H N P D I S C U S S I O N P A P E R Synopsis of Results on the Impact of Community- Based Health Insurance on Financial Accessibility to Health Care in Rwanda Pia Schneider and Francois Diop September 2001 SYNOPSIS OF RESULTS ON THE IMPACT OF COMMUNITY-BASED HEALTH INSURANCE ON FINANCIAL ACCESSIBILITY TO HEALTH CARE IN RWANDA Pia Schneider and François Diop September 2001 Health, Nutrition and Population (HNP) Discussion Paper This series is produced by the Health, Nutrition, and Population Family (HNP) of the World Bank's Human Development Network (HNP Discussion Paper). The series provides a vehicle for publishing preliminary and unpolished results on HNP topics to encourage discussion and debate. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors and should not be attributed in any manner to the World Bank, to its affiliated organizations, or to members of the Bank's Board of Executive Directors or the countries they represent. Citation and the use of material presented in this series should take into account this provisional character. For free copies of papers in this series, please contact the individual authors whose name appears on the paper. Inquiries about the series and submissions should be directed to the Editor in Chief. Submissions should have been previously reviewed and cleared by the sponsoring department that will bear the cost of publication. No additional reviews will be undertaken after submission. The sponsoring department and authors bear full responsibility for the quality of the technical contents and presentation of material in the series. Since the material will be published as presented, authors should submit an electronic copy in a predefined format as well as three camera-ready hard copies (copied front to back exactly as the author would like the final publication to appear). Rough drafts that do not meet minimum presentational standards may be returned to authors for more work before acceptance. The Editor in Chief of the series is Alexander S. Preker (apreker@worldbank.org); For information regarding this and other World Bank publications, please contact the HNP Advisory Services (healthpop@worldbank.org) at: Tel (202) 473-2256; and Fax (202) 522-3234. ISBN 1-932126-08-2 © 2001 The International Bank for Reconstruction and Development / The World Bank 1818 H Street, NW Washington, DC 20433 All rights reserved. ii Synopsis of Results on the Impact of Community-Based Health Insurance on Financial Accessibility to Health Care in Rwanda Pia Schneider,a François Diopb aHealth economist, Partnerships for Health Reform (PHR); Abt Associates Inc., Bethesda, MD bSenior health economist, Partnerships for Health Reform (PHR); Abt Associates Inc., Bethesda, MD Prepared for the Partnerships for Health Reform Bethesda, Maryland, USA, Version: September, 2001 Abstract: This paper evaluates the impact of prepayment schemes on access to health care for poor households, based on household survey data. Rwanda is one of the poorest countries in the world. After the genocide in 1994, public health care services were provided for free to patients, financed by donors and the government. In 1996, the Ministry of Health reintroduced prewar level user charges. By 1999, utilization of primary health care services had dropped from 0.3 in 1997 to a national average of 0.2 annual consultations per capita. This sharp drop in health service use combined with growing concerns about rising poverty, poor health outcome indicators, and a worrisome HIV prevalence among all population groups motivated the Rwandan government to develop community-based health insurance to assure access to the modern health system for the poor. In early 1999, the Rwandan Ministry of Health, in collaboration with the local communities, and the technical support of the USAID-funded Partnerships for Health Reform project (PHR) started the process to pilot test prepayment schemes in three districts. At the end of their first operational year, the 54 schemes counted more than 88,000 members. The findings presented in this paper reveal that insurance enrollment is determined by household characteristics such as the health district of household residence, education level of household head, family size, distance to the health facility, and radio ownership, whereas health and economic indicators did not influence enrollment. Insurance members report up to five times higher health service use than nonmembers. The analysis confirms findings reported by PHR based on provider data: health insurance has significantly improved equity in health service use for members while at the same time out-of-pocket spending has gone down per episode of illness. Keywords: prepayment schemes; enrollment in community-based health insurance; equity in health service use; Rwanda. Disclaimer: The findings, interpretations, and conclusions expressed in the paper are entirely those of the authors, and do not necessarily represent the views of the World Bank, its Executive Directors, or the countries they represent. Correspondence Details: Pia Schneider, Abt Associates Inc., 4800 Montgomery Lane, Suite 600, Bethesda, MD 20814-5341. Tel. (301) 347-5376. Fax (301) 913-0562. Email: Pia_Schneider@abtassoc.com; Web: www.phrproject.com iii iv Contents PREFACE.............................................................................................................................................. VII ACKNOWLEDGMENTS ......................................................................................................................IX I. INTRODUCTION............................................................................................................................. 