimates members was: (1) currently living with cancer; or (b) hospitalized of the economic burden attributed to any one condition are due to cancer in the year preceding the survey, whether or not likely to be upwardly biased. Second, socioeconomic and currently alive. This definition included most diagnosed cancers, demographic differences across households can influence with the exception of those who died from cancer in the year healthcare expenditures and other coping strategies in response preceding the survey without being hospitalized. In our survey to illness. Finally, aggregate calculations also include impacts of data, about 54% of the cancer cases were reported as hospitalized cancer on households unaffected by cancer, such as via an in the year preceding the survey, which is less than the annual overall slowing of economic growth that lowers employment hospitalization rates (of around 70%) reported for cancer patients rates and incomes [7]. in developed countries such as Australia for which similar data are For these reasons, we focus specifically on households contain- available [13]. Control households did not meet this definition. ing a member reporting as being diagnosed with cancer, i.e., ‘cancer-affected’ households, and compare their economic out- 4. Matching Variables comes with households with similar socioeconomic and demo- Cancer-affected households and control households were graphic characteristics that do not include a member with cancer. matched on several indicators, including educational status of Matching on observable household characteristics allows selection the head of household, house-type, land ownership, water source, of control households that are similar in socioeconomic charac- sanitation type, major source of livelihood, demographic structure teristics and demographic composition to the sample of households (number of children, young adults and the elderly, proportion of affected with cancer. household members that is female), caste (whether ‘scheduled caste We used propensity score matching (PSM) to compare the or tribe’ or not, or ‘other backward caste’ or not), religion, rural/ outcomes for a household containing a person with cancer to a urban status and whether a member of the household was covered matched household with no cancer case [11]. This generally by social or private insurance. Dummies for 71 NSS sub-areas of consists of two stages. In the first stage, the probability that a residence were also included. household contains a member with cancer (the ‘propensity score’) is predicted based on household socioeconomic and demographic 5. Outcome Variables characteristics. This (pre-processing) stage involves the estimation Healthcare utilization and expenditure of of a logit or probit model. In this paper the first stage consisted of households. We used several outcome variables for healthcare estimating the following logit model: utilization and spending. Healthcare utilization variables included hospital admissions per household member in the one year preceding the survey, length of hospital stay (in days) per ebXi P((Ci ~1)=Xi )~ household member, and outpatient visits per household member 1zebXi in the 15 days preceding the survey. Health expenditure variables Here Ci indicates whether household i contains a member with included household out-of-pocket (OOP) expenditure per member cancer. The vector Xi indicates household demographic and for outpatient care in the 15 days preceding the survey, and socioeconomic characteristics, and b is a vector of the parameters household spending per member for hospital (inpatient) care in the to be estimated. In the second stage, cancer-affected households one year preceding the survey. Household non-Medical consumption were matched to control households with similar propensity scores spending. Treatment expenses and loss of income associated using STATA, version 12.0. For balance checking, for each with cancer are likely to influence spending on items other than covariate used in the regression model that generated the healthcare in the cancer-affected household [14]. Household propensity scores, we compared the means between the cancer- consumption spending, net of healthcare expenditures, per affected households and matched control households using a t-test. member in the 15-day preceding the survey was used to capture this effect. 2. Main Source of Data Financing of out-of-Pocket healthcare spending. To This paper is based on anonymized survey data collected by assess the degree of illness-related financial stress faced by the National Sample Survey organization (NSSO), a department households, we inquired whether the household borrowed funds of the Indian Ministry of Statistics and Programme Implemen- or sold assets to support out of pocket spending on health care. tation. We have sought and obtained permission from the Two indicator variables were used to capture this: whether a NSSO to use this data in our research. This cross-sectional household borrowed funds or sold assets to finance inpatient (in survey collected nationally representative data on morbidity and the one year preceding the survey) and whether a household health care utilization in 2004 and covered nearly 74,000 borrowed or sold assets to finance outpatient care in the 15 days households (383,000 individuals of all ages). The sampling was preceding the survey [15–16]. based on a multi-stage stratified design [12]. In India, NSSO Work force participation among adults and health surveys are conducted roughly once a decade, with the children. We compared adult workforce participation rates last being in the year 2004. These surveys are developed with (the number of members aged 15 years and over who are currently extensive pilot testing and constitute the most reliable source of working, divided by all members in the household aged 15 years national-level information on household healthcare utilization and over) among cancer-affected household members and and expenditures. Information on socioeconomic and demo- members of matched control households. A similar measure was graphic characteristics, insurance status, healthcare use, medical constructed for comparing workforce participation rates among and non-medical spending, sources of healthcare financing and children aged 5–14 years of cancer-affected and control house- work status for all household members was collected in person holds. from a single key adult respondent. Implications for members without cancer in cancer- Affected households. In a cancer-affected household, the PLOS ONE | www.plosone.org 2 August 2013 | Volume 8 | Issue 8 | e71853 The Economic Burden of Cancer burden of cancer might be felt in the form of changed workforce cancer-affected and control households using the nearest neighbor participation, healthcare spending and non-medical consumption and stratification approaches (columns 2 and 3), a comparison