Downloaded from http://bmjopen.bmj.com/ on November 30, 2015 - Published by group.bmj.com Open Access Research Changes in addressing inequalities 102438 in access to hospital care in Andhra Pradesh and Maharashtra states of India: a difference-in-differences study using repeated cross-sectional surveys Mala Rao,1,2 Anuradha Katyal,3 Prabal V Singh,3 Amit Samarth,4 Sofi Bergkvist,3 Manjusha Kancharla,5 Adam Wagstaff,6 Gopalakrishnan Netuveli,7 Adrian Renton1 To cite: Rao M, Katyal A, ABSTRACT Singh PV, et al. Changes in Strengths and limitations of the study Objectives: To compare the effects of the Rajiv addressing inequalities Aarogyasri Health Insurance Scheme of Andhra ▪ This study uses two cross-sectional surveys to in access to hospital care in Andhra Pradesh and Pradesh (AP) with health financing innovations compare changes between the Indian states of Maharashtra states of India: a including the Rashtriya Swasthya Bima Yojana (RSBY) Andhra Pradesh and Maharashtra, in hospital difference-in-differences in Maharashtra (MH) over time on access to and out- inpatient care-related expenditures and beha- study using repeated cross- of-pocket expenditure (OOPE) on hospital inpatient viours before and after the roll-out of the sectional surveys. BMJ Open care. Aarogyasri and Rashtriya Swasthya Bima Yojana 2014;4:e004471. Study design: A difference-in-differences (DID) study schemes. doi:10.1136/bmjopen-2013- using repeated cross-sectional surveys with parallel ▪ The study, based on a survey of 18 696 house- 004471 control. holds, has shown that health innovations in Setting: National Sample Survey Organisation of India Andhra Pradesh had a greater beneficial effect on ▸ Prepublication history for (NSSO) urban and rural ‘first stratum units’, 863 in AP hospital inpatient care-related expenditures and this paper is available online. and 1008 in MH. access than innovations in Maharashtra. The To view these files please Methods: We used two cross-sectional surveys: as a Aarogyasri scheme is likely to have contributed visit the journal online baseline, the data from the NSSO 2004 survey collected to these impacts in Andhra Pradesh. (http://dx.doi.org/10.1136/ bmjopen-2013-004471). before the Aarogyasri and RSBY schemes were launched; ▪ The study also highlights the implications of the and as postintervention, a survey using the same findings for policy and practice, and additional Received 19 November 2013 methodology conducted in 2012. interventions are necessary to address gaps in Revised 6 May 2014 Participants: 8623 households in AP and 10 073 the availability of and access to care. Accepted 13 May 2014 in MH. ▪ The study is only able to compare the effects of Main outcome measures: Average OOPE, large OOPE health innovations over time across the two and large borrowing per household per year for inpatient states but does not allow the drawing of infer- care, hospitalisation rate per 1000 population per year. ences on the impacts of individual initiatives. Results: Average expenditure, large expenditures and ▪ The study uses the difference in differences large borrowings on inpatient care had increased in MH methodology and its findings may have been and AP, but the increase was smaller in AP across these affected by unobservable differential changes three measures. DIDs for average expenditure and large between the two states. borrowings were significant and in favour of AP for the rural and the poorest households. Hospitalisation rates also increased in both states but more so in AP, although INTRODUCTION the DID was not significant and the subgroup analysis In 2005, member states of the WHO commit- presented a mixed picture. ted themselves to developing their health Conclusions: Health innovations in AP had a greater financing systems to deliver universal cover- beneficial effect on inpatient care-related expenditures age (UC), so that all people have access to than innovations in MH. The Aarogyasri scheme is likely health services and do not suffer financial to have contributed to these impacts in AP, at least in hardship when paying for them.1 WHO’s For numbered affiliations see part. However, OOPE increased in both states over time. end of article. 2010 World Health Report2 recommended Schemes such as the Aarogyasri and RSBY may result in inter alia that countries reduce reliance on some positive outcomes, but additional interventions Correspondence to may be required to improve access to care for the most direct payments, and improve equity of Dr Mala Rao; vulnerable sections of the population. access, including through the introduction of m.rao@uel.ac.uk prepayment schemes. However, one recent Rao M, Katyal A, Singh PV, et al. BMJ Open 2014;4:e004471. doi:10.1136/bmjopen-2013-004471 1 Downloaded from http://bmjopen.bmj.com/ on November 30, 2015 - Published by group.bmj.com Open Access systematic review of the impact of national health insur- inform India’s road map towards universal health cover- ance schemes in low-income and middle-income coun- age. Both schemes belong to the new generation of pub- tries3 found only weak evidence of increased use of licly funded government-sponsored health insurance healthcare and reduced out-of-pocket expenses, with the schemes, principally aimed at providing financial protec- poorer benefiting less, while others4 5 concluded that tion to the poor against catastrophic health shocks, health insurance improved healthcare access and use, as which, for these schemes, the Government has defined well as financial protection in most cases, but had no as inpatient hospital care.9 Both schemes have been conclusive impact on health status. Both highlighted the subject to some assessment in their early phases, but a need for more rigorous assessments of such schemes. recent editorial13 highlighted the necessity of further India has one of the highest levels of out-of-pocket and repeated evaluations to enhance the credibility and health expenditure (more than 80% of private health accountability of existing schemes and to identify those expenditure).6 The aim of our study was to explore how which deserve a scale-up. The Aarogyasri and RSBY recently introduced health financing initiatives have schemes and the other recent health sector innovations affected access to and out-of-pocket expenditure in AP and MH, which form the major backdrop and (OOPE) on inpatient hospital care in the Indian states complex healthcare architecture against which these of Andhra Pradesh (AP) and Maharashtra (MH). have been launched, and which may have contributed to the changes in outcomes we have explored in our study, are described below. BACKGROUND In its Eleventh Five Year Plan (2007–2012),7 the Government of India (GOI) sought to increase public The Rajiv Aarogyasri Health Insurance Scheme of AP expenditure on health and to strengthen investment in In 2007, AP launched a pioneering new state-wide, fully the rural health infrastructure through the National state-funded health insurance scheme, the Rajiv Rural Health Mission. Its Twelfth Five Year Plan (2012– Aarogyasri Community Health Insurance Scheme 2017)8 reflected the recommendation of the Planning (Aarogyasri),11 to provide treatment for serious and life- Commission’s High Level Expert Group for general tax- threatening illnesses. The specific objectives include: to ation to be the principal source of healthcare financing. improve access of poor families to quality medical care It proposed the development of government-funded (meaning low-frequency, high-cost specialist care) and health insurance schemes, building on the evidence treatment of identified diseases requiring hospitalisation from experimental schemes being introduced across through an identified network of healthcare providers, many states. In India, health is primarily a state rather to provide financial cover for catastrophic illnesses than national responsibility.6 While it is recognised that which have the potential to wipe out the lifetime savings political will and good governance are essential, both of poor families and to provide UC to the urban and the political mobilisation of funds for health schemes the rural poor in the state,14 albeit for the conditions and the effectiveness and efficiency of funded schemes covered in the benefits package. All families with a will be enhanced by robust evidence which documents ‘below poverty line’ (BPL) ration card, that is, those on as to whether schemes achieve objectives, what works an annual income below US$1384 ( 75 000) in urban well and the main challenges they face. areas and US$1107 ( 60 000) in rural areas, and includ- In India, evaluation is not routine, even of large, costly ing individuals with pre-existing medical conditions are public healthcare programmes. Nevertheless, some automatically enrolled and the scheme was estimated to assessments have been carried out, for example, of the cover approximately 20.4 million poor and lower middle Yeshasvini Co-operative Farmers Health Care Scheme of class families, comprising about 85% of the state’s popu- Karnataka, the longest running state-supported health lation in 2009.9 Enrollees make no contribution, the insurance scheme for the informal sector in India,9 with annual benefit is a maximum of US$4500 ( 200 000) promising results in terms of increased utilisation of and per family per year and there is no limit on the size of reduced borrowing for healthcare services.10 However, the family.14 A total of 942 medical and surgical proce- these early models covered small populations and dures across 31 clinical specialties14 are provided and offered limited benefits, so that the policy implications the benefits include all inpatient costs—associated inves- of conclusions drawn from even the best evaluations tigations, food, transport and medicines for 10 