1 II. BACKGROUND................................................................................................................................ 2 III. DATA SOURCES AND METHODOLOGY ................................................................................... 3 A. MODEL 1: DEMAND FOR HEALTH INSURANCE ..................................................................... 4 B. DESCRIPTION OF VARIABLES INCLUDED IN MODEL 1........................................................ 5 C. MODEL 2: ACCESS TO THE MODERN HEALTH CARE SYSTEM........................................... 5 D. DESCRIPTION OF VARIABLES INCLUDED IN MODEL 2........................................................ 6 E. MODEL 3: FINANCIAL IMPACT OF HOUSEHOLD OUT-OF-POCKET HEALTH EXPENDITURES................................................................................................................................... 6 F. DESCRIPTION OF VARIABLES INCLUDED IN MODEL 3 ........................................................ 7 IV. RESULTS............................................................................................................................................ 7 A. DESCRIPTION OF THE SAMPLE GROUP.................................................................................... 8 B. MODEL 1: WHO DEMANDS HEALTH INSURANCE?.............................................................. 11 C. MODEL 2: EQUITY IN FINANCIAL ACCESSIBILITY TO PROFESSIONAL CARE.............. 13 D. MODEL 3: AVERAGE OUT-OF-POCKET HEALTH EXPENDITURES PER EPISODE OF ILLNESS............................................................................................................................................... 15 V. DISCUSSION AND CONCLUSION................................................................................................ 20 VI. REFERENCES ................................................................................................................................. 21 v vi PREFACE In January 2000, Dr. Gro Harlem Bruntland, Director General of the World Health Organization (WHO), established a Commission on Macroeconomics and Health (CMH) to provide evidence about the importance of health to economic development and poverty alleviation. This HNP Discussion Paper is based on a Report on community financing submitted in September 2001 to Working Group 3 of the CMH. The mandate of Working Group 3 was to examine alternative approaches to domestic resources mobilization, risk protection against the cost of illness and resource allocation. The working group was chaired by Professor Alan Tait (Former Deputy Director of Fiscal Affairs, International Monetary Fund, and currently Honorary Fellow at University of Kent at Canterbury and Honorary Fellow at Trinity College, Dublin) and Professor Kwesi Botchewey (Director of Africa Research and Programs at the Harvard Center for International Development). Professor Jeffery D. Sachs (Chairman of the Commission and Director of the Harvard Center for International Development) presented the findings of the CMH in a Report that was submitted to WHO on December 20, 2001--Macroeconomics and Health: Investing in Health for Economic Development. The report of the CMH recommended a six-pronged approach to domestic resource mobilization at low- income levels: "(a) increased mobilization of general tax revenues for health, on the order of 1 percent of GNP by 2007 and 2 percent of GNP by 2015; (b) increased donor support to finance the provision of public goods and to ensure access for the poor to essential health services; (c) conversion of current out- of-pocket expenditure into prepayment schemes, including community financing programs supported by public funding, where feasible; (d) a deepening of the HIPC (Highly Indebted Poor Countries) initiative, in country coverage and in the extent of debt relief (with support form the bilateral donor community); (e) effort to address existing inefficiencies in the way in which government resources are presently allocated and used in the health sector; and (f) reallocating public outlays more generally from unproductive expenditure and subsidies to social-sector programs focused on the poor." Most community financing schemes have evolved in the context of severe economic constraints, political instability, and lack of good governance. Usually government taxation capacity is weak, formal mechanisms of social protection for vulnerable populations absent, and government oversight of the informal health sector lacking. In this context of extreme public sector failure, community involvement in the financing of health care provides a critical but insufficient first step in the long march toward improved access to health care by the poor and social protection against the cost of illness. The CMH stressed that community financing schemes are no panacea for the problems that low-income countries face in resource mobilization. They should be regarded as a complement to--not as a substitute for--strong government involvement in health care financing and risk management related to the cost of illness. Based on an extensive survey of the literature, the main strengths of community financing schemes are the degree of outreach penetration achieved through community participation, their contribution to financial protection against illness, and