that of members without cancer, or in the ways healthcare is financed excluded all households experiencing at least one death in the [17–20]. Therefore, we compared household-level outcomes (per preceding year, and a third that excluded 1% of the households person outpatient care visits, hospital admissions and health with the highest levels of cancer-related inpatient spending. spending) of members without cancer in cancer-affected house- Cancer-affected households experience an extra 15.8 to 17.3 holds and control households. We also compared healthcare hospital admissions per 100 members annually, and an extra 5.6 to utilization for ‘non-major conditions’ (specifically excluding stroke, 7.6 outpatient visits per 100 members in the 15 days preceding the diabetes, cancer, CVD and injuries) for the two sets of matched survey, compared to matched control households. Cancer-affected households. households also reported more days spent in hospitals per member. Per person outpatient visits of members without cancer 6. Subgroup Analysis by Socioeconomic Status (in cancer-affected households) were lower – by between 2 and 3 Low levels of insurance coverage might lead to the economic visits for every 100 members – than for household members in burden of cancer being felt differentially across households of matched controls in the 15 days preceding the survey. Per person different economic status. The survey did not collect information outpatient visits for non-major health conditions in the 15-days on indicators of wealth or income. Instead, we used the education preceding the survey were also lower in cancer-affected house- status of the head of household measured by years of schooling holds, by between 2.8 and 4.7 visits per 100 members, compared completed, as a proxy for household socioeconomic status (SES) to matched controls. [21]. To assess whether the economic burden varied with Columns 2 through 5 of Table 4 present results on household educational attainment of head of household, we conducted consumption and labor force participation. Out-of-pocket health additional matching analyses comparing cancer-affected and expenditures are significantly higher in households with cancer control households separately for two subsets of population: one compared to controls, by between INR 3,576 and INR 4,438 per consisting of households where the educational status of the head member for inpatient expenditures in the year preceding the of household was above the median (high SES households), and survey and by between INR 66 and INR 85 per member in the 15 one of households where the educational status of the head of days preceding the survey per person for outpatient visits. household was below or equal to the median (low SES) for the full Expenditures on non-major health conditions and on healthcare set of households. for individuals other than the person with cancer in cancer- affected households were lower, relative to controls. Spending per 7. Robustness Checks household member on non-medical consumption was lower in Several robustness checks were carried out by assessing results cancer-affected households than matched controls, by amounts from alternative matching methods such as nearest neighbor and between INR 27 and INR 85 per household member in the 15 stratification; by estimating results after excluding from the data days preceding the survey, although the differences were not the 1% of households with the most out-of-pocket expenditures on always statistically indistinguishable from zero. cancer inpatient treatment to lower the risk of a few outliers Cancer-affected households rely to a significantly greater extent influencing the results; and by excluding any household with at on borrowing or asset sales for financing out-of-pocket spending least one death because differential underreporting of deaths in on treatment than matched controls: between 33.5% and 39.2% household surveys. Further, to address possible risks of upward for inpatient spending in the year preceding the survey; and bias in estimates from using hospitalized cancer cases in our between 3.3% to 4.0% for outpatient spending in the 15-days definition of cancer-affected households, we estimated a separate preceding the survey. Current workforce participation rates set of results when generating propensity scores by including an among household members aged 15 years and above are lower indicator of hospitalization in the propensity score equation. by between 2.4% and 3.2% for households with cancer relative to matched controls. Workforce participation rates were higher among adult members in cancer-affected households when the Results individual with cancer was excluded from consideration – by Tables 1 and 2 present summary statistics for households with between 0.80% and 1.90% – but the results are not statistically cancer, unmatched control households and matched control significant. Differences in work-force participation rates among households (using nearest neighbor matching). Table 1 reveals children (aged 5–14 years) between cancer-affected and control that many of the means of the healthcare use and out-of-pocket households were also statistically indistinguishable from zero and spending indicators for matched cancer-affected and control varied from 0.01% to 20.30% in absolute magnitude. households were closer to each other than to unmatched controls. We also conducted a separate set of analyses, in which an Thus, a simple comparison of cancer households to a random indicator of hospitalization was included as a matching variable to selection of households not reporting cancer may yield upwardly generate propensity scores (in addition to the core socioeconomic biased estimates of the economic burden of cancer. Across a wide and demographic characteristics). Consequently, a large number range of household socioeconomic and demographic characteris- of households whose members were hospitalized for reasons other tics used for matching, the sample means of matched (control) than cancer were included as controls and lowered estimates of the households were considerably closer to those of cancer-affected economic impacts due to cancer. However, as indicated in Tables 5 households than their unmatched counterparts (Table 2): t-tests for and 6, although our estimates of the economic burden on differences in sample means of cancer-affected and control households are lower, they do not affect the basic direction of our households in the matched dataset (after nearest neighbor conclusions. matching) showed no statistically significant differences at the As part of a subgroup analysis, Table 7 presents results for 5% level. Column 5 of Table 2 shows the relevant t-statistic and analyses in which the population is divided into two sub-groups by the associated p-value (in parentheses) for the tests. education of household head: education greater the median (high Columns 2 through 