days fol- were unclear. lowing discharge. One year follow-up packages including This scenario has changed in the past 5 years with the consultation, medicines and diagnostics are also avail- launch of two schemes, the Rajiv Aarogyasri11 able for 125 procedures requiring longer periods of Community Health Insurance Scheme (Aarogyasri) of follow-up.9 Aarogyasri has unique features including AP and Rashtriya Swasthya Bima Yojana (RSBY) cur- Aarogyamithras (health system navigators), outreach rently offered in 30 states and union territories of health camps delivered by participating hospitals to India,12 including AP’s neighbouring state of MH. Their educate, screen and case-find and a state-of-the-art infor- scale in terms of population coverage and range of treat- mation technology-based management system. At the ments offered significantly enhances their potential to time of this study, 353 public and private sector hospitals 2 Rao M, Katyal A, Singh PV, et al. BMJ Open 2014;4:e004471. doi:10.1136/bmjopen-2013-004471 Downloaded from http://bmjopen.bmj.com/ on November 30, 2015 - Published by group.bmj.com Open Access were ‘empanelled’ to provide services to Aarogyasri conducted in the Amravati district of MH, which has a beneficiaries. large tribal population.19 In MH,17 utilisation rates have In 2009, a descriptive study of Aarogyasri, based on an been reported to be lower than in other states and the analysis of claims data and a survey of beneficiaries,15 male:female enrolment ratio is 6.5:3.5. concluded that while the scheme was beginning to reach its intended beneficiaries, uptake was lower among Other major health sector initiatives in AP and MH scheduled castes and tribes. This was confirmed by Fan Both states have a complex healthcare landscape with et al16 who used variations in programme roll-out over numerous programmes in place. There are several initia- time and districts to evaluate the scheme using National tives which have been launched during the past decade, Sample Survey data collected before and after its some of which are common to both states and driven by launch. They reported reduced OOPE in this initial national strategies, and others which owe their existence phase but no major impact on catastrophic healthcare to state-level enterprise, innovation and political expenditure. Inspired by Aarogyasri and mindful of the support. The most notable programmes with the poten- political benefits of introducing popular health reforms, tial to impact on patient care are described below. other states have launched health financing innovations The National Rural Health Mission (NRHM) was similar to this model. launched in 2005 nationwide, with the key aim of redu- cing maternal and infant mortality.20 Government RSBY in MH reports suggest that its notable achievements include an RSBY was launched across a number of states by the increase in institutional deliveries; in AP from 1.25 Ministry of Labour, GOI in 200817 and provides access to million in 2005–2006 to 1.46 million by 2011–201221 and free inpatient hospital care up to US$550 ( 30 000) per in MH from 1.1 million to 1.63 million,22 achieving an family per year.18 Households which meet the criteria institutional delivery rate of approximately 92% in both based on the much more limiting definition of poverty states. Also common to both states is the ‘104 health and numbers of poor families provided for each State by information help line’ launched in AP in 2008 and in the GOI Planning Commission are eligible to enrol, and MH in 201123 to provide medical advice and informa- pay a contribution of US$0.55 ( 30) at registration and tion based on validated algorithms and disease summar- at each annual renewal.9 Up to five family members, ies, to direct callers to appropriate health facilities or to including those with pre-existing conditions, can be receive complaints against a public sector health facility. covered, and personal information including biometric In AP, the help line and call centre were subsumed data are collected prior to the issue of a smart card with within the Aarogyasri infrastructure by 2011. encoded details of the family. Seven hundred proce- In MH, the RSBY was preceded by the Jeevandayee dures covering 18 broad categories of interventions, scheme launched in 1997 with the objective of reducing which would generally be included under the umbrella catastrophic OOPE on inpatient care in the BPL popula- of ‘secondary’ care, are provided and the benefit tion.24 Potential beneficiaries were required to apply for packages include the intervention, public transport costs funding after a diagnosis was confirmed and the scheme limited to US$1.8 ( 100) per visit and US$18.2 ( 1000) covered serious illness such as cardiac and renal disease per year and posthospitalisation drugs for 5 days. and cancer. However, the scheme uptake has been low, and Networked hospitals are required to provide free out- while it has continued to run in parallel to the RSBY, only patient consultations (which have only recently been 66 853 procedures (4456 procedures per year in a state introduced; Kurian OC, personal communication), but with 112.37 million people) have been approved during other costs such as ambulatory diagnostics and medi- the scheme’s lifetime.25 Since 2006, MH has also had a cines have to be borne by the beneficiaries, except if scheme in place which mandated 20% of the beds in investigations lead to inpatient admissions within a day.9 private hospitals to be made available for free or at subsi- A pilot of the RSBY scheme was launched in one district dised rates to poor patients (Kurian OC). It has been esti- of AP, but only after the start of our household survey. mated that around 10 000 private beds are available for the In MH, enrolment began in August 2009, and by poor across MH, equivalent to approximately 20% of the mid-2013, approximately two million of the eligible four total bed capacity of the public sector. Although the imple- million families were enrolled in the scheme,12 which is mentation of the scheme is reported to be erratic (Kurian being implemented in 31 of 35 districts in the state. OC), it may have had some positive impact on access to Enrolment had extended to 26 districts prior to June hospital inpatient care for serious illness. 2012, when our household survey began. Notably, only Launched in 1995–1996, the Navasanjeevani Yojana 15 of 1215 hospitals contracted for RSBY funded services scheme is exclusive to the 15 tribal districts of MH and are from the public sector.12 Early assessments of was to improve maternal and infant mortality in these the scheme’s impact nationally suggest that although the vulnerable populations.26 It has focused on strengthen- rate of hospitalisations has increased, awareness of the ing primary health and nutrition services and access to scheme was poor and remains a barrier to uptake. safe drinking water. Notable variations in enrolment and scheme awareness A service available in AP but not in MH is the ‘108’ were also observed by a descriptive study of RSBY scheme, launched in 2005 to provide a state-of-the-art Rao M, Katyal A, Singh PV, et al. BMJ Open 2014;4:e004471. doi:10.1136/bmjopen-2013-004471 3 Downloaded from http://bmjopen.bmj.com/ on November 30, 2015 - Published by group.bmj.com Open Access medical emergency response service.27 At the time of AP and MH have a broadly similar development our study, 802 ambulances catered to approximately profile as shown by the data below. AP’s other socio- 3500 emergencies per day.28 economically similar neighbouring states of Karnataka and Tamil Nadu had already introduced Aarogyasri-like schemes, and Odisha and Chattisgarh, the only other OBJECTIVES OF THE STUDY neighbours, had comparatively higher levels of socio- Our objective was to compare the effects of health inno- economic deprivation. HREBs were measured in AP and vations over time on access to and OOPE on inpatient MH by two waves of household survey before (2004) and care in AP and MH and to assess whether the AP initia- after (2012) the introduction of Aarogyasri and RSBY. tives had larger or smaller beneficial effects than those found in MH. These differential effects are likely to be Survey design substantially due to the Aarogyasri scheme in AP and Baseline survey: 2004 the RSBY in MH. In this paper, we report findings from We used the original data from the NSSO 60th decennial a study which compared these changes. The findings do round household survey undertaken in 200432 to estimate not allow us to draw inferences on the impacts of indi- baseline HREB estimates for AP and MH (table 1). This vidual initiatives, but nevertheless contribute new knowl- was the most recent round measuring morbidity profiles, edge on the impact and role of the innovations, provide use of healthcare services including hospitalised and non- lessons for other programmes, and strengthen the evi- hospitalised treatments and expenditures incurred. The dence base for policy on UC in India. household survey used a multistage stratified sampling methodology to identify a representative random popula- METHODS tion sample and an interviewer completed questionnaire Overview to obtain measures of HREB along with sociodemographic None of the aforementioned initiatives—including the household expenditure and other information. Aarogyasri and RSBY schemes in which we are especially interested—was piloted in a systematic way, let alone via a Follow-up survey: 2012 carefully designed randomised control trial. Following We used the same household survey design and Medical Research Council (MRC) Guidance29 30 and best methods to collect postintervention data in AP and MH practice, we therefore opted for a cross-sectional survey as those used by NSSO. Briefly, the household survey design