increase in access to health care by low-income rural and informal sector workers. Their main weaknesses are the low volume of revenues that can be mobilized from poor communities, the frequent exclusion of the poorest from participation in such schemes without some form of subsidy, the small size of the risk pool, the limited management capacity that exists in rural and low- income contexts, and their isolation from the more comprehensive benefits that are often available through more formal health financing mechanisms and provider networks. vii The work by the CMH proposed concrete public policy measures that governments can introduce to strengthen and improve the effectiveness of community involvement in health care financing. This includes: (a) increased and well-targeted subsidies to pay for the premiums of low-income populations; (b) use of insurance to protect against expenditure fluctuations and use of reinsurance to enlarge the effective size of small risk pools; (c) use of effective prevention and case-management techniques to limit expenditure fluctuations; (d) technical support to strengthen the management capacity of local schemes; and (e) establishment and strengthening of links with the formal financing and provider networks. The report presented in this HNP Discussion Paper has made a valuable contribution to our understanding of some of the strengths, weaknesses and policy options for securing better access for the poor to health care and financial protection against the impoverishing effects of illness, especially for rural and informal sector workers in low-income countries. Alexander S. Preker Chief Economist Health, Nutrition, and Population The World Bank viii ACKNOWLEDGMENTS The authors are grateful to the World Health Organization (WHO) for having provided an opportunity to contribute to the work of the Commission on Macroeconomics and Health and to the World Bank for having published the report as an HNP Discussion Paper. This analysis is supported by the Partnerships for Health Reform Project, which is administered by Abt Associates Inc. and funded by the United States Agency for International Development (USAID). Helpful comments on this analysis were received from Charlotte Leighton, A.K. Nandakumar, and Manjiri Bhawalkar. ix x I. INTRODUCTION In response to declining health service utilization after the reintroduction of user fees for services and drugs in public health facilities in 1996, the Rwandan Ministry of Health (MOH) decided to pilot test alternative health care financing and provider payment methods. During the first six months in 1999, the MOH developed an on-going collaboration with the local communities, 54 community-based health insurance (CBHI) plans--each of them partnering with a health center. The MOH selected three pilot districts, Kabutare, Byumba and Kabgayi, to pilot test these prepayment schemes (PPS). The three districts were chosen based on the extent of their health infrastructure, the repeated demand for technical assistance from the population in developing and implementing CBHI, and the districts' political will to participate in the health insurance pilot experience. By July 1, 1999, all 54 health centers, which had collaborated in the set up of their CBHI in the three pilot districts, signed a contract with their partnering health insurance scheme. Thereafter, the district population started to enroll in the schemes, which are democratically managed by their members as mutual health associations. At the end of the first year, membership in the 54 health insurance plans mounted to 88,303 individuals, corresponding to 8 percent of the total population of the three districts (Table 1). Technical and financial assistance for the entire design, development, implementation and evaluation phase was provided to the Rwandan government by Partnerships for Health Reform (PHR), a project funded by USAID and administered by Abt Associates1. Table 1: Community-Based Health Insurance in Rwanda, First Year Performance (7/1999­6/2000) Pilot Districts with CBHI Indicators Byumba Kabgayi Kabutare All 3 Districts All Prepayment Schemes (Status on 6/30/2000): Total Number of PPS 21 17 16 54 Total Target Population in Districts 459,329 368,020 288,160 1,115,509 Total Population Enrolled 48,837 21,903 17,563 88,303 Average Number of Members per PPS 2,326 1,288 1,098 1,635 First year average PPS Enrollment Rate 10.6% 6.0% 6.1% 7.9% Source: PHR TR No. 61) The objective of this paper is to respond to two questions on the prepayment schemes' impact. First, what are the population groups that enroll in community-based health insurance schemes; and second, does health insurance membership improve financial accessibility to care without increasing the burden of out- out-pocket health expenditures. These questions are addressed by presenting a synopsis of the findings from the household survey conducted by PHR in the three health districts. The impact of prepayment schemes on insurance and providers' utilization, cost, and finances has been analyzed from monthly routine data collected from providers and health insurance schemes over a two-year period in the three districts. Findings of this detailed analysis on the financial sustainability of CBHI and their impact on health care providers are presented in the PHR Technical Report No. 61. This paper will therefore 1See: Schneider, P. et al. 2000, PHR TR 45 for detailed information on the development and implementation of prepayment schemes in Rwanda. 