5 of Table 3 present results for healthcare SES) and below or equal to the median (low SES). Comparisons utilization under multiple specifications: a direct comparison of between the two groups are thus based on outcomes relative to PLOS ONE | www.plosone.org 3 August 2013 | Volume 8 | Issue 8 | e71853 The Economic Burden of Cancer Table 1. Summary of Outcome Variables by Cancer-Affected or Control Households. Cancer-Affected Control Households – Control Households – Outcome Variable (reference period) Households Matched Unmatched Healthcare utilization Hospital admissions, per household member (1 year) 0.275 (0.26, 0.29) 0.110* (0.10, 0.12) 0.098 (0.097, 0.099) Public hospital admissions, per household member (1 year) 0.130 (0.12, 0.14) 0.048* (0.04, 0.06) 0.046 (0.0447, 0.0465) Length of hospital stay (in days), per household member (1 year) 4.349 (3.84, 4.86) 1.134* (0.91, 1.36) 0.940 (0.916, 0.965) Outpatient visits, per household member (15 days) 0.201 (0.18, 0.22) 0.145* (0.13, 0.16) 0.121 (0.120, 0.123) Outpatient visits of non-cancer patients, per household member (15 days) 0.112 (0.10, 0.13) 0.145* (0.13, 0.16) 0.120 (0.119, 0.122) Outpatient visits for non-major conditions, per household member (15 days) 0.073 (0.06, 0.08) 0.120* (0.10, 0.14) 0.098 (0.096, 0.099) Consumption Inpatient OOP expenditure (INR), per household member (1 year) 5311 (4514, 6108) 1,079* (702, 1457) 764 (733, 794) Outpatient OOP expenditure (INR), per household member (15 days) 118.33 (92.10, 144.56) 45.16* (33.94, 56.38) 34.16 (33.01, 35.30) Outpatient OOP expenditure (INR) on non-major health conditions, 15.54 (11.16, 19.92) 33.49* (24.18, 42.80) 24.34 (23.46, 25.22) per household member (15 days) Outpatient OOP expenditure (INR) for members without cancer, 28.03 (26.20, 35.87) 45.16* (33.94, 56.38) 33.14 (32.03, 34.24) per household member (15 days) Non-medical consumption expenditure (INR), per household member (15 days) 294 (252, 337) 321 (301, 342) 322 (319, 324) Percentage of households reporting borrowing or selling assets 51.40 (47.98, 54.82) 15.77* (13.28, 18.26) 15.72 (15.46, 15.98) to finance inpatient care (1 year) Percentage of households reporting borrowing/selling assets to 6.46 (4.78, 8.14) 3.11* (1.92, 4.30) 2.45 (2.34, 2.56) finance outpatient care (15 days) Workforce Participation Percentage of household members aged 15+ who are working 48.50 (46.69, 50.31) 50.90 (49.11, 52.69) 53.79 (53.59, 53.99) Percentage of household members without cancer aged 15+ who are working 52.58 (50.30, 54.86) 50.90 (49.11, 52.69) 53.84 (53.64, 54.04) Number of observations 821 824 72,582 Note: Estimates are based on calculations by authors using raw data from National Sample Survey data for 2004. Reference period refer to a household report in the period (15 days or 1 year) immediately preceding the survey. The data presented refer to all households, regardless of whether there was a death in the household. 95% confidence intervals are reported in parentheses underneath the means for each statistic. *indicates that the treatment and matched controls are significantly different at the 5% level. doi:10.1371/journal.pone.0071853.t001 their (respective) matched controls. In general, both high and low households experienced lower non-medical consumption that is SES cancer-affected households experienced more hospital both statistically significant and slightly larger as a share of non- admissions, more days in hospitals and increased outpatient visits. medical expenditures of matched controls (16.4%). However, high SES households experienced a greater increase in private hospital admissions, and a smaller increase in outpatient Discussion visits (per member) relative to matched controls. Indeed, high SES cancer-affected households experienced a decline in outpatient As expected, cancer-affected households experience a greater visits among members without cancer. This was not observed number of hospital admissions (inpatient stays) and outpatient among low-SES households: and almost the entire difference in visits, compared to matched controls. Given the low population the change in overall number of outpatient visits among the two coverage of health insurance in India and a poorly run public groups can be explained by the declining outpatient use by sector, it is not surprising to find a large burden of out-of-pocket members without cancer in high SES households. spending on households affected by cancer [22]. The additional After accounting for controls, high SES households also spend expenditures (per member) incurred on inpatient care by cancer- more per member on inpatient and outpatient care compared to affected households annually is equivalent to 36% to 44% of low SES cancer-affected households. The former also spend less annual household expenditures of matched controls (of INR out of pocket on the healthcare of members without cancer, and 9,988). Roughly 34% to 42% percent of all spending (INR 15,343 on non-major health conditions; and they borrowed less compared annually) by an average cancer-affected household is for out-of- to low SES cancer-affected households. Among high SES cancer- pocket treatment inpatient and outpatient expenses. Households affected households, adult labor force participation is lower than with higher SES spend more on healthcare out of pocket as a matched controls but statistically indistinguishable from zero. Low percentage of total spending (60%) relative to their low SES SES cancer-affected households experienced lower adult labor counterparts (53%). force participation relative to controls, with the difference also Out-of-pocket expenses associated with treatment for cancer being larger in absolute magnitude when compared to adult and any loss of income affect expenditures of a cancer-affected workforce participation among high SES households. Finally, high household on non-medical consumption. We would expect SES households also see a statistically insignificant difference in household non-medical consumption be lower when some non-medical consumption relative to controls (the decline is 14.5% members have cancer, unless the household was able to effectively of non-medical expenditures of matched controls). Low SES insure against associated financial risks [14–15]. We find that PLOS ONE | www.plosone.org 4 August 2013 | Volume 8 | Issue 8 | e71853 The Economic Burden of Cancer Table 2. Summary of Matching Variables by Cancer-Affected or Control Households. Cancer-Affected Control Households Control Households – Matching Variable Households – Matched Unmatched t-statistic Percentage rural households (%) 63.22 [59.91, 66.52] 61.88 [58.56, 65.19] 63.91 [63.56, 65.26] 0.56 (0.58) Age structure Percentage of members aged 0–14 years (%) 25.49 [24.02, 26.95] 25.29 [23.77, 26.81] 28.42 [28.25, 28.59] 0.18 (0.86) Percentage of members aged 15–29 years (%) 27.25 [25.79, 28.72] 27.92 [26.31, 29.52] 26.50 [26.33, 26.68] 20.55 (0.59) Percentage of members aged 60+ (%) 11.72 [10.36, 13.08] 11.56 [10.20, 12.93] 12.16 [12.00, 12.32] 0.16 (0.87) Level of schooling of household head Percentage Illiterate (%) 31.32 [29.41, 33.23] 32.58 [30.53, 34.63] 38.67 [38.44, 38.91] 20.87 (0.38) Percentage with primary schooling (%) 29.60 [27.83, 31.37] 29.20 [27.44, 30.96] 29.25 [29.06, 29.44] 0.32 (0.75) Percentage with secondary schooling (%) 9.87 [8.79, 10.94] 10.02 [8.80, 11.23] 7.65 [7.54, 7.76] 20.18 (0.86) Percentage with a graduate degree (%) 6.89 [5.78, 8.00] 5.90 [4.80, 7.00] 5.13 [5.02, 5.23] 1.24 (0.21) Percentage female (%) 50.25 [49.16, 51.34] 50.42 [49.12, 51.72] 49.07 [48.93, 49.20] 20.20 (0.85) Energy source, drinking water and sanitation Percentage of households using cooking gas (%) 34.35 [31.10, 37.60] 35.69 [32.42, 38.96] 28.58 [28.25, 28.90] 20.57 (0.57) Percentage of households with piped water access (%) 41.66 [38.28, 45.03] 41.66 [38.29, 45.02] 41.71 [41.35, 42.07] 0.00 (0.99) Percentage of households with latrine with septic tank (%) 38.98 [35.64, 42.31] 39.71 [36.36, 43.05] 33.32 [32.98, 33.67] 20.30 (0.76) Percentage of households with covered drainage (%) 20.10 [17.35, 22.84] 18.88 [16.21, 21.55] 17.44 [17.16, 17.72] 0.62 (0.53) Caste Percentage scheduled caste/tribe households (%) 22.53 [19.67, 25.39] 22.66 [19.80, 25.52] 28.48 [28.15, 28.81] 20.06 (0.95) Percentage ‘other’ backward caste households (%) 39.22 [35.88, 42.56] 39.22 [35.88, 42.56] 37.55 [37.19, 37.90] 0.00 (0.99) Religion Percentage Hindu (%) 81.49 [78.83, 84.14] 83.19 [80.54, 85.75] 79.44 [79.15, 79.74] 20.81 (0.37) Percentage Muslim (%) 12.06 [9.83, 14.29] 10.35 [8.27, 12.43] 11.81 [11.58, 12.04] 1.09 (0.27) Percentage self-employed (%) 24.48 [21.54, 27.43] 24.48 [21.54, 27.42] 24.16 [23.85, 24.48] 0.00 (0.99) Percentage insured (%) 3.41 [2.17, 4.65] 3.41 [2.17, 4.65] 2.20 [2.10, 2.31] 0.00 (0.99) Geographic region Percentage in northern region (%) 12.42 [10.17, 14.68] 13.64 [11.30, 15.99] 10.35 [10.13, 10.57] 20.74 (0.46) Percentage in western region (%) 15.10 [12.65, 17.55] 14.13 [11.75, 16.51] 16.23 [15.96, 16.50] 0.56 (0.58) Percentage in southern region (%) 26.67 [23.65, 29.70] 27.41 [24.36, 30.45] 23.74 [23.43, 24.05] 20.33 (0.74) Percentage in eastern region (%) 19.61 [16.89, 22.33] 18.88 [16.21, 21.55] 19.08 [18.79, 19.36] 0.37 (0.71) Percentage in central region (%) 19.49 [16.78, 22.20] 19.49 [16.78, 22.19] 19.11 [18.82, 19.39] 0.00 (0.99) Number of observations 821 824 72,582 1,645 Note: Estimates are based on calculations by authors using raw data from National Sample Survey data for 2004. The data presented refer to all households, regardless of whether there was a death in the household. Propensity score calculations used 71 NSS sub-region dummies rather than geographic region. The t-test reported in column 5 refers to a comparison of means between matched cancer-affected and control households; p-values are reported in parentheses below the t-statistic. *indicates that the treatment and matched controls are significantly different at the 5% level. doi:10.1371/journal.pone.0071853.t002 cancer-affected households have lower expenditures per person. Households also cope with the costs of cancer by increasing the The estimated adverse impacts on consumption spending per burden on other members on unaffected members [14,17–20,23]. person net of health spending ranged between INR 649 to INR Outpatient visits and out-of-pocket healthcare spending per 2,058 annually, indicating that Indian households are unable to member were both significantly lower in cancer-affected house- fully protect themselves against financial risks from cancer, relative holds compared to matched controls, once the individual with to controls. However, this decline is considerably less than the cancer was excluded from the analysis. However, when data are additional estimated inpatient out-of-pocket expenditures (INR broken down by SES, this conclusion holds only for higher SES 3,576 to INR 4,438). This conclusion also holds for households households. We also compared healthcare utilization and expen- with different SES. Higher SES cancer-affected households diture for non-major conditions (i.e. other than cancer, stroke, experience lower (by INR 1,545) non-medical expenses and heart disease, injuries and diabetes) on the assumption that the higher inpatient care out of pocket expenses of INR 6,127, relative occurrence of a serious condition would lower healthcare use for to matched controls. Similarly low SES households had non- the former. We found that per person outpatient care use for non- medical expenses that were INR 992 less than matched controls, major conditions was lower by between 3 and 5 visits per 100 but higher annual inpatient expenses of INR 3,223. Thus persons among cancer-affected households relative to matched households are relying on other ways to protect their non-health controls, and this conclusion holds even if data are broken down spending against cancer-related expenditures. into two SES groups. Expenditures on non-major health PLOS ONE | www.plosone.org 5 August 2013 | Volume 8 | Issue 8 | e71853 The Economic Burden of Cancer Table 3. Effects of Cancers on Household Healthcare Utilization in India. Excluding Excluding Households with 1% All Matched All Matched Households with Most Expensive Households (by Households (by Death (by nearest Cancer Cases (by Outcome Indicator (reference period) nearest neighbor) stratification) neighbor) nearest neighbor) Hospital admissions, per household member (1 year) 0.165 (,0.01) 0.173 (,0.01) 0.165 (,0.01) 0.158 (,0.01) Public hospital admissions, per household member (1 year) 0.082 (,0.01) 0.085 (,0.01) 0.074 (,0.01) 0.081 (,0.01) Length of hospital stay, per household member (1 year) 3.215 (,0.01) 3.378 (,0.01) 3.323 (,0.01) 3.303 (,0.01) Outpatient visits, per household member (15 days) 0.056 (,0.01) 0.069 (,0.01) 0.076 (,0.01) 0.057 (,0.01) Public outpatient visits, per household member (15 days) 0.029 (,0.01) 0.035 (,0.01) 0.031 (,0.01) 0.025 (,0.01) Outpatient visits of non-cancer patients per household 20.033 (,0.01) 20.021 (0.013) 20.018 (0.146) 20.031 (0.01) member (15 days) Outpatient visits for non-major health conditions, 20.047 (,0.01) 20.032 (,0.01) 20.028 (,0.01) 20.039 (,0.01) per household member (15 days) Number of treated (and matched control) observations 821 (824) 821 (71,761) 735 (743) 813 (817) Notes: INR = Indian Rupees. Values in parentheses refer to p-values that the matched cancer-affected and control outcomes differ in a two-tailed test. ‘Non-major’ health conditions refer to all health conditions except cancer, heart disease, stroke, injuries & diabetes. doi:10.1371/journal.pone.0071853.t003 conditions were also lower, although in this case, the results are workers are mostly employed in the informal sector with limited being driven primary by lower expenses among high SES social security benefits. Overall, cancer-affected