in which we seek to minimise selection bias, to used a multistage stratified sampling methodology with control for confounding variables and to reduce the effects the ‘First Stage Units’ (FSUs) identical to those used by of chance. Specifically, we compare changes in hospital NSSO in their 66th round (2008–2009),33 the latest inpatient care-related expenditures and behaviours round for which FSUs had been mapped. However, the (HREB) in AP and MH before and after the roll-out of the FSUs were not the same as those in NSSO 2004, our Aarogyasri scheme in AP and the RSBY scheme in MH. baseline survey, rapid urbanisation having changed sub- The difference in changes between AP and MH is not an stantially the urban–rural landscape of both states and estimate of a specific initiative. Rather, it tells us whether, thus the geographical basis for sampling units. on balance, the AP initiatives have had larger (or smaller) The interviewer-completed household survey question- beneficial effects than the MH initiatives, and if so how naire was pretested and then piloted in both states prior to much more (or less) beneficial they have been. Since the the survey. All respondents provided written informed NRHM was common to both states and the MH-specific consent for participation. Questions addressed the follow- initiatives were quite small in scale or unlikely to affect ing: household composition and sociodemographic HREBs, any difference in change between the two states is characteristics of members, household expenditure, quite likely to be mainly due to the differential effects of health expenditure (outpatient and inpatient) and means the Aarogyasri and RSBY programmes. of its financing, healthcare-seeking behaviour, factors MH Indicator AP 112.37 Population (2011 census, in millions) 84.66 10.20 % Scheduled Caste (2001 census)* 16.60 8.90 % Scheduled Tribe (2001 census)* 6.20 101 314 Per capita income 2011–2012 (in )† 71 540 35 Number of districts 23 5314 Households covered in National Sample Survey 5059 Organisation (NSSO) 60th round (2004–2005) 10 073 Households covered in the study 2012 8623 *Note that the 2011 census data for social groups are not yet published. †Source: Presentation on Annual Plan 2012–2013 and Five Year Plan 2012–2017.31 4 Rao M, Katyal A, Singh PV, et al. BMJ Open 2014;4:e004471. doi:10.1136/bmjopen-2013-004471 Downloaded from http://bmjopen.bmj.com/ on November 30, 2015 - Published by group.bmj.com Open Access Table 1 Urban and rural populations and households surveyed in 2004 and 2012 in Andhra Pradesh and Maharashtra Andhra Pradesh Maharashtra NSSO 60th round Our survey NSSO 60th round Our survey 2004–2005 2012 2004–2005 2012 Population 76 210 007* 84 665 533† 96 878 627* 112 372 972† Urban population 20 808 940* 28 353 745† 41 100 980* 50 827 531† Rural population 55 401 067* 56 311 788† 55 777 647* 61 545 441† Total households (urban) 4 397 138* 6 778 225† 8 403 224* 10 813 928† Total households (rural) 12 607 167* 14 246 309† 11 173 512* 13 016 652† Total households 17 004 305* 21 024 534† 19 576 736* 23 830 580† FSUs (urban) 183 372 267 504 FSU (rural) 325 491 265 504 Total households surveyed (urban) 1824 3715 2664 5038 Total households covered (rural) 3235 4908 2650 5035 *2001 Census. †2011 Census. The NSSO 66th round had 492 rural FSUs in Andhra Pradesh, but 1 FSU was found to be uninhabited. FSU, First Stratum Unit; NSSO, National Sample Survey Organisation of India. affecting access to healthcare and awareness and percep- any questionnaires reported incorrect were sent back to tions of the quality of the Aarogyasri scheme (AP only). the field for resurvey. The research team carried out a The survey questions in 2012 were identical to those from final validation and review of the data. the NSSO 2004.34 Additional questions specific to the Aarogyasri and other relevant schemes were also added. Outcome measures A survey of 18 696 households across 2 states and 1871 Average inpatient expenditure per household per year locations within the states is a challenging undertaking. Average OOPE for inpatient care during the 1 year prior The survey design had several features intended to assure to the survey was estimated from questionnaire the quality of data collected. Few academic institutions responses for AP and MH from the baseline and have the internal capacity to carry out such large surveys, follow-up data. Reimbursements for inpatient expend- and consequently the Social and Research Institute of iture were deducted from the total where households IMRB International, a leading market research agency, had received them. was selected to carry out the survey. The Institute has field survey teams based in every Indian state, conversant Large out-of-pocket inpatient expenditure in local languages and dialects and trained to carry out Owing to the limited data on household consumption in surveys in the socioeconomic development sector. Its the 2004 NSSO health survey, we did not estimate ‘cata- clients include the GOI (for whom the national Family strophic health expenditure’. Instead, we constructed a Health Survey data are collected), the World Bank and measure of ‘large’ OOPE. The Aarogyasri Health Care other UN organisations. A group of NSSO consultants in Trust data on expenditure incurred by the Government of AP and the Indian Socioeconomic Research Unit, Pune AP per case in 2012 were examined14 and the mean was were recruited to support the training of the field survey estimated as US$419 ( 23 000). A household was deemed teams and data verification. to have incurred ‘large’ expenditure if OOPE for inpatient We planned three levels of verification of the study care was equal to or greater than this threshold. data: the first to be undertaken by the survey agency, the second to be carried out by the study team and the third by the agencies mentioned above. Survey teams for Large borrowing each district were accountable to a field supervisor who We estimated the total amount borrowed by a household was responsible for checking the household listing and to meet the expenditure of all the inpatient episodes of data entry on a daily basis. The study team also accom- that family during the previous year. A household was panied the field staff to survey sites on a regular basis. considered to have incurred a ‘large borrowing’ if the Data collected from 250 households in each state borrowing was equal to or exceeded the BPL threshold (approximately 2.5% of the surveyed households) and set by the Government of AP: 70 000 for urban families 186 of the FSU listings (approximately 10%) were inde- and 65 000 for rural households. These prices have pendently verified by the agencies in the villages and been deflated to 2004 levels. urban blocks in order to ensure that the sampling method and administration of the questionnaire survey Hospitalisation rate were being correctly applied. The data entry was carried This was estimated as the number of individuals hospita- out by the Institute using a double entry method and lised during the previous year, per 1000 population. Rao M, Katyal A, Singh PV, et al. BMJ Open 2014;4:e004471. doi:10.1136/bmjopen-2013-004471 5 Downloaded from http://bmjopen.bmj.com/ on November 30, 2015 - Published by group.bmj.com Open Access Variations in outcomes were examined between male- household lives in a rural or urban location, three headed and female-headed households, and rural and dummy variables capturing the household’s social group urban populations, as well as across social groups and (the lowest is the excluded category) and four asset economic groups represented by asset quintiles. quintile dummies (the bottom is the excluded category). Owing to the limited data on household consumption In the regression, yit is the outcome, state is a dummy in the 2004 NSSO health survey which made the estima- variable with 0 for MH and 1 for AP, and survey is a tion of wealth difficult, we have opted instead to dummy variable with 0 for the 2004 survey and 1 for the measure household living standards using an asset or 2012 survey. The coefficient for the interaction term, b3 , wealth index based on information on ownership of gives the DID estimate, Y DD . Robust SEs of Y DD were cal- household durables, dwelling type, etc, using principal culated to account for clustering of households within component analysis to estimate weights35–37 for each FSUs using Stata survey commands. A positive value for indicator. The indicators used were limited to those col- Y DD suggested that the change in the outcome in AP lected in the 2004 NSSO: type of structure of the dwell- was more than the change in MH and that the negative ing unit, type of toilet, type of fuel used for cooking and value would suggest the reverse. source of drinking. Data from the 2004 and 2012 surveys An advantage of a regression based DID estimate is were pooled so that the index captures changes in living this ability to use covariates which can account for differ- standards between the 2 years; there are therefore more ential trajectories in the two states. In addition to this, households in the top quintile in 2012 than in 2004. we did subgroup analysis stratifying for different covari- The statistical software Stata V.11 was used to generate ates. This is particularly relevant in the case of sched- the index. uled tribes whose proportion increased in MH in the follow-up survey. Deflation of follow-up expenditure estimates Subgroups were not mutually adjusted for the analysis The 2012 expenditure data including the threshold for due to sample size restrictions in relation to some of large expenditures were deflated using the consumer them. price index of the GOI38 to reflect 2004 prices. Role of the funding sources Analysis The external funding sources had no role in study Our repeat