1 respond to the question about the insurance impact on households' financial accessibility to the modern health care system by focusing on information collected in the household survey. The following section provides a brief summary of the design, development, and implementation of prepayment schemes in Rwanda, which took place from January 1999 until September 2000. The third section introduces the method used to address the research questions and presents the household survey data and variables used. Results are presented in the fourth section; and the fifth section has a discussion and conclusion. II. BACKGROUND In January 1999, the Rwandan MOH initiated the design and development phase of prepayment plans by creating a strategic steering committee, which was headed by the Director of Health Care and included representatives and stakeholders from the central and regional levels. The design and implementation of health insurance modalities and management features were discussed and agreed upon during 28 district- level workshops attended by community and health care representatives and in a series of community gatherings with the local population. Proposals stemming from these district and community meetings were shared with the central steering committee, which provided feedback and advice to the communities. As a result of this on-going discussion between the central and local level, the scheme features were designed, the legal, contractual, and financial tools were developed, and workshop participants were trained and prepared to manage the 54 prepayment schemes, each entering into partnership with a health center on July 1, 1999. Under Rwandan law, the schemes are deemed mutual health associations, headed by an executive bureau with four volunteers, elected by and among the scheme members during a CBHI general assembly. At district level, the schemes have formed a federation. Six members have been elected by and among all PPS executive bureau representatives in their general assembly to constitute the district federation of prepayment schemes. The federation is the partner to the district hospital as well as to the health district and other authorities. Each prepayment bureau has signed a contract with the affiliated health center, and each federation with the district hospital, defining in 17 articles, the rules of collaboration between the insurer and provider. According to the schemes' by-laws, members are invited at least once a year to attend the prepayment scheme general assembly. Individuals and households, who would like to be insured, pay at the time of enrollment an annual premium of 2,500 francs2 per family up to seven persons to the CBHI affiliated with their "preferred" health center3. In case of sickness, members contact first their preferred health center, which is usually their nearest public or church-owned facility. Health centers play a gatekeeper function, and hospital services are covered for members only if referred by their preferred health centers. This is done to dissuade members and providers from frivolous use of more expensive hospital services. PPS membership entitles members--after a one-month waiting period--to a basic health care package covering all services and drugs provided in their preferred health center, including ambulance transfer to the district 2July 1999: RWF 2,500 = US$7.50. 3 Premium rates were set by taking into account existing user fees and by assuming that utilization rates would increase by 25 percent compared to baseline levels. See TR 45: chapter 3.1.1. 2 public or church-owned hospital, where a limited package is covered4. Members pay a 100 francs5 copayment for each health center visit. The MOH was concerned that the availability of health insurance to a population group with an accumulated demand for health services could lead to adverse selection and moral hazard, causing health care costs to rise. Therefore, the MOH recommended to the district workshop participants to incorporate into the design of the prepayment schemes a provider-payment mechanism that will set the necessary incentives for providers to improve their productivity while controlling for unnecessary use of health services. After several discussions between providers and future scheme managers, the workshop participants selected capitation provider payment to the health center, as a measure to control cost escalations caused by supply-side induced increases in demand for health care. Each prepayment bureau disburses monthly one-twelfth of its accumulated premium fund, of which 5 percent is withheld to cover the scheme's administrative costs, 10 percent is paid to the district's prepayment federation, while the rest is paid as capitation payment to the partnering health center. The federation reimburses the district hospital for covered services provided to members, paying per episode of illness (for C-sections, malaria, and pediatrics), and by service for overnight stays and physician consultations for all other illnesses. Thus, insurance members share their hospital costs on a district level and health center costs on a health center catchment area level6. III. DATA SOURCES AND METHODOLOGY The analysis presented in this paper is based on data collected in the prepayment scheme household survey, conducted by PHR in collaboration with the Rwandan National Population Office (ONAPO). Data collection took place during 40 days in October/November 2000. The household survey includes 2,518 households that were successfully interviewed