households have a households (Table 7). The lack of change in outpatient care use lower labor force participation rate than matched controls of among members without cancer in low SES households likely between 2.4 and 3.2 percentage points. However, workforce reflects the already low outpatient care use among them, so further participation of members without cancer in cancer-affected declines associated with cancer are unlikely to be significant. households was 0.8 to 1.9 percentage points higher than control Income losses from days missed at work by the sick person and households, although the results are statistically indistinguishable their caregivers could arise because more than 90% of India’s from zero (see studies that analyzed workforce participation in Table 4. Effects of Cancers on Household Consumption and Workforce Participation in India. Excluding Excluding Households with 1% All Matched All Matched Households with most expensive Households (by Households (by Death (by nearest cancer cases (by Outcome Indicator (reference period) nearest neighbor) stratification) neighbor) nearest neighbor) Household Consumption Inpatient OOP expenditure (INR), per household member (1 year) 4,232.08 (,0.01) 4,437.66 (,0.01) 4,044.30 (,0.01) 3,575.74 (,0.01) Outpatient OOP expenditure (INR), per household member (15 days) 73.18 (,0.01) 78.91 (,0.01) 85.31 (,0.01) 65.68 (,0.01) Outpatient OOP expenditure (INR) on non-major health conditions, 217.95 (,0.01) 212.80 (,0.01) 210.99 (,0.01) 211.99 (,0.01) per household member (15 days) Outpatient OOP expenditure (INR) for members without cancer, 217.13 (0.01) 211.40 (,0.01) 210.55 (0.05) 213.78 (0.03) per household member (15 days) Non-medical consumption expenditure (INR), per household 226.69 (0.265) 249.75 (0.020) 284.60 (0.033) 232.15 (0.174) member (15 days) Percentage of households reporting borrowing or selling assets 0.356 (,0.01) 0.361 (,0.01) 0.392 (,0.01) 0.335 (,0.01) to finance inpatient care (1 year) Percentage of households reporting borrowing/selling assets 0.033 (,0.01) 0.040 (,0.01) 0.039 (,0.01) 0.033 (,0.01) to finance outpatient care (15 days) Workforce Participation Percentage of household members aged 15+ who are working 22.40 (0.07) 23.00 (,0.01) 22.80 (,0.06) 23.20 (0.02) Percentage of household members without cancer aged 15+ who 1.70 (0.26) 1.10 (,0.36) 1.90 (0.26) 0.80 (0.59) are working Percentage of household members aged 5–14 years who are working 20.30 (0.50) 20.30 (0.32) 0.01 (0.91) 20.10 (0.88) Number of treated (and matched control) observations 821 (824) 821 (71,761) 735 (743) 817(813) Notes: INR = Indian Rupees. Values in parentheses refer to p-values that the matched cancer-affected and control outcomes differ in a two-tailed test. ‘Non-major’ health conditions refer to conditions other than cancer, heart disease, stroke, injuries and diabetes. doi:10.1371/journal.pone.0071853.t004 PLOS ONE | www.plosone.org 6 August 2013 | Volume 8 | Issue 8 | e71853 The Economic Burden of Cancer Table 5. Robustness Check – Effects of Cancers on Healthcare Utilization in India. Excluding Households All Matched All Matched Excluding Households with 1% most expensive Households (by Households (by with Death (by cancer cases (by nearest Outcome Indicator (reference period) nearest neighbor) stratification) nearest neighbor) neighbor) Hospital admissions per household member (1 year) 0.045 (,0.01) 0.044 (,0.01) 0.030 (0.013) 0.041 (,0.01) Public hospital admissions per household member (1 year) 0.021 (0.03) 0.027 (,0.01) 0.018 (0.058) 0.023 (0.017) Length of hospital stay per household member (1 year) 2.059 (,0.01) 2.142 (,0.01) 2.028 (,0.01) 2.013 (,0.01) Outpatient visits per household member (15 days) 0.057 (,0.01) 0.052 (,0.01) 0.056 (,0.01) 0.036 (,0.01) Public outpatient visits per household member (15 days) 0.027 (,0.01) 0.025 (,0.01) 0.032 (,0.01) 0.028 (,0.01) Outpatient visits of non-cancer patients per household 20.033 (,0.01) 20.037 (,0.01) 20.037 (,0.01) 20.051 (,0.01) member (15 days) Outpatient visits for non-major health conditions, per 20.040 (,0.01) 20.040 (,0.01) 20.046 (,0.01) 20.052 (,0.01) household member (15 days) Number of Treated (and Matched Control) Observations 821 (801) 821 (71,555) 735 (725) 813 (800) Notes: Robustness check refers to propensity scores generated with a hospitalization. INR = Indian Rupees. Values in parentheses refer to p-values that the matched cancer-affected and control outcomes differ in a two-tailed test. ‘Non-major’ health conditions refer to conditions excluding cancer, heart disease, stroke, injuries & diabetes. doi:10.1371/journal.pone.0071853.t005 other disease contexts [24,25]). These estimates do not indicate than in high SES households. These ideas do not carry over, impacts on hours of work or earnings, but suggest that the adverse however, for child workers where the differences between cancer- impact of cancer on household earnings may be partly countered affected households and matched controls are small and statisti- by rising workforce participation among healthier household cally insignificant, even when data are broken down by SES. One members. Even here though, adult workforce participation possible explanation is that most of the response in household declines much more in low SES households (relative to controls) economic activities associated with cancer occurs among adults. Table 6. Robustness Check – Effect of Cancers on Household Consumption and Workforce Participation in India. Excluding Households with Excluding 1% most All Matched All Matched Households with expensive cancer Households (by Households (by Death (by nearest cases (by nearest Outcome Indicator (reference period) nearest neighbor) stratification) neighbor) neighbor) Household Consumption Inpatient OOP expenditure (INR), per household member (1 year) 3,403.94 (,0.01) 3,401.52 (,0.01) 2,942.23 (,0.01) 2,534.72 (,0.01) Outpatient OOP expenditure (INR), per household member (15 days) 72.14 (,0.01) 66.31 (,0.01) 75.42 (,0.01) 49.50 (,0.01) Outpatient OOP expenditure (INR) on non-major health conditions, per 216.58 (,0.04) 220.82 (,0.01) 218.02 (,0.01) 229.67(,0.01) household member (15 days) Outpatient OOP expenditure (INR) for members without cancer, per 218.16 (,0.06) 223.95 (,0.01) 229.96 (,0.01) 219.95 (,0.01) household member (15 days) Non-medical consumption expenditure (INR), per household member (15 216.91 (0.509) 236.27 (0.094) 244.23 (0.089) 237.16 (0.164) days) Percentage of households reporting borrowing or selling assets to finance 0.171 (,0.01) 0.169 (,0.01) 0.154 (,0.01) 0.137 (,0.01) inpatient care (1 year) Percentage of households reporting borrowing/selling assets to finance 0.028 (0.01) 0.032 (,0.01) 0.035 (,0.01) 0.020 (,0.072) outpatient care (15 days) Workforce Participation Percentage of household members aged 15+ who are working 20.50 (0.683) 21.80 (0.051) 22.40 (0.088) 22.00 (0.129) Percentage of household