cross-sectional surveys do not allow for esti- design, data collection, analysis, interpretation or report- mation of within-individual household changes in out- ing or in submission decision. comes over time. Our analysis therefore focused on estimating outcomes averaged across states, and in com- paring changes in these over time between AP and MH. RESULTS If we assume that outcome determinants other than A total of 5314 and 5059 households from MH and AP Aarogyasri and RSBY remained stable in the two states were surveyed by the NSSO in 2004 (table 2). Our over time or followed a parallel change, then a survey in 2012 included 10 073 (MH) and 8623 (AP) difference-in-differences (DID) analysis will uncover the households. net effect of Aarogyasri over and above RSBY. The DID of outcome ðY DD Þ is Changes in average in-patient expenditure Table 3 (top panel) shows average baseline levels of ðY AP 2012 À Y 2004 Þ À ðY 2012 À Y 2004 Þ AP MH MH inpatient expenditure. The table also shows the real terms change (deflated to 2004 prices) in these out- where the subscripts and superscripts for Y refer to the comes at follow-up and the DID estimate comparing AP respective states and the years when the surveys were with MH. DIDs for overall results are shown unadjusted, carried out. CIs were calculated from the SE Y DD and as well as adjusted for the effects of the covariates. the p value for the null-hypothesis ðY DD ¼ 0Þ was tested Breakdowns by sex of the head of the household, social using the Wald test ast ¼ Y DD =SEY DD with one degree of group, urban/rural location and asset quintiles are also freedom. Y DD was estimated using ordinary least square shown. regression: Overall, average inpatient expenditure increased in real terms in the states between 2004 and 2012, but the yit ¼ b0 þ b1 statei þ b2 surveyt þ b3 ðstate  surveyÞit increase was significantly greater in MH (unadjusted Xm DID=−498.2 , 95% CI −792.9 to −203.5, p=0.0009). þ b3þk covariatek þ 1 The direction in terms of a greater increase in MH was k ¼1 evident across all subgroups of analysis except the The basic DID results are obtained using the above richest asset quintile. However, the DIDs reached signifi- regression with covariates excluded. The adjusted DID cance in male-headed households (DID=−513.7 , 95% results are obtained using the above regression with m=9 CI −843.9 to −183.4, p=0.0023), scheduled castes (DID= covariates, namely the gender of the head of the house- −708.7 , 95% CI −1234.3 to −183.2, p=0.0082), all hold, a dummy variable capturing whether the ‘other’ social groups (DID=−1110.46 , 95% CI −1868 6 Rao M, Katyal A, Singh PV, et al. BMJ Open 2014;4:e004471. doi:10.1136/bmjopen-2013-004471 Downloaded from http://bmjopen.bmj.com/ on November 30, 2015 - Published by group.bmj.com Open Access Table 2 Sociodemographic characteristics of baseline and follow-up samples Number (%) of households 2004 Number (%) of households 2012 Subgroups Maharashtra Andhra Pradesh Maharashtra Andhra Pradesh All 5314 5059 10 073 8623 Head of household Male 4785 (90.0) 4433 (87.6) 8543 (84.8) 7418 (86.0) Female 529 (10.0) 626 (12.4) 1530 (15.2) 1205 (14.0) Social group Scheduled Tribes 413 (7.8) 296 (5.9) 1364 (13.5) 883 (10.2) Scheduled Castes 809 (15.2) 974 (19.3) 2235 (22.2) 1797 (20.8) Other excluded 1644 (30.9) 2317 (45.8) 1899 (18.9) 3419 (39.7) All other groups 2448 (46.1) 1472 (29.1) 4571 (45.4) 2524 (29.3) Location Rural 2650 (49.9) 3235 (63.9) 5035 (50.0) 4908 (57.0) Urban 2664 (50.1) 1824 (36.1) 5038 (50.0) 3715 (43.0) Asset quintile Lowest 1260 (23.7) 1594 (31.5) 996 (9.9) 826 (9.6) Second 1016 (19.1) 1237 (24.5) 1841 (18.2) 1286 (14.9) Third 772 (14.5) 753 (14.9) 2228 (22.1) 2121 (24.60) Fourth 857 (16.1) 744 (14.7) 2373 (23.6) 3072 (35.6) Fifth 1408 (26.5) 730 (14.4) 2633 (26.1) 1318 (15.3) to −352.9, p=0.0041), rural households (DID=−504 , both male-headed and female-headed households, there 95% CI −801.9 to −206.0, p=0.0009) and the poorest was a greater increase in hospitalisation in AP, but this (DID=−1001.3 , 95% CI −1751 to −251.7, p=0.0089) reached moderate statistical significance only for female- and middle asset quintiles (DID=−798.1 , 95% CI headed households (DID=27.6, 95% CI 1.1 to 54.1, −1362.9 to −233.3, p=0.0056). p=0.0415). There is an increase in hospitalisations among scheduled tribes in MH and a reduction in AP (DID= Large expenditures for inpatient care −19.8, 95% CI −37.3 to −2.3, p=0.0272), but the opposite Proportions of households incurring large expenditures picture was seen among ‘other excluded’ groups with an showed an increase in both states (table 4), but the increase in AP and a reduction in MH (DID=12.5, 95% increase was smaller in AP for the sample as a whole as CI 1.2 to 23.9 p=0.0309). In scheduled castes, hospitalisa- well as for all the groups except for the second asset tions had increased in both states but more so in AP, quintile. The DID was strongly significant for the house- while in the ‘other’ group there was a small increase in holds overall (adjusted DID=−1.8, 95% CI −3 to −0.7, MH and a small reduction in AP. In the poorest quintile, p=0.0009), but this was not observed for any of the sub- the increase in hospitalisation was significantly greater in groups of analysis. MH (DID=−14.4, 95% CI −28 to −0.31, p=0.0451). Changes in large borrowing for inpatient care In both states, proportions of households incurring LIMITATIONS OF THE STUDY large borrowings to meet inpatient expenses increased DID estimations aimed at assessing the impacts of inter- from 2004 to 2012 (table 5). However, there was a con- ventions assume that both populations demonstrate sistent pattern of smaller increases in AP for the overall similar characteristics prior to the start of the interven- population, as well as all subgroups (except the richest tion, and that ‘unobservables’ follow a common trend; asset quintile) with DIDs strongly significant for the under such circumstances, any differences in changes overall population (adjusted DID=−4, 95% CI −6.6 to observed over time between the two populations are −1.4, p=0.0032), scheduled tribes (DID=−5.5, 95% CIs attributable to the interventions.39 40 Despite AP and −9.3 to −1, p=0.0048), rural households (DID=−4.7, MH having broadly similar socioeconomic profiles, as 95% CIs −7.3 to −2.1, p=0.0007) and all asset quintiles well as our DID analysis taking account of a number of except the richest (the poorest asset quintile DID=−9.0, covariates, there may have been other factors resulting 95% CI −14.0 to −4.4, p=0.0002). in unobserved differential changes between the two populations to which the results of the DID analysis may Hospital utilisation for inpatient care be at least partially attributable. Overall, hospitalisation rates have increased in AP and A second limitation could arise from the impact of MH (table 6), but more so in AP (5.6/1000 population vs other public health programmes implemented during 2.2), although the DID was not statistically significant. the period 2004–2012. The most significant of these is The subgroup analysis presented a mixed picture. For the National Rural Health Mission launched in 2005, Rao M, Katyal A, Singh PV, et al. BMJ Open 2014;4:e004471. doi:10.1136/bmjopen-2013-004471 7 8 Open Access Downloaded from http://bmjopen.bmj.com/ on November 30, 2015 - Published by group.bmj.com Table 3 Change in average inpatient expenditure (in ) in Maharashtra and Andhra Pradesh between 2004 and 2012 Baseline mean (95% CI) Change 2004:2012 mean (95% CI) DID Subgroups Maharashtra Andhra Pradesh Maharashtra Andhra Pradesh Mean (95% CI) p Value Household inpatient expenditure All 1091.6 (978.4 to 1204.8) 723.5 (527.5 to 919.5) 942.8 (749.9 to 1135.6) 444.55 (221.5 to 667.6) −498.2 (−792.9 to -203.5) 0.0009 Adjusted for covariates −565.8 (862.9 to −268.6) 0.0002 Head of household Rao M, Katyal A, Singh PV, et al. BMJ Open 2014;4:e004471. doi:10.1136/bmjopen-2013-004471 Male 1132.9 (1015 to 1251) 758 (419.7 to 1096.3) 935 (727 to 1143.01) 1074.9 (555.9 to 1593.8) −513.7 (−843.9 to −183.4) 0.0023 Female 757.1 (555.5 to 1014.6) 341.1 (222.3 to 460.01) 421.3 (164.7 to 678.0) 589.9 (307.22 to 872.8) −484.9 (−1075.6 to 105.9) 0.1076 Social group Scheduled Tribes 376.6 (231.7 to 521.6) 432.7 (212.52 to 652.9) 1153.1 (803.3 to 1502.9) 675.2 (163.2 to 1187.2) −477.9 (−1097.7 to 142) 0.1307 Scheduled 696.7 (500.2 to 893.2) 432.6 (305.6 to 559.4) 1464.1 (1039.9 to 1888.4) 755.4 (444.9 to 1065.9) −708.7 (−1234.3 to −183.2) 0.0082 Castes Other excluded 1028.6 (838.4 to 1218.8 562.4 (463.7 to 662) 928.9 (532.9 to 1324.9) 767.9 (569.6 to 966.2) −161 (−603.7 to 281.7) 0.4758 All other groups 1424.5 (1222.3 to 1626.7) 1306.2 (627.8 to 1984.6) 734.9 (427.9 to 1041.7) −375.6 (−1068.5 to 317.4) −1110.46 (−1868 to −352.9) 0.0041 Location Rural 897.8 (768.1 to 1027.5) 571.4 (496.2 to 646.6) 1084.7(826.3 to 1343.1) 580.7 (432.2 to 729.2) −504 (−801.9 to −206.0) 0.0009 Urban 1343.5 (1146.1 to 1540.9) 1113.5 (466.2 to 1760.8) 753.6 (458.7 to 1048.6) 92.3 (−586.92 to 771.5) −661.3 (−1401.5 to 78.864) 0.0799 Quintile Poorest 656.3 (498.0 to 814.6) 391.5 (319 to 464.1) 1692.5 (1053.3 to 2331.7) 691.2 (298.9 to 1083.5) −1001.3 (−1751 to −251.7) 0.0089 2nd 786.5 (583.5 to 989.5) 443.3 (356.5 to 530.2) 979.3 (599.4 to 1359.2) 839.5 (465.7 to 1213.3) −139.8 (−672.5 to 393) 0.607 Middle 1062.7 (738.8 to 1386.1) 862.1 (577.8 to 1146.5) 1011.8 (550.2 to 1473.4) 213.7 (−112.1 to 539.6) −798.1 (−1362.9 to −233.3) 0.0056 4th 1241.7 (894.4 to 1589.1) 1819 (337.5 to 3302.5) 803.6 (328.7 to 1278.5) −644.3 (−2128.3 to 839.7) −1447.9 (−3005.2 to 109.5) 0.0684 Richest 1818.6 (1505.5 to 2131.8) 908.3 (682.1 to 1133.4) 252.3 (−193.4 to 698.1) 362.1 (15.3 to 708.9) 109.7 (−454.80 to 674.3) 0.7031 DID, difference in differences. Downloaded from http://bmjopen.bmj.com/ on November 