in the three pilot districts, and number 11,583 individuals. The sample was designed to provide information on the impact of prepayment schemes on households' enrollment and health care seeking behavior, as well as the related financial implications. The sample was based on the same sampling frame as the Rwandan Demographic and Health Survey (DHS), 2000, covering 11 health regions in Rwanda7. Households for the prepayment household survey in the three districts have been sampled at random from a list of primary sample households from sample cells identified in the national DHS sample, rendering the household survey sample representative to the district level. The prepayment household survey used three structured questionnaires for data collection, first, a socioeconomic household questionnaire, second, a curative questionnaire, and third a preventive care questionnaire. The household questionnaire collected information on households' and individuals' sociodemographic and economic characteristics including household expenditures for consumer goods, 4The Kabgayi PPS covers at the hospital full episodes of C-section, Malaria and non-surgical pediatrics, whereas the Kabutare and Byumba PPS cover C-section full episode, as well as each service of physician consultation, and overnight stay at the district hospitals. 5July 1999: RWF 100 = US$ 0.3. 6See Schneider et al. (2001)a for detailed analysis of PPS impact on utilization, cost and finances in health centers. 7The DHS has been conducted by the Rwandan National Population Office (ONAPO) in collaboration with Macro International and USAID in 2000/01. Households for the DHS have been selected as primary sample units from sample cells identified for the Living Condition Monitoring Survey (LCMS), conducted by the Ministry of Finance in collaboration with UNDP in 2000/1. 3 health, and education, and participation in CBHI. The curative care questionnaire was addressed to household members who were sick two weeks prior to the interview, and the preventive care questionnaire was used to interview women of childbearing-age who had delivered a child in the last five years or who were pregnant during the year preceding the interview8. The three pilot districts (Byumba, Kabgayi, Kabutare) are similar in their socioeconomic situation. There is little urban activity, and the population is mainly active in agriculture and animal husbandry. Households are assumed to be equally poor with few of them owning cattle, which are a sign of wealth. For the analysis, the sample population is divided into two groups: first, CBHI members in pilot districts, and second, CBHI nonmembers in pilot districts. Three models are used in this paper to estimate first, the probability of buying health insurance for specific population groups in the three pilot districts; second, the probability of access to basic health care services for the insured and uninsured population groups in pilot districts; and third, the estimated out-of- pocket health expenditures per episode of illness for all sick individuals and for those who sought professional care, based on a set of explanatory variables. For each categorical variable used in the three models, one category has been selected as a reference category. Odds ratios are estimated in the logit regression models for each category to estimate the factor that measures the magnitude of the difference in relation to the reference category. Interaction effects were tested for significance. A. MODEL 1: DEMAND FOR HEALTH INSURANCE The following model estimates the probability of CBHI enrollment for households in pilot districts. The objective is to determine if the poorest buy basic health insurance, and if the poor benefit from a re- distribution from richer members of the financial pool. The willingness to join CBHI is a discrete choice--to join or to not join. A logit regression model is used to determine households' CBHI enrollment probability, and the extent to which this decision is influenced by specific sociodemographic and--economic characteristics. The hypothesis to be tested is that the CBHI member and nonmember households do not differ in their socioeconomic characteristics. In a logit regression, the dependent variable "demand for insurance" Di , will equal 1 if individuals buy insurance, or zero otherwise. Formally, the logit model can be written as a linear function of the explanatory variables: (1) Li = b1 + b2 X2i + .... + bk Xki and (2) Pi (D for CBHI membership) = 1 / ( 1 + 1/e ) Li The second equation shows that the conditional probability of buying insurance Pi is a non-linear function of the explanatory variables Xi, which represents a series of attributes assumed to have caused a household to buy health insurance membership in the three pilot districts. We will estimate the unknown coefficients bi which are the weights assigned each of the households' sociodemographic and economic characteristics in the probability that Di = 1 for given Xi. Insurance was an option only for households within pilot districts. Therefore, the logit regression was performed with weighted household survey data from pilot districts, based on household heads as the unit of analysis. 8Diop and Schneider (2001a). 