members without cancer aged 15+ who 3.60 (0.018) 2.30 (0.052) 2.20 (0.174) 2.00 (0.168) are working Percentage of household members aged 5–14 years who are working 20.60 (0.18) 20.30 (0.27) 20.00 (0.96) 20.30 (0.46) Number of treated (and matched control) observations 821 (801) 821 (71,555) 735 (725) 813 (800) Notes: Robustness check refers to propensity-score matching that included hospitalization. INR = Indian Rupees. Values in parentheses refer to p-values that the matched cancer-affected and control outcomes differ in a two-tailed test. ‘Non-major’ health conditions refer to conditions other than cancer, heart disease, stroke, injuries and diabetes. doi:10.1371/journal.pone.0071853.t006 PLOS ONE | www.plosone.org 7 August 2013 | Volume 8 | Issue 8 | e71853 The Economic Burden of Cancer Table 7. Burden of Cancer by Educational Status of Head of Household, 2004. Outcome Indicator (reference period) By Education of Household Head Above Median Below or Equal to (low SES) Median (high SES) Healthcare Utilization Hospital admissions, per household member (1 year) 0.160 (,0.01) 0.168 (,0.01) Public hospital admissions, per household member (1 year) 0.067 (,0.01) 0.090 (,0.01) Length of hospital stay, per household member (1 year) 3.412 (,0.01) 3.355 (,0.01) Outpatient visits, per household member (15 days) 0.047 (0.02) 0.089 (,0.01) Public outpatient visits, per household member (15 days) 0.026 (,0.01) 0.047 (,0.01) Outpatient visits of non-cancer patients per household member (15 days) 20.034 (0.09) 20.007 (0.61) Outpatient visits for non-major health conditions, per household member (15 days) 20.027 (0.06) 20.026 (0.01) Household Consumption Inpatient OOP expenditure (INR), per household member (1 year) 6,127.36 (,0.01) 3,223.36 (,0.01) Outpatient OOP expenditure (INR), per household member (15 days) 79.32 (0.02) 69.90 (,0.01) Outpatient OOP expenditure (INR) on non-major health conditions, per household member (15 days) 236.04 (,0.01) 21.15 (0.75) Outpatient OOP expenditure (INR) for members without cancer, per household member (15 days) 240.52 (0.04) 1.59 (0.76) Non-medical consumption expenditure (INR), per household member (15 days) 263.50 (0.22) 240.80 (0.02) Percentage of households reporting borrowing or selling assets to finance inpatient care (1 year) 0.311 (,0.01) 0.426 (,0.01) Percentage of households reporting borrowing/selling assets to finance outpatient care (15 days) 0.003 (0.83) 0.071 (,0.01) Workforce Participation Percentage of household members aged 15+ who are working 21.60 (0.40) 23.80 (0.04) Percentage of household members without cancer aged 15+ who are working 1.50 (0.50) 1.10 (0.59) Percentage of household members aged 5–14 years who are working 0.20 (0.59) 20.40 (0.65) Number of treated (and matched control) observations 467 (470) 353 (354) Note: Authors’ estimates using a dataset consisting of only matched households using the nearest neighbor method. Values in parentheses refer to p-values. doi:10.1371/journal.pone.0071853.t007 Alternatively, given that only 2.4% of the sample of children aged burden of cancer – be it in terms of public subsidies or out-of- 5–14 years reported working in the survey, understanding the pocket spending by households – may not be as large as one might burden of cancer on children in the form of work participation conclude in the absence of matching. Moreover, we use a and/or caregiving activities with any statistical precision may nationally representative household survey that contains detailed require collecting larger samples. information on healthcare utilization and out-of-pocket expendi- Households cope with increased requirements for health ture on health services by individuals, the methods by which out- spending associated with cancer by borrowing or selling assets. of-pocket spending was financed, along with information on However, the extent to which they rely on sales of assets and individual-level workforce participation. Finally, our study findings borrowing varies inversely with economic status: there is an almost rely on multiple checks for robustness. Our results hold up across 11 percentage point difference in household reliance on borrowing different matching methods, as well as matching after excluding and sale of assets for financing inpatient care in high SES cancer- households with a death from any cause, and matching after affected households, compared to low SES households. In sum, excluding the top 1% households with the most out of pocket not only can cancer have long-term implications for household spending on cancer treatment. economic well-being if borrowing costs are high or if income There are limitations to our study. Our identification of cancer- earning assets are sold, but also it may exacerbate pre-existing affected households relies on self-reports, which may lead to economic inequalities. Our results also suggest a difference in the inaccurate estimation of cancer cases. It is noteworthy though, that way households at different levels of SES respond to the economic the prevalence of cancer cases in the survey data we used was challenges of cancer. Specifically, higher-SES households seem to 0.22%, which was very close to the prevalence estimate reported adjust by lowering their spending on outpatient care for members by GLOBOCAN (2008) of 0.23% among individuals aged 15- without cancer and for non-major health conditions, along with years or older. some decline in non-medical expenses and reliance on borrowing/ From the health survey we used, there are an estimated 102 assets. Low SES households, however, have a narrower set of thousand deaths due to cancer in 2004, compared to 634 thousand options: more borrowing and sale of assets and lower non-medical deaths per GLOBOCAN (2008) estimates based on cancer consumption. registries in India, and about 560 thousand deaths in 2000 based Our study has multiple strengths. Matching cancer-affected on the Million Deaths Study [3,4]. Underestimates of cancer households to ‘control’ households (not affected by cancer) on deaths arise partly because 49% of deaths reported in the survey large set of observable socioeconomic and demographic charac- are not assigned a cause. However, another reason is that deaths teristics reduces confounding that arises from non-random (from any cause) are undercounted in NSS surveys. Cancer impact assignment of cancer. Our results suggest that the economic estimates could be biased downwards if healthcare utilization and PLOS ONE | www.plosone.org 8 August 2013 | Volume 8 | Issue 8 | e71853 The Economic Burden of Cancer expenditures are concentrated in the time immediately preceding have in the data are relatively more severe cancer cases thereby death, and a disproportionate number of cancer deaths are what we are capturing is, in fact, the economic burden of cancer excluded. On the other hand, a disproportionate share of cancer for the more severe cases. deaths