30, 2015 - Published by group.bmj.com Open Access Table 4 Change in the proportion (%) of households incurring large health expenditures for inpatient care in Maharashtra and Andhra Pradesh between 2004 and 2012 Baseline mean (95% CI) Change 2004:2012 mean (95% CI) DID Andhra Subgroups Maharashtra Pradesh Maharashtra Andhra Pradesh Mean (95% CI) p Value Large inpatient expenditure ( 23 000 deflated to 2004 figures) All 6.7 (6 to 7.3) 3.4 (2.9 to 3.9) 3.1 (2.1 to 4.1) 2.2 (1.5 to 2.8) −0.91 (−2.1 to 0.27) 0.1302 Adjusted for covariates −1.8 (−3 to −0.7) 0.0009 Head of household Male 6.8 (6.2 to 7.6) 3.5 (3.1 to 4) 3.1 (2.0 to 4.1) 2.1 (1.5 to 2.9) −0.8 (−2.1 to −0.4) 0.1928 Female 5.0 (3.7 to 6.4) 2.8 (1.9 to 3.7) 3.9 (2 to 5.8) 2 (6.6 to 3.4) −1.8 (−4.2 to 0.50) 0.1222 Social group Scheduled 2.2 (1.3 to 3.1) 1.4 (0.5 to 2.2) 5.3 (3.5 to 7) 3.5 (1.9 to 5.1) −1.7 (−4.1 to 0.61) 0.1478 Tribes Scheduled 6.1 (4.6 to 7.7) 2.1 (1.5 to 2.6) 4.3 (2.3 to 6.3) 3 (1.9 to 4.1) −1.2 (−3.5 to 1.01) 0.2785 Castes Other excluded 5.9 (4.7 to 7.0) 3.3 (2.7 to 3.8) 2.9 (1.1 to 4.6) 2.7 (1.7 to 3.6) −2.1 (−2.2 to 1.8) 0.8389 All other groups 8.3 (7.5 to 7.9) 5.4 (4.4 to 6.3) 2.2 (0.9 to 3.6) 0.51 (−0.7 to 1.7) −1.7 (−3.5 to 0.04) 0.0628 Location Rural 1.9 (1.5 to 2.2) 0.9 (0.79 to 1.1) 1.7 (1.1 to 2.3) 1.3 (0.09 to 1.6) −0.45 (−1.1 to 0.25) 0.2098 Urban 12.9 (11.9 to 14) 9.7 (8.9 to 10.7) 4.4 (3.0 to 5.7) 3.9 (2.6 to 5.3) −0.7 (−2.4 to 1.5) 0.6350 Quintile Poorest 1.8 (1.3 to 2.4) 1.1 (0.8 to 1.4) 3.7 (2.2 to 5.2) 1.7 (0.7 to 2.7) −0.2 (−3.8 to −0.19) 0.0307 2nd 2.7 (1.9 to 3.5) 1.2 (0.9 to 1.5) 2.1 (0.93 to 3.3) 2.2 (1.4 to 3.1) 0.9 (−1.4 to 1.6) 0.9079 Middle 6.9 (4.9 to 8.9) 4.2 (3.1 to 5.3) 1.3 (−1 to 3.6) 0.9 (−1.2 to 1.4) −1.2 (−3.9 to 1.4) 0.3596 4th 10.7 (8.7 to 12.6) 7.6 (5.9 to 9.2) 1.8 (−0.57 to 4.3) −0.036 (−1.9 to 1.80) −1.9 (−4.9 to 1.2) 0.2268 Richest 13.5 (12 to 14.9) 9.6 (8.1 to 11.2) 0.3 (−1.6 to 2.2) −0.6 (−2.8 to 1.6) −0.9 (−3.7 to 2) 0.5601 DID, difference in differences. mainly to improve maternal and child health through DISCUSSION the revitalisation of rural primary care and child and We found that average expenditure, large expenditures maternal health services. A key assumption of our study and large borrowings on inpatient care had increased in is that the impacts of the NRHM in terms of healthcare MH and AP, but the increase was consistently smaller in expenditure for maternal and child healthcare would AP across these three outcome measures, which may be have been similar in both states, as this was a nationwide suggestive of Aarogyasri having a somewhat larger effect development. Despite the improvements in the public than RSBY. Similar increases in institutional deliveries sector maternal and child health services sought by the across the two states, as well as low levels of utilisation of NRHM, it is widely recognised that the public, including RSBY and Jeevandayee schemes in MH, may further BPL families, continues to pay OOP for private health- strengthen this explanation. care. We have assumed that this behaviour is likely to be The increase in average OOPE on inpatient care in similar across the two states. Other health initiatives, AP and MH reflects a pattern observed nationwide.16 41 such as the Navsanjeevani Yojana and the 104 helpline, The Aarogyasri scheme may have contributed to the were unlikely to have had an impact on inpatient care or more favourable trajectory in AP directly and indirectly, expenditure. The 108 scheme had the potential, in AP, in that the scheme may have contributed to a reduction to influence hospitalisation rates by helping more house- in the prices of interventions and an increase in compe- holds to visit hospitals when seriously ill. However, the tition among healthcare providers. The evaluation of effect is likely to be small as the majority of even serious the Yeshasvini scheme also found a significant reduction illnesses do not result in a 108 call for transport. The in the price of surgical interventions.10 Our findings implementation of the RSBY and the scheme to make may suggest that the positive effects of Aarogyasri private hospital beds available for the poor in MH may detected by other studies15 16 at an early stage of the have diluted the DID, although both schemes are known roll-out of the scheme have been sustained. Automatic to have been only partially implemented. enrolment on the scheme, near universality of coverage Lastly, the 2004 NSSO survey, which served as our and no requirement for enrollee contributions may have baseline, was carried out between January and June contributed to the significant DIDs in male-headed 2004. Our end-line 2012 survey was carried out over a households, scheduled castes, rural households and the period of 3 months from June to September. The mor- poorest and middle asset quintiles. However, these bene- bidity and mortality patterns recorded in different time fits were not demonstrated in some of the most vulner- periods may vary, and could have influenced the data. able groups—female-headed households and scheduled Rao M, Katyal A, Singh PV, et al. BMJ Open 2014;4:e004471. doi:10.1136/bmjopen-2013-004471 9 10 Open Access Downloaded from http://bmjopen.bmj.com/ on November 30, 2015 - Published by group.bmj.com Table 5 Change in the proportion (%) of households’ large borrowings for inpatient care in Maharashtra and Andhra Pradesh between 2004 and 2012 Baseline mean (95% CI) Change 2004:2012 mean (95% CI) DID Subgroups Maharashtra Andhra Pradesh Maharashtra Andhra Pradesh Mean (95% CI) p Value Proportion of households having large borrowings All 7.5 (6.7 to 8.2) 3.8 (3 to 4.5) 8.9 (6.8 to 11) 5.3 (3.4 to 7.2) −3.7 (−6.4 to −0.908) 0.0100 Adjusted for covariates −4 (−6.6 to −1.4) 0.0032 Rao M, Katyal A, Singh PV, et al. BMJ Open 2014;4:e004471. doi:10.1136/bmjopen-2013-004471 Head of household Male 7.8 (7.0 to 8.5) 3.9 (3.1 to 4.7) 9.8 (6.6 to 1.3) 5.3 (3.3 to 7.3) −3.6 (−6.6 to −0.62) 0.0187 Female 5 (3 to 7) 2.9 (1.7 to 4.1) 8.(6.6 to 11.2) 5 (2.9 to 7.23) −4.7 (−8.3 to −1) 0.0137 Social group Scheduled Tribes 3.6 (2.3 to 4.9) 2.4 (0.93 to 3.9) 11 (8.9 to 14) 5.8 (2.8 to 8.8) −5.5 (−9.3 to −1.8) 0.0048 Scheduled Castes 7.2 (5.5 to 8.8) 3 (1.7 to 4.2) 9.6 (6.6 to 13) 5.8 (3.4 to 8.3) −3.8 ( −7.5 to 0.03) 0.0518 Other excluded 8.0 (6.8 to 9.2) 3.5 (2.7 to 4.4) 8 (5.8 to 10.3) 5.3 (3.2 to 7.4) −2.8 (−5.7 to 0.19) 0.0661 All other groups 8 (7.15 to 8.8) 0.052 (.040 to 0.064) 8.8 (5.9 to 12) 4.7 (2.2 to 7.4) −4.1 (−7.9 to −0.4.0) 0.0302 Location Rural 6.5 (5.6 to 7.5) 0.03 (0.024 to 0038) 10 (8.5 to 12) 5.8 (3.9 to 7.6) −4.7 (−7.3 to −2.1) 0.0007 Urban 8.7 (7.3 to 10) 0.056 (0.048 to 0.064) 7.0 (4.5 to 9.5) 4 (1.1 to 6.9) −3.0 (−6.7 to 0.68) 0.1081 Quintile Poorest 5.2 (3.9 to 6.5) 0.025 (0.016 to 0.033) 12.1 (7.8 to 16) 3.1 (1.3 to 0.049) −9 (−14 to −4.4) 0.0002 2nd 0.064 (0.048 to 0.08) 0.027 (0.021 to 0.032) 0.095 (0.070 to 0.12) 0.052 (0.034 to 0.070) −0.043 (−0.073 to −0.013) 0.0062 Middle 0.074 (0.050 to 0.098) 0.048(0.031 to 0.065) 0.10 (.073 to 0.133) 0.044 (0.013 to 0.076) −0.059 (−0.100 to −0.017) 0.0069 4th 0 .087(0.063 to 0.110) 0.06 (.041 to 0.078) 0.083 (0.061 to 0.104) 0.039 (0.014 to 0.064) −0.044 (−0.075 to −0.012) 0.0076 Richest 0.10 (0.090 to 0.12) 0.064 (0.049 to 0.079) 0.045 (0.0035 to 0.086) 0.049 (−0.0068 to 0.105) 0.0045 ( −0.062 to 0.071) 0.8937 Rao M, Katyal A, Singh PV, et al. BMJ Open 2014;4:e004471. doi:10.1136/bmjopen-2013-004471 Downloaded from http://bmjopen.bmj.com/ on November 30, 2015 - Published by group.bmj.com Table 6 Changes in hospitalisation (per 1000 population) in Maharashtra and Andhra Pradesh between 2004 and 2012 Baseline mean (95% CI) Change 2004:2012 mean (95% CI) DID Subgroups Maharashtra Andhra Pradesh Maharashtra Andhra Pradesh Mean (95% CI) p Value Hospitalisations per 1000 population All 41.3 (37.3 to 45.2) 31.5 (27.8 to 35.3) 2.2 (−4.7 to 9.1) 5.6 (−1.1 to 12.3) 3.4 (−5.9 to 12.7) 0.4636 Adjusted 0.7 (−8.6 to 10.2) 0.8685 Head of household Male 41.0 (37.1 to 44.9) 31.7 (16.9 to 34.0) 1.9 (−5.1 to 8.8) 4.4 (−2.4 to 11.1) 2.5 (−6.9 to 11.9) 0.5966 Female 51.6 (30.6 to 72.5) 25.5 (27.7 to 35.7) 13.9 (−7.53 to 35.4) 41.5 (24.6 to 58.4) 27.6 (1.1 to 54.1) 0.0415 Social group Scheduled Tribes 23.7 (14.2 to 33.1615) 35.5 (21.9 to 49.1) 17.1 (5.8 to 28.5) −2.7 (−16.95 to 11.5) −19.8 (−37.3 to −2.3) 0.0272 Scheduled Castes 42.4 (36.3 to 48.4) 29.5 (22.6 to 36.5) 3.9 (−7.9 to 15.7) 7.6 (−2.6 to 17.7) 3.7 (−11.4 to 18.7) 0.6268 Other excluded 44.2 (38.2 to 50.2) 29.9 (25.5 to 34.3) −1.9 (−11.2 to 7.3109) 10.6 (3.4 to 17.8) 12.5 (1.2 to 23.9) 0.0309 All other groups 42.5 (37.1 to 47.9) 34.9 (29.0 to 40.9) 0.93 (−7.3 to 9.1) −1.1 (−9.4 to 7.4) −2.0 ( −13.5 to 9.4) 0.7235 Location Rural 36.6 (32.2 to 41) 28.9 (26.4 to 31.5) 9.5 (3.5 to 15.4) 8.8 (2.2 to 15.3) −0.69 (−9.3 to 7.9) 0.8725 Urban 48.2 (43.7 to 53) 38.2 (35.5 to 41.2) −7.9 (−14.5 to −1.3) −2.5 (−12.1 to 7.1) 5.4 (−5.8 to 16.6) 0.3358 Quintile Poorest 31.4 (25.9 to 36.9) 27.5 (22.8 to 32.1) 20.7 (9 to 32.8) 6.4 (−1.7 to 14.4) −14.4 (−28 to −0.31) 0.0451 2nd 36.5 (28.3 to 44.7) 27.1 (21.9 to 32.4) 8 (0.5 to 15.4) 8.9 (−0.56 to 18.4) 0.9 (−10 to 12.5) 0.8746 Middle 47.8 (33.6 to 62.1) 35.3 (29.0 to 41.6) −1.5 (−17.9 to 14.8) 4.1 (−0.56 to 13.7) 5.6 (−12.8 to 24.0) 0.5457 4th 46.9 (39.9 to 53.9) 41.7 (32.7 to 50.8) −4.0 (−12 to 3.4) −5.1 (−15.7 to 5.4) −0.75 (−13.3 to 11.9) 0.9056 Richest 51.0 (46 to 56.1) 36.5 (25.8 to 47.2) −13.1 (−18.9 to −7.1) 1.5 (−19.5 to 22.6) 15.0 (−5.8 to 3.6) 0.1665 Open Access 11 Downloaded from http://bmjopen.bmj.com/ on November 30, 2015 - Published by group.bmj.com Open Access tribes. This is consistent with the findings of other covers common hospital procedures are addressing a studies,5 16 which reported that the slightly less vulner- large hitherto unmet need for