4 B. DESCRIPTION OF VARIABLES INCLUDED IN MODEL 1 The response to enroll in CBHI is the dependent variable in this model and is made primarily by the head of household, based on a set of independent variables X. These explanatory variables are classified into demographic, socioeconomic, and health attributes of the household. Table 2 presents their sample size, mean, and standard deviation. Table 2: Summary Statistics: Independent Variables used to Determine Probability of CBHI Independent Variables Mean Std. Dev. N (Households) District Kabgayi 0.41 0.492 2,518 District Byumba 0.22 0.411 2,518 District Kabutare 0.37 0.482 2,518 Male household head 0.66 0.474 2,518 Average age household head 44.33 15.789 2,512 Household head 40 years and older 0.83 0.377 2,518 Household head attended school 0.54 0.498 2,515 Household with child <5 0.07 0.249 2,518 Household with pregnancy in past year 0.03 0.167 2,518 Household with cattle 0.18 0.386 2,503 Household with radio 0.34 0.475 2,505 Household with 5 and more members 0.45 0.498 2,518 Household size 4.57 2.255 2,518 Less than 30 min. from household cell to 0.38 0.487 2,518 h-facility Income Quartiles 1 0.26 0.437 2,518 2 0.25 0.431 2,518 3 0.25 0.435 2,518 4 0.24 0.430 2,518 C. MODEL 2: ACCESS TO THE MODERN HEALTH CARE SYSTEM Patients' health-seeking behavior was measured based on weighted data for individuals who reported sickness during the two weeks preceding the interview in the household survey and who responded to the curative care questionnaire. As in the first model, the second model applies a logit regression model to estimate the probability of entering (or not entering) the modern health care system for the insured and uninsured in pilot districts. Access probabilities are estimated based on specific sociodemographic and economic household characteristics that determine a sick individual's care-seeking behavior. The hypothesis is tested that the sick who access health care do not significantly differ in their sociodemographic, economic, and health characteristics. Therefore, the logit regression was performed with weighted curative survey data from pilot districts, based on sick individuals as the unit of analysis. The logit model, based on equation (1) presented in the first model, leads to the following definition of the probability of accessing modern health care: (3) Pi (Access to professional care) = 1 / ( 1 + 1/e ) Li 5 Where, X represents a set of explanatory variables that are assumed to have caused a sick person to seek care with a professional provider at a health center or district hospital during the two weeks prior to the interview. D. DESCRIPTION OF VARIABLES INCLUDED IN MODEL 2 The decision to seek professional care is influenced by households' socioeconomic conditions, insurance status, and the sick individuals' health status. These explanatory variables are summarized in Table 3, showing for each attribute the sample size, mean, and standard deviation. Table 3: Summary Statistics of Explanatory Variables for Probability of Professional Visit Independent Variable Mean Std. Dev N (Sick Ind) District Kabgayi 0.23 0.421 3,130 District Byumba 0.37 0.482 3,130 District Kabutare 0.40 0.491 3,130 Prepayment member in pilot district 0.06 0.236 3,130 Male patient 0.42 0.494 3,130 Average age patient 24.60 21.003 3,127 Patient age 0-5 years 0.23 0.423 3,127 Patient with pregnancy in past year 0.07 0.251 3,130 Patient spent 4 and more days in bed 0.56 0.497 1,599 Household with 5 and more members 0.57 0.495 3,130 Household head attended school 0.56 0.496 3,130 Household with cattle 0.21 0.404 3,130 Household with radio 0.36 0.481 3,130 Less than 30 min. from household cell to health facility 0.39 0.488 3,130 Income Quartiles 1 0.23 0.421 3,130 2 0.26 0.441 3,130 3 0.26 0.439 3,130 4 0.24 0.430 3,130 E. MODEL 3: FINANCIAL IMPACT OF HOUSEHOLD OUT-OF-POCKET HEALTH EXPENDITURES The third model is a log-linear regression that serves to estimate first sick individuals' total average out- of-pocket spending per episode of illness, and second, total out-of-pocket spending conditioned on the positive use of health care services9. The model is a linear regression for the logarithm of total health- related spending per episode of illness of the sick, and for the logarithm of total health-related spending for the sick who reported at least one visit. The logarithmic transformation of health expenditures per 9Manning, G., et al., 1987. Yip, W., and P. Berman, 2001. 6 episode of illness eliminates skewness in the distribution of health expenses among users, yielding roughly normal error distributions. The model can be written as follows: (3a) Log(total illness related out-of-pocket spending) = a + b X + e , (3b) Log(total illness related out-of-pocket spending¦ visit >0) = a + b X + e where X represents a set of continuous and dummy attributes assumed to influence patients' health expenditures. Detailed health expenditures are reported by episode of illness, which includes spending before and during a professional care visit, and will show to what extent patients rely on alternative sources of care outside the formal health sector. The regressions were performed with weighted curative survey data from pilot districts, based on sick individuals as the unit of analysis. It is assumed that the amount spent on non-professional medicine will be higher for patients whose access to professional care is limited by financial barriers. F. DESCRIPTION OF VARIABLES INCLUDED IN MODEL 3 Model 3 uses the same variables as in the second model and adds the variable professional care visit to estimate out-of-pocket health expenditures for those individuals who were sick, and for those who sought professional care. Table 4 presents sample size, mean, and standard deviation for each explanatory