included hospitalizations (95% of cancer cases resulting in Our estimates of the economic burden of cancer are for 2004, deaths reported in the survey were hospitalized in the preceding 12 and can be expected to have risen considerably since then, both on months) and this could bias estimates of the economic burden of account of general inflation as well on account of changes in the cancer upwards given the definition of a cancer-affected household technology of cancer treatment. Information on the impact of the includes all households that had a member hospitalized due to latter on treatment costs is unavailable for India. However, cancer in the year preceding the survey. As noted earlier, consumer price index data suggest that the treatment costs are estimated annual hospitalization rate for cancer in our survey likely to be at least 70% higher at present compared to 2004 [28]. population in cancer-affected households is not inordinately high at 54%, well below rates seen in a developed country with Conclusions excellent prevalence and hospital separations data, e.g. 70% in Our finding provides much better understanding of economic Australia. However, we also addressed this issue directly by burden of cancer at the household level. To our knowledge, ours is undertaking additional analyses limited to households that did not the first paper to estimate this burden for households in a experience deaths (since the incidence of hospitalization was much developing country. Our use of matching also helps to address higher in hospitalized cases), irrespective of cause. Although the potential confounding associated with socioeconomic and demo- absolute values of coefficient estimates are lower, the direction of graphic differences in estimating the economic burden of cancer. the results is similar. It is noteworthy that if households with cancer The major policy implication of our findings is the need for deaths are excluded, we end up with an estimate of 0.93 million protection against the financial risks from cancer for Indian cancer cases from NSS data in 2004 (after using sampling weights) households. This is not surprising given the heavy reliance on out versus 1.07 million for GLOBOCAN (2008). of pocket spending in financing healthcare in India: government Matching methods cannot account for unobservable factors that financing accounts for only about one-fourth of India’s aggregate drive household risks for cancer. For example, our 2004 data does health spending of 4.5% of GDP, and most of the residual not include information on tobacco and alcohol consumption, spending is in the form of out of pocket spending by households. dietary history or obesity in the household. Nor do we have Our results also show that households across the SES spectrum information on the occupational history of household members face a large economic risk from cancer, and the risk seems that could affect cancer risks. Unavailability of this information particularly serious for low SES households who rely on borrowing may lead to the exclusion of at-risk households from matched and asset sales to a much greater extent to finance their healthcare. controls and can bias our estimated economic burden of cancer Unlike in 2004 the year our survey was conducted, over the last upwards if the excluded households are also at risk for acquiring few years, a number of publicly funded health insurance schemes other serious illnesses and increased health care use. Our matched have emerged in India [29]. These schemes and existing control households reported higher levels of health care use and subsidized public sector health facilities are likely to have provided out of pocket spending compared to the average survey household, improved protection against the economic costs of cancer. so this risk is lower. Separately, we tried to address this issue by However, coverage is not comprehensive in many of these including a hospitalization indicator in constructing propensity programs and non-poor households are often ineligible to benefit scores prior to matching. As a result, the estimated economic from such programs. For example, the Rashtriya Swasthya Bima burden of cancer was lower, but the direction of the conclusions Yojana (RSBY) is a fully government funded insurance scheme remained unchanged. To effectively address this subject would targeted only at the poor and provides coverage for hospital-based require longitudinal data, which the NSS surveys do not currently treatments including cancer, at both private and public healthcare collect. facilities. Although the scheme now covers more than one hundred Biased estimates may also have resulted because we did not million people in India, the financial cover it provides is relatively have information on differences in physical access (such as distance small (INR 30,000) for a family of four and is inadequate given the to health facilities) which can influence the diagnosis of cancer and financial costs of cancer treatment [30]. More generous publicly any associated health spending. The absence of information on financed schemes are in place in a few Indian states such as distance to health facilities from the propensity score equation Andhra Pradesh and Tamil Nadu, and they cover a broader could potentially cause an upward bias in our estimates of out of category of households than just the poor but these comprise only pocket spending and healthcare use if, as is likely, individuals with a subset of India’s population and face issues of long-term financial better access also happen to be wealthier on average. We sought to sustainability. address this through the inclusion of three sets of variables in the Apart from the limited financial cover for the poor, a key propensity score equation, two of them being a range of controls of conclusion of our paper is that higher SES households also face a living conditions (access to drains, piped water, etc.) and rural serious financial risk from cancer. The problem of lack of coverage residence that are likely to be positively correlated with physical of the non-poor is exacerbated by various age-limits that exist for access [26]. In addition, we used 71 indicators of location used by voluntary private insurance coverage in India, and a public sector the NSSO to indicate sub-regions with varying economic and that is underfunded and provides poor quality services for all climatic characteristics and these may capture many region- population sub-groups. specific differences related to physical access to health services. These gaps in coverage necessitate thinking more broadly about The survey also did not collect information on the severity of pooling mechanisms than the strategy of just targeting the poor cancer. If information on severity were available, we could have with publicly funded programs. However, expanded coverage for used a variation of the propensity score matching method for