inpatient care. The 108 able may benefit more from such schemes, with non- scheme in AP may be an additional albeit smaller con- financial barriers undermining access to services for the tributor. It may also be explained by a supplier-induced most vulnerable socioeconomic groups. The case sum- demand. An assessment of health provider behaviour maries below illustrate the obstacles they face. A similar and governance is an important strand of the evaluation observation was reported by the evaluation of the of health financing schemes46 but was outside the remit Mexican Seguro Popular health scheme which showed of our study. We would, however, strongly recommend that a third of the treatment-cluster households who an impact evaluation focusing on healthcare supply to were automatically affiliated were unaware of this fact.42 complement our evidence. The most likely explanation of an Aarogyasri effect The greater increase in hospitalisation in AP, in may be strengthened by the changes in terms of large female-headed households and ‘other excluded groups’, borrowings which have also increased over time in both which are two of the most vulnerable population groups, states but less so in AP across all groups of analysis is encouraging. However, the reduction in hospitalisa- except the richest asset quintile. A multicountry analysis tions among scheduled tribes and the poorest asset of household catastrophic health expenditure high- quintile in AP are of concern and suggest that if the lighted that increasing the availability of health services poor are to secure the benefits appropriated by the near is critical to improving health in poor countries, but it poor or more often by the rich,47 the provision of more could also raise the proportion of households facing comprehensive health schemes is essential, which com- catastrophic expenditure unless financial risk protection bines the tertiary and secondary care focus of Aarogyasri policies are given a high priority.43 Borrowing of com- and the secondary care benefits of RSBY with attention paratively small amounts is less impactful and may be a paid to minimise barriers such as the widespread influ- result of improved access to financial markets and also ence of illiteracy and lack of awareness, which limit supports consumption smoothening. During recent access to even schemes such as Aarogyasri that are decades, AP has in particular witnessed a significant rise apparently highly inclusive and non-discriminatory, as in microfinance institutions and debt due to high levels well as distance to facilities. Our case summaries illus- of interest levied by the more recent entrants to this trate this. Furthermore, health financing reforms such market.44 45 Nevertheless, our results suggest that as the Jamkesmas48 in Indonesia and the Seguro increases in large borrowings associated with inpatient Popular in Mexico,42 both countries similar to India in healthcare were smaller in AP. Strongly significant DIDs terms of population and growing economies, include in scheduled tribes, rural households and the poorest outpatient and inpatient care, and curative as well as and second asset quintiles and moderately significant preventive services, suggesting that a more comprehen- DIDs in female-headed households, scheduled castes sive service is possible to implement and worthy of and other excluded social groups may also point to consideration. Aarogyasri beginning to offer greater access to health- In summary, health innovations in AP had a greater care when families are faced with serious illness. beneficial effect on hospital inpatient care-related In terms of large OOPE, a significant DID was found expenditures than innovations in MH. The Aarogyasri only for the total households, but the direction of all scheme is likely to have contributed to these impacts in DIDs except for the second asset quintile was the same AP, at least in part. However, in both states, OOPE as for average expenditure on inpatient care and large increased over time, in keeping with the picture borrowings, that is, in favour of AP. It is perfectly pos- reported nationwide.49 The most likely explanations are sible that the non-significant differences for the majority that the poor spend the largest proportion of OOPE on of subgroups are due at least in part to the RSBY, the drugs,50 an expenditure not adequately addressed by the availability of private sector hospital beds for the poor health financing schemes, including the RAS, which and the Jeevandayee schemes reducing large OOPE in provide only ‘follow-up’ medicines for limited periods MH, and that without these schemes the expenditure after discharge from hospital. Yet chronic disease such would have been greater. In AP, 70% of survey house- as diabetes may result in a one-off cardiovascular inter- holds reported that they were covered by the Aarogyasri vention funded by the RAS, as well as lifelong medica- scheme, but only 25% of these ‘covered’ households tion required to be paid for out-of-pocket. Other were aware that the benefit package was limited. It is possible explanations are medical inflation and rising also possible therefore that in AP some families seek costs of initial consultation and diagnostic investigation hospital care assuming that the Aarogyasri scheme pro- of symptoms, often in the rapidly growing private sector vides comprehensive cover, but are faced with large outpatient services, prior to hospitalisation under the expenditures when their treatments fall outside the cover of the RAS or the other health financing schemes. limits. Hospitalisations also increased, and while it is generally The mixed picture in relation to the increasing rate of assumed that in developing countries this is likely to hospitalisations in both states may suggest that the address genuine need, its impact on health status is not Aarogyasri and, in particular, the RSBY scheme which known. 12 Rao M, Katyal A, Singh PV, et al. BMJ Open 2014;4:e004471. doi:10.1136/bmjopen-2013-004471 Downloaded from http://bmjopen.bmj.com/ on November 30, 2015 - Published by group.bmj.com Open Access IMPLICATIONS FOR POLICY AND PRACTICE Furthermore, differences in data and methodology may Despite these uncertainties, Aarogyasri is perceived, with result in even very well-designed evaluations of the same some justification across India, as a successful scheme, programme producing contrasting findings.52 Our evalu- and is being rapidly replicated across the states. Since ation is the first to use a survey methodology—the best July 2012, MH too has joined the list of states offering possible in the hierarchy of evaluation methodologies, an Aarogyasri-like scheme, the Rajiv Gandhi Jeevandayee when a randomised control trial is not achievable—to Arogya Yojana.51 This study has highlighted that such evaluate the health financing reforms in two large states schemes may result in some positive outcomes. of India. Despite that, the study has limitations which we Although the study was not designed to elicit the specific have acknowledged. For example, this study has exam- features of the scheme to which any comparatively ined health-related expenditure and behaviours at only greater benefits may be attributable, and which deserve two points in time. To establish trends, it is suggested to be replicated in other states, evidence from systematic that more than two time points are needed. However, it reviews3 points to the need for schemes to be more needs to be recognised that schemes such as the ‘comprehensively’ designed to maximise their positive Aarogyasri themselves may not have longevity in their impact. Aarogyasri’s design in terms of its aims to original form and may change or be replaced in address financial and non-financial barriers—being fully response to changing needs and policy imperatives. King state funded, the systematic administrative implementa- et al42 acknowledged that their assessment of the tion of Aarogyasri across the state so that it is now almost Mexican Seguro Popular programme (at 10 months) was universal, automatic enrolment, allowing access to the undertaken at an early stage, but it was nevertheless scheme via a ration card which most poor families own, recognised as having provided important evidence of the wide spectrum of treatments offered, the large impacts, albeit early ones. Therefore, a realistic aim of number of healthcare providers empanelled—is perhaps an evaluation in this rapidly changing health delivery responsible as a ‘comprehensive’ package for the greater landscape in India would be to study change over time, impact. even if that is limited initially to two time points, as well However, the study also suggests that improved access as to continue to evaluate the system repeatedly, and use to healthcare and the reduction of the overall burden of the evidence to reshape health delivery to be responsive OOPE, especially in the most vulnerable sections of the to future socioeconomic and epidemiological trends. In population,52 are likely to require additional interven- addition, the evaluation was not designed to assess pro- tions that address gaps in the availability of care and vider behaviours. Many other pertinent questions, such provide patients appropriate pathways that support their as the impact of the schemes on the overall economy of journey from a strong and comprehensive primary care healthcare, cannot be answered by a single evaluation. service where they may be informed of their entitle- However, these are recognised problems which can only ments and investigated for their initial symptoms to be addressed through continuous assessments, as other appropriate hospitals for the treatment of serious illness. evaluations have shown. Our study has nevertheless pro- Besides, inpatient care is only consumed