variable. Table 4: Summary Statistics of Explanatory Variables for Estimated Out-of-Pocket Spending per Episode of Illness Independent Variable Mean Std. Dev N (Sick Ind) All sick with 1+ professional care visit 0.16 0.371 3,130 District Kabgayi 0.23 0.421 3,130 District Byumba 0.37 0.482 3,130 District Kabutare 0.40 0.491 3,130 Prepayment member in pilot district 0.06 0.236 3,130 Male patient 0.42 0.494 3,130 Average age patient 24.60 21.003 3,127 Patient age 0-5 years 0.23 0.423 3,127 Patient with pregnancy in past year 0.07 0.251 3,130 Patient spent 4 and more days in bed 0.56 0.497 1,599 Household with 5 and more members 0.57 0.495 3,130 Household head attended school 0.56 0.496 3,130 Household with cattle 0.21 0.404 3,130 Household with radio 0.36 0.481 3,130 Less than 30 min. from household cell to 0.39 0.488 3,130 health facility Income Quartiles 1 0.23 0.421 3,130 2 0.26 0.441 3,130 3 0.26 0.439 3,130 4 0.24 0.430 3,130 IV.RESULTS 7 A. DESCRIPTION OF THE SAMPLE GROUP Table 5 and 6 describe sociodemographic and economic characteristics for the sample population included in the household survey conducted in the three pilot districts. Males head almost 82 percent of PPS member households. One third of the heads of member household have 5 or more years of schooling, 41 percent of the household heads are in the age group of 40-59 years and belong to households in higher expenditure quartiles as compared to the nonmember households in the same districts. Distance to the health facility also seems to be an important criterion as almost 50 percent of the member households live within 15 minutes of the health facility. Membership starts to taper off as the distance to the health facility increases. Table 5: Descriptive Sample Characteristics (Column percentages sum to 100 within each category) Characteristics Pilot Districts Head of Household Nonmembers (n PPS Members Total 2,337) (n 181) (n 2,518) Gender Female 35.4% 17.8% 34.1% Male 64.6% 82.2% 65.9% Level of schooling Never 47.0% 28.4% 45.7% Primary <5 22.6% 22.5% 22.6% Primary =>5 24.2% 33.3% 24.9% Above primary 6.2% 15.9% 6.9% Age group < 30 17.1% 17.6% 17.2% 30-39 25.2% 27.2% 25.4% 40-59 36.3% 41.2% 36.7% 60 & + 21.3% 14.0% 20.8% Income quartiles, household 1 26.1% 19.8% 25.6% 2 24.7% 24.2% 24.6% 3 25.0% 29.9% 25.3% 4 24.3% 26.1% 24.4% Time distance (min) from household to health facility 15 37.6% 48.6% 38.4% 45 14.6% 20.8% 15.1% 75 25.6% 24.2% 25.5% 105 22.1% 6.4% 21.0% Almost 61 percent of insured households interviewed in the survey said they have enrolled all individuals living in the household. Those who have not, usually excluded young adults above the age of 18, still 8 living in the household, and who are supposed to enroll either in a group category or as individuals10. Table 6 presents households' monthly average per capita monetary expenditures for each expenditure quartile, as well as the average household size for insured and uninsured households. Independent means tests were performed to compare differences. Households' monetary expenditures were used as proxy to classify households in income quartiles. Insured households number on average significantly more individuals than the uninsured in pilot districts. The possibility of signing up in CBHI as a family up to seven members for the same annual premium might have been an incentive for larger households to enroll with all their family members. Each of the two groups shows a decreasing average household size with higher expenditure quartiles. Table 6: Household Characteristics, by Income Quartile Household Characteristics Nonmembers PPS Members Total Monthly average per capita expenditure (RWF) Quartile 1 333 347 334 Quartile 2 1,050 1,007 1,047 Quartile 3 2,241 2,056 2,225 Quartile 4 8,154 9,367 8,247 Total RWF 2,884 3,370 2,919 Average household size, number of individuals Quartile 1 4.6 5.5 4.7 Quartile 2 4.7 5.6 4.8 Quartile 3 4.6 5.8 4.7 Quartile 4 4.1 4.8 4.1 Total Individuals 4.5 5.5*** 4.6 Note: t-tests were performed to compare the average values of the insured with the uninsured sample. *** Significant at 1 percent level of significance. Smaller households in higher expenditure quartiles pay the same premium per household as larger families in lowest expenditure quartiles. Depending on members' service use and financial contribution to health for uncovered services, this negative relationship between household size and income status can lead to a cross-subsidies from the smaller, predominantly richer, to the larger and poorer families in the prepayment health insurance pool. Table 7 presents the summary statistics for monthly per capita monetary expenditures for households living in the pilot districts. The following histogram visualizes that the monetary expenditure data are very right-skewed, confirming that this is an equally poor population with very few reporting high monetary expenditures: the 90th percentile amount is 5,975 francs compared to the maximum amount of 192,950 francs. Table 7: Summary Statistics on Distribution of Monthly Monetary Expenditures Per Capita 10The enrollment category "household" includes two adults and all children up to the age of 18 living in the same household. Other household members will need to enroll in a group or individual category. 