the cancer (and other conditions requiring expensive treatment such as multiple treatment case to assess the economic burden on heart disease) is challenging in a resource constrained health sector households by severity of condition [27]. In the absence of this such as in India as it competes with other pressing healthcare information, all we can estimate is the average burden for different needs for the poor, including the needs for child survival, malaria levels of cancer severity. It is nevertheless possible that what we and tuberculosis, as it charts out a path for achieving universal PLOS ONE | www.plosone.org 9 August 2013 | Volume 8 | Issue 8 | e71853 The Economic Burden of Cancer coverage over the next 10–15 years [31]. An interim mechanism Acknowledgments may well be a regulatory environment that encourages private The authors acknowledge helpful research assistance from Rachel sector pooling to fund cancer care and expanded publicly funded Silverman and Mingzhu Zhou. cover for the poor for expensive to treat medical conditions. From a longer term perspective, mechanisms to lower the risk of Author Contributions acquiring cancer, including improved screening might also be prioritized. Conceived and designed the experiments: AM AK ME VF. Performed the experiments: AM AK VF. Analyzed the data: AM AK. Contributed reagents/materials/analysis tools: AM AK ME VF. Wrote the paper: AM AK ME VF. References 1. Institute of Health Metrics and Evaluation (2013) GBD Profile: India. Available: 17. Lilly M, Laporte A, Coyte P (2010) Do they care too much to work? The http://www.healthmetricsandevaluation.org/sites/default/files/country- influence of caregiving intensity on the labour force participation of unpaid profiles. Accessed 2013 June 26. caregivers in Canada. Journal of Health Economics 29: 895–903. 2. Ferlay J, Shin HR, Bray F, Forman D, Mathers C, et al. (2010) Estimates of 18. Ciani E (2011) Informal Adult Care and Caregivers’ Employment in Europe. worldwide burden of cancer in 2008: GLOBOCAN 2008. International Journal Labour Economics 19: 155–164. of Cancer 127: 2893–2917. 19. Shaw W, Patterson T, Semple S, Ho S, Irwin M, et al. (1997) Longitudinal 3. Dikshit R, Gupta PC, Ramasundarahettige C, Gajalakshmi V, Aleksandrowicz analysis of multiple indicators of health decline among spousal caregivers. Annals L, et al. (2012) Cancer mortality in India: a nationally representative survey. of Behavioural Medicine 19: 101–109. Lancet 379: 1807–1816. 20. Roberts A (1992) The labor market consequences of family illness. Journal of 4. GLOBOCAN (2008) The GLOBOCAN project. Available: http://globocan. Mental Health Economics and Policy 2: 183–195. iarc.fr/. Accessed 2013 February 10. 21. Jolliffe D (1997) Whose education matters in the determination of household 5. Popkin B, Horton S, Kim S, Mahal A, Shuigao J (2001) Trends in diet, income? Evidence from a developing country. Discussion paper #39. nutritional status, and diet-related non-communicable diseases in China and Washington, D.C.: International Food Policy Research Institute, Food, India: the economic costs of the nutrition transition. Nutrition Reviews 59(12): Consumption and Nutrition Division. 379–390. 22. Mahal A (2010) Health financing in India. In: Mahal A, Debroy B and Bhandari 6. Murthy N, Chaudhry K, Rath G (2008) Burden of cancer and projections for L, editors, India Health Report. New Delhi: Business Standard Press. 109–126. 2016, Indian scenario: gaps in the availability of radiotherapy treatment facilities. Asia Pacific Journal of Cancer 9: 671–677. 23. Yamano T, Jayne T (2004) Measuring the impacts of working-age adult 7. Bloom D, Cafiero E, Jane-Llopis E, Abrahams-Gessel S, Bloom L, Fathima S, et mortality on small-scale farm households in Kenya. World Development 32: 91– al. (2011) The Global Economic Burden of Non-Communicable Diseases. 119. Boston, MA: Harvard School of Public Health and World Economic Forum. 24. Harris A (2008) Chronic disease and labour force participation in Australia: an 8. Abegunde DO, Mathers CD, Adam T, Ortegon M, Strong K (2007) The endogenous multivariate probit analysis of clinical prevalence data. Research burden and costs of chronic diseases in low-income and middle-income paper no. 25/08. Clayton, Australia: Monash University, Centre of Health countries. Lancet 370: 1929–1938. Economics. 9. John R, Ross H (2011) Global cancer facts & figures. 2nd edition. Atlanta: 25. Passey ME, Shrestha RN, Bertram MY, Schofield DJ, Vos T, et al. (2012) The American Cancer Society. impact of diabetes prevention on labour force participation and income of older 10. Reddy K, Patel V, Jha P, Paul V, Shiva Kumar AK, et al. (2011) Towards Australians: an economic study. BMC Public Health 12: 16. achievement of universal health care in India by 2020: A call to action, The 26. Malhotra C, Kyung Do Y (2012) Socioeconomic disparities in health system Lancet 377: 760–8. responsiveness in India. Health Policy and Planning doi:10.1093/heapol/ 11. Rosenbaum P, Rubin D (1983) The central role of the propensity score in czs051. observational studies for causal effects. Biometrika 70: 41–55. 27. Morena-Serra R (2009) Health programme evaluation by propensity score 12. National Sample Survey Organization (NSSO) (2006) Morbidity, health care matching: Accounting for treatment intensity and health externalities with an and the conditions of the aged, NSSO 60th Round (January-June 2004). New application to Brazil. Health Economics Group Working Paper 09/05, Delhi: National Sample Survey Organisation. Ministry of Statistics and University of York, United Kingdom. Programme Implementation. Government of India. 28. Government of India (2012) Economic Survey 2011–12. New Delhi, India: 13. Australian Institute of Health and Welfare (2012) Australian Health Statistics Government of India, Ministry of Finance. 2010–11. Health Services Series no. 43, Category no. HSE 117. Canberra: 29. Fan VY, Karan A, Mahal A (2012) State health insurance and out-of-pocket Australia. health expenditures in Andhra Pradesh, India. International Journal of Health 14. Gertler P, Gruber J (2002) Insuring consumption against illness. American Care Finance and Economics 12: 1–27. Economic Review 92(1): 51–70. 30. Ministry of Labour and Employment (2008) RSBY Guidelines: Guide- 15. Islam A, Maitra P (2012) Health shocks and consumption smoothing in rural lines13.3.08_rev.pdf (New), Annexure I. Available: http://www.rsby.gov.in/ households: Does microcredit have a role to play? Journal of Development Documents.aspx?id = 25/.Accessed 2013 June 30. Economics 97(2): 232–43. 31. John Jacob T, Dandona L, Sharma VP, Kakkar M (2011) Continuing challenge 16. Wagstaff A (2008) Measuring financial protection in health. Policy Research of infectious disease in India. The Lancet 377: 252–269. Working Paper 4554. Washington, D.C.: The World Bank. PLOS ONE | www.plosone.org 10 August 2013 | Volume 8 | Issue 8 | e71853