by a small pro- duced sufficient insights to enable policy leaders to portion of households in a year,53 and this is especially improve programme effectiveness and, importantly, to true of tertiary care, while many more will seek out- undertake further assessment. Our household survey patient services and referral to inpatient care, should data provide a valuable baseline for future monitoring this be required. Others have strongly recommended and analyses of trends in both states. the strengthening of the primary care base as an essen- This study needs to be followed up with further and tial means to universal health coverage, and this study repeated evaluations as AP’s and MH’s schemes evolve; confirms their view.41 The key implications for AP are to to assess the impacts of redesign and to help health explore how best the most advantageous features of policy leaders achieve their aspiration of universal access Aarogyasri can be extended to include secondary and to good quality healthcare. primary care, while those for MH may be to build on and unify its menu of currently available schemes to Three faces of the Aarogyasri scheme create an evidence-based comprehensive health delivery The beneficiary system. These conclusions may be applicable to other Patient A, a 65-year-old widow, is from a tribal back- states with similar health financing schemes. ground. She is an unskilled labourer, supporting a The design of health financing systems as well as their family of 5, with a monthly income of 4000 (US$74). evaluations is complex and challenging, as the mountain She was referred to a municipal hospital with chest pain of available evidence suggests.41 42 52 54 Even the and underwent heart surgery. The hospital, which is a ground-breaking Seguro Popular health insurance pro- part of the Aarogyasri network, provided free care and gramme of Mexico, which used a cluster-randomised her total out-of-pocket expenses amounted to 850 (US trial design with strong government support, demon- $16$) for initial transport. In addition to the surgery, strated some but not all anticipated outcomes, contrary she received free food, money for transport home and to expectations,42 and a key recommendation was that follow-up medicines. She is very satisfied with the service continued assessment of the programme was needed. she received. Rao M, Katyal A, Singh PV, et al. BMJ Open 2014;4:e004471. doi:10.1136/bmjopen-2013-004471 13 Downloaded from http://bmjopen.bmj.com/ on November 30, 2015 - Published by group.bmj.com Open Access The excluded progress and data collation and verification, commented on drafts of the Patient B, a mother, has a BPL card, but her daughter report and helped prepare the references. AK undertook the data collation, verification and analysis, assisted with the survey and questionnaire design does not, as she was abandoned by her husband who is a and survey implementation and prepared the tables for the report. AS led the government employee and entitled to free healthcare. literature review, assisted with the study and questionnaire design, survey The mother was unable to secure free care using her implementation and preparation and analysis of baseline data, and BPL card for her seriously ill daughter who paid commented on drafts of the report. MK helped with the data analysis. AW out-of-pocket at a private facility where she was offered a devised the methodology for the estimation of the programme impacts, advised during the data-collection and data-preparation stages, wrote and hysterectomy for the relief of her gynaecological symp- implemented the computer code for the model estimation, helped to oversee toms. Her daughter’s healthcare costs were met by the the production of the results, and contributed text to the report. GN provided family selling a number of household assets and her technical advice on accounting for the complex survey structure in the granddaughter discontinuing her education to take up analysis, developed a STATA equation, helped to compute an asset index, paid work. The patient’s daughter is severely depressed advised on the output tables, verified the analysis and commented on drafts of the report. AR helped develop a conceptual framework for the evaluation, and does not speak to anyone. advised on funding proposals, the study design, analytical methodology and presentation of results and contributed text to the report. MR wrote the first The uninformed draft of the paper and its redrafts in accordance with the comments of all Patient C, a 43 year-old, had severe stomach pain one other authors and reviewers. night. Although the family had a BPL card, her husband Funding The study was funded by the International Development Research and son, rushed her to a private hospital nearby, which Centre, Canada, the Wellcome Trust, the UK Department for International was not part of the Aarogyasri network. They were Development, and Rockefeller Foundation. The World Bank supported Adam Wagstaff’s contribution to the study. unaware of how to access the Aarogyasri scheme hospi- tals, which were further away from home, and the local Competing interests MR, AK, PS, AS and SB have support from the primary health services being inadequate, patient C had Rockefeller Foundation, Wellcome Trust, International Development Research Centre, Canada and Department for International Development, UK. not had her initial symptoms investigated. The treatment was funded through a loan from a private moneylender. Ethics approval The study protocol and questionnaire for the 2012 survey were reviewed and approved by the Research Ethics Committee of the On discharge, she has not attended follow-up, as the Administrative Staff College of India, Hyderabad, which hosted the study. family cannot afford transport or medicines. Household survey questionnaires include signed consent by the head of the household or another adult representative of the household. Author affiliations 1 Provenance and peer review Not commissioned; externally peer reviewed. Institute for Health and Human Development, University of East London, London, UK Data sharing statement The 2004 National Sample Survey Organisation of 2 Administrative Staff College of India, Hyderabad, Andhra Pradesh, India India household survey questionnaire and data are available to the public. The 3 ACCESS Health International, Hyderabad, Andhra Pradesh, India 2012 household survey questionnaire, and the full anonymised household 4 SughaVazhvu Healthcare, Thanjavur, Tamil Nadu, India survey dataset will be available with open access as soon as the data analyses 5 Indian School of Business, Hyderabad, Andhra Pradesh, India and submissions for publication are completed. 6 Development Research Group (DECRG), The World Bank, Washington, DC, Open Access This is an Open Access article distributed in accordance with USA 7 the terms of the Creative Commons Attribution (CC BY 3.0) license, which Institute for Health and Human Development, University of East London & permits others to distribute, remix, adapt and build upon this work, for ESRC International Centre for Life Course Studies in Society and Health, commercial use, provided the original work is properly cited. See: http:// University College London, London, UK creativecommons.org/licenses/by/3.0/ Acknowledgements The authors thank Bhimasankaram Pochiraju, Sundaresh Peri, C Ravi and Rahul Ahluwalia for their advice and guidance at various stages of the study design and data analysis, and to colleagues at the University of East London and Administrative Staff College of India, REFERENCES Hyderabad for their administrative support. The authors also thank 1. Resolution WHA58.33. Sustainable health financing, universal P Suryanarayana, CM Reddy, D Chakrapani and K Pandu Ranga Reddy and AY coverage and social health insurance. In: Fifty-eighth World Health Jadhav and his colleagues for their contribution to the training of the survey Assembly, Geneva, 16–25 May 2005. Geneva: World Health Organization, 2005. http://apps.who.int/gb/ebwha/pdf_files/WHA58/ teams, verification of the survey and data collection and acknowledge the WHA58_33-en.pdf (accessed 26 Mar 2013). IMRB International Social Research Institute team’s support in carrying out 2. World Health Organization. Health Systems Financing. Path to the survey. The authors are grateful to Somil Nagpal and Joseph Kutzin for Universal Coverage. Geneva: World Health Organization, 2010. commenting on the report and helping us to improve it, to Sujatha Rao for World Health Organization Library Information Services. her encouragement and valuable suggestions and ideas throughout the 3. Acharya A, Vellakkal S, Taylor F, et al. Impact of national health insurance for the poor and the informal sector in low and middle course of the study and to Jay Bagaria for her help during the early stage of income countries: a systematic review. London: EPPICentre, Social the development of the proposal. Finally, the authors thank PV Ramesh Science Research Unit, Institute of Education, University of London, without whose constant support, guidance and encouragement this study 2012. http://www.dfid.gov.uk/r4d/PDF/Outputs/SystematicReviews/ would not have been possible. Health-insurance-2012Acharya-report.pdf 4. Giedion U, Diaz BY. A review of the evidence. In: Escobar ML, Contributors MR conceived and designed the study, applied for funding, and Griffin CShaw RP, eds. The impact of health insurance in low-and was responsible for the supervision and management of all aspects of the middle-income countries. Washington, DC: Brookings Inst Press, study as well as the dissemination of its results. She is the guarantor. SB 2011. shared responsibility for the conception of the study, applications for funding, 5. Giedion U, Alfonso E, Diaz. The impact of universal coverage schemes in the developing world: a review of the existing evidence. study design and data collation and analysis, contributed to the questionnaire Universal Health Coverage (UNICO) studies series; no. 25. design and commented on drafts of the report. PS contributed to the Washington, DC: The World Bank, 2013. conception of the study and study design, led the questionnaire design and 6. National Commission on Macroeconomics and Health. Report of the survey implementation, including training of survey staff, monitoring survey National Commission on Macroeconomics and Health. New Delhi: 14 Rao M, Katyal A, Singh PV, et al. BMJ Open 2014;4:e004471. doi:10.1136/bmjopen-2013-004471 Downloaded from http://bmjopen.bmj.com/ on November 30, 2015 - Published by group.bmj.com Open Access Ministry of Health and Family Welfare of India, 2005. Ministry of 32. National Sample Survey Organization. Morbidity, healthcare and Health and Family Welfare. condition of the aged. New Delhi: Ministry of Statistics and 7. Eleventh Five Year Plan (2007–2012). Vol 2: Social Sector. Programme Implementation, 2004. Government of India Planning Commission. 2008. http:// 33. National Sample Survey Organization. Household consumer planningcommission.nic.in/plans/planrel/fiveyr/11th/11_v2/11th_vol2. expenditure 66th round. New Delhi: Ministry of Statistics and pdf (accessed 26 Mar 2013). Programme Implementation, 2009–10. 8. Faster, Sustainable and More Inclusive Growth. An Approach to the 34. National Sample Survey Organization. Household consumer Twelfth Five Year Plan (2012-17). Government of India Planning expenditure in India. New Delhi: Ministry of Statistics and Commission. 2011. http://planningcommission.nic.in/plans/planrel/ Programme Implementation, 2005. 12appdrft/appraoch_12plan.pdf (accessed 26 Mar 2013). 35. Booysen F, Van der Berg S, Burger R. Using an Asset Index to 9. La Forgia G, Nagpal S. Government-Sponsored Health Insurance in Assess Trends in Poverty in Seven Sub-Saharan African Countries. India. Are You Covered? Washington, DC: World Bank, 2012. doi:10. World Dev 2008;36:1113–30. 1596/978-0-8213-9618-6 36. Vyas S, Kumarnayake L. Constructing socio-economic status 10. Agarwal A. ‘Impact evaluation of India’s ‘Yeshasvini’ community indices: how to use principal components analysis. Health Policy based health insurance programme. Health Econ 2010;19:5–35. Plan 2006;21:459–68. 11. The Rajiv Aarogyasri scheme. https://www.aarogyasri.org/ASRI/ 37. O’Donnell O, Van Doorslaer E, Wagstaff A, et al. Analyzing health index.jsp (accessed 26 Mar 2013). equity using household survey data: a guide to techniques and their 12. RSBY scheme status. http://www.rsby.gov.in/statewise.aspx? implementation, chap 6. Washington, DC: World Bank Institute, state=35 (accessed 30 Apr 2013). 2008. Measurement of Living Standards. http://siteresources. 13. Mahal A. Learning and getting better: rigorous evaluation of health worldbank.org/INTPAH/Resources/Publications/459843- policy in India. National Medical Journal of India, 2011:325–27. 1195594469249/HealthEquityFINAL.pdf (accessed 9 May 2013). 14. Aarogyasri Health Care Trust Annual Report 2011-2012. https:// 38. Ministry of Finance. Economic Survey 2011-12. New Delhi: www.aarogyasri.org/ASRI/EXT_IMAGES/documents/Annual_ Government of India, 2012. http://indiabudget.nic.in/ Report_201011.pdf (accessed 26 Mar 2013). 39. Poverty Reduction and Social Development Department. A User’s 15. Rao M, Ramachandra S, Bandyopadhyay S, et al. Addressing Guide to Poverty and Social Impact Analysis. Washington: World healthcare needs of people living below the poverty line: a rapid Bank, 2003. assessment of the Andhra Pradesh Health Insurance Scheme. Natl 40. Gertler P, Martinez S, Premand P, et al. Impact evaluation in Med J India 2011;24:335–41. practice. Washington: World Bank, 2011. 16. Fan V, Karan A, Mahal A. ‘State Health Insurance and Out of-Pocket 41. Selvaraj S, Karan A. Why Publicly-Financed Health Insurance Health Expenditures in Andhra Pradesh, India.’ CGD Working Paper Schemes are ineffective in providing financial risk protection. Econ 298. Washington, DC: Center for Global Development, 2012. http:// Pol Wkly 2012;XLVII:60–8. www.cgdev.org/content/publications/detail/1426275 (accessed 15 42. King G, Gakidou E, Imai K, et al. Public policy for the poor? Mar 2013). A randomised assessment of the Mexican universal health 17. Palacios R, Das J, Sun C, eds. India’s health insurance scheme for programme. Lancet 2009;373:1447–54 the poor: evidence from the early experience of Rashtriya Swasthya 43. Xu K, Evans DB, Kawabata K, et al. Household catastrophic health Bima Yojana. New Delhi: Center for Policy Research, 2011. expenditure: a multicountry analysis. Lancet 2003;362:111–17. 18. Rashtriya Swasthya Bima Yojana overview and scheme details. 44. Naga Sridhar G. Microfinance institutions: Moving beyond AP crisis. http://www.rsby.gov.in (accessed 30 Apr 2013). The Business Line [Newspaper Online]. 2012. http://www. 19. Rathi P. http://www.priorities2012.com/documents/3d-4-patreek.pdf thehindubusinessline.com/companies/microfinance-institutions- 20. NRHM Mission Document. http://www.nird.org.in/brgf/doc/Rural% moving-beyond-ap-crisis/article4253239.ece 20HealthMission_Document.pdf (accessed 30 Apr 2013). 45. Naga Sridhar G. Life after microfinance in AP. The Business Line 21. State HMIS data analysis, April’10–March’11 Andhra Pradesh, [Newspaper Online]. 2013. http://www.thehindubusinessline.com/ prepared by NHSRC. http://cfw.ap.nic.in/nrhm/pdf/AP_Analysis_ opinion/life-after-microfinance-in-ap/article4273183.ece Apr2010-Mar2011.pdf (accessed 2 May 2013). 46. Prasad NP, Raghavendra P. Healthcare models in the era of medical 22. NRHM achievements in Maharashtra. http://www.nrhm.maharashtra. neoliberalism. A study of Aarogyasri in Andhra Pradesh. Econ Pol gov.in/achievements.htm (accessed 2 May 2013). Wkly 2012;XLVII:118–26. 23. Health information helpline. http://www.hmri.in/oursolutions- 47. Health Systems 20-20. An Evaluation of the effects of the National healthinformation.html (accessed 30 Apr 2013). Health Insurance Scheme in Ghana. Bethesda: Health Systems 24. About the project/scheme. http://www.maha-arogya.gov.in/ 20-20, 2009. projectandschemes/Jeevandaiaarogya/default.htm (accessed 48. http://www.jointlearningnetwork.org/programs/compare/benefits/ 30 Apr 2013). 142%2C16 25. Jeevandayee scheme performance: Information received from Rajiv 49. Ghosh S. Catastrophic payments and impoverishment due to Gandhi Jeevandayee Arogya Yojana Society Government of out-of-pocket health spending. Econ Pol Wkly 2011;XLVI:63–70. Maharashtra on 17 May 2013. 50. Garg CC, Karan AK. Reducing out-of-pocket expenditures to reduce 26. Projects and schemes. http://maha-arogya.gov.in/projectandschemes/ poverty: a disaggregated analysis at rural-urban and state level in itdpnavsanjivani%5Cdefault.htm (accessed 30 Apr 2013). India. Health Policy Plan 2009;24:116–28. 27. Mallepeddi R, Pernefeldt H, Bergkvist S. Andhra Pradesh health 51. Rajiv Jeevandayi scheme details. http://www.jeevandayee.gov.in/ sector reform, a narrative case study, Technical paper No.7 for The RGJAY/FrontServlet?requestType=CommonRH&actionVal= Rockefeller Foundation–Sponsored Initiative on the Role of the RightFrame&page=undefined%3E%3E%3Cb%3ERGJAY%3C/b% Private Sector in Health Systems in Developing Countries. 3E&pageName=RGJAY&mainMenu=About&subMenu=RGJAY 2009:42–8. (accessed 30 Apr 2013). 28. EMRI. http://www.emri.in/states.html (accessed 30 Apr 2013). 52. Tangcharoensathien V, Swasdiworn W, Jongudomsuk P, et al. 29. Developing and evaluating complex interventions: new guidance. Universal Coverage Scheme in Thailand: equity outcomes and London: Medical Research Council, 2008. future agendas to meet challenges. Geneva: World Health 30. Cesar GV, Jean-Pierre H, Jennifer B. Evidence-based public health: Organization, 2010. moving beyond randomized trials. Am J Public Health 53. Dilip TR. On Publicy-Financed Health Insurance Schemes. Is the 2004;94:400–5. analysis premature? Econ Pol Wkly 2012;XLVII8:79–80. 31. Presentation on Annual Plan 2012-13 and Five Year Plan 2012-17. 54. Vallakal S, Ebrahim S. Publicly-Financed Health Insurance http://planningcommission.nic.in/plans/stateplan/Presentations12_13/ Schemes. Concerns about impact assessment. Econ Pol Wkly maharashtra1213.pdf (accessed on 25 Mar 2013). 2013;XLVIII:24–7. Rao M, Katyal A, Singh PV, et al. BMJ Open 2014;4:e004471. doi:10.1136/bmjopen-2013-004471 15 Downloaded from http://bmjopen.bmj.com/ on November 30, 2015 - Published by group.bmj.com Changes in addressing inequalities in access to hospital care in Andhra Pradesh and Maharashtra states of India: a difference-in-differences study using repeated cross-sectional surveys Mala Rao, Anuradha Katyal, Prabal V Singh, Amit Samarth, Sofi Bergkvist, Manjusha Kancharla, Adam Wagstaff, Gopalakrishnan Netuveli and Adrian Renton BMJ Open 2014 4: doi: 10.1136/bmjopen-2013-004471 Updated information and services can be found at: http://bmjopen.bmj.com/content/4/6/e004471 These include: References This article cites 8 articles, 2 of which you can access for free at: http://bmjopen.bmj.com/content/4/6/e004471#BIBL Open Access This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 3.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/3.0/ Email alerting Receive free email alerts when new articles cite this article. Sign up in the service box at the top right corner of the online article. Topic Articles on similar topics can be found in the following collections Collections Health services research (783) Public health (1259) Notes To request permissions go to: http://group.bmj.com/group/rights-licensing/permissions To order reprints go to: http://journals.bmj.com/cgi/reprintform To subscribe to BMJ go to: http://group.bmj.com/subscribe/