9 Monetary Expenditures Households N 2,518 Mean RWF 2,919.1 Std. Error of Mean 126.798 Median 1,475.5 Std. Deviation 6,362.787 Minimum 0 Monetary Expenditures ( continued) Households N Maximum 192,950 Percentiles 10 267.0 25 624.5 50 1,475.5 75 3,190.0 90 5,975.0 Monthly Monetary Expenditure 3000 2000 1000 Households Std. Dev = 6362.79 of Mean = 2919.1 Nbr 0 N = 2518.07 500 0.025000.045000.065000.085000.01050001.0 25000 145000165000.10 85000 .0 .0 .0 RWF per capita The histogram shows that Rwandan households living in these rural districts are poor. They live mostly from subsistence farming in areas with a high-density population. Rwanda is recovering from the recent civil war, and an estimated 10 percent of the male population is still missing. Muller (1997) found in a household survey conducted in 1983 that the average land area farmed by Rwandan households is very small with 1.24 ha and households produced agricultural product worth an average of $51 per capita per year, 90 percent of which is used for consumption. Findings in our household survey, conducted 17 years later, estimated annual monetary expenditure is approximately $100 per capita for these rural households. 10 The Rwandan Ministry of Finance is conducting a living standard survey in Rwanda, which will provide more insight into the socioeconomic conditions of Rwandan households. B. MODEL 1: WHO DEMANDS HEALTH INSURANCE? The means comparison in Tables 3.1 and 3.2 has shown that, compared to the uninsured, insured households are more likely to be headed by a male individual, who has attended some schooling. Also proportionally more CBHI member households were likely to come from higher income quartiles and from larger households. The logit regression results presented in Table 8 show that the level of education of household head, family size, household's district of residence, distance to the health facility, and radio ownership are the major determining factors that influence whether to join a health insurance or not. Households' health and economic indicators did not influence demand for health insurance. Radio ownership is indicative of a household's ability to access information and exposure to advertising about the CBHI. It may also be seen to a certain extent as an economic indicator for these very poor households. Households who live in Kabgayi are more than twice, and those in Byumba almost 15 times more likely to buy health insurance as households in Kabutare. Household heads who attended school are 103 percent more likely to enroll in health insurance than the illiterate. Large households with five and more members are 60 percent more likely to buy insurance than smaller households. This is likely because, irrespective of the family size (up to seven members), households pay 2500 francs membership fee per year11. Therefore, larger families effectively pay less per household member. Households who live within 30 minutes of their health facility have a 296 percent higher probability of joining than those who live further away. This latter result might have been influenced by health centers' and prepayment schemes' awareness campaign, which could have been more intense in the vicinity of a health facility. Households who own a radio are 47 percent more likely to enroll than those without radio, another result that might have been caused by the regular awareness campaign transmitted by radio. Although, male-headed households are 55 percent more likely to join than female-headed, and households with pregnant women are 23 percent more likely to join, these results are not significant. Other economic attributes, such as household cattle ownership and different income quartiles were not significant in the demand for health insurance. Households in the lowest and lower income quartile were as likely to enroll as those in the fourth income quartile. Table 8: Logit regression results for households' probability to demand community-based health insurance (prepayment schemes) 11Households up to seven members pay 2500 francs per year. Individual enrollment costs 2000 francs, and enrollment in a group for eight and more individuals costs 530 francs per person per year. Premiums are slightly higher in Kabgayi due to the larger hospital coverage (Household: 2600 francs, Individual: 2200 francs and group enrollment per person: 550 francs). 11 Insurance Membership in Pilot Districts Explanatory Variable Reference Category Variable Odds Ratio S.E. Sign Kabgayi district Kabutare district 3.51*** 0.362 0.001 Byumba district 15.80*** 0.268 0.000 Male hh head Female hh head 1.55 0.253 0.084 HH head, age 40+ HH head, younger than 40 1.13 0.239 0.598 HH head, attended school HH head, illiterate 2.03*** 0.196 0.000 Large HH size, 5+ Small HH size, less than 5 1.60*** 0.189 0.013 HH with child <5 No child <5 0.87 0.488 0.768 HH with pregnancy in past No pregnancy in past year year 1.23 0.674 0.761 Less than 30min to health More than 30min to health facility facility 3.96*** 0.187 0.000 HH with cattle No cattle 1.28 0.210 0.237 HH with radio No radio 1.47*** 0.184 0.038 Quartile 1 Quartile 4 1.19 0.264 0.513 Quartile 2 1.21 0.244 0.437 Quartile 3 1.15 0.228 0.535 Ancillary statistics: N (households) 2,474 - 2 Log likelihood 1054.901 Goodness fit (chi-squared test) 236.998 Degree of freedom 14 Nagelkerke R Square 22% Note: Hh=Household. Z-tests were performed to test the probability of enrollment for each characteristic in a logit model. *** Significant at 1 percent level of significance. ** Significant at 5 percent level of significance. The findings from the first model respond to the question, who enrolls in CBHI. Households living in Byumba and Kabgayi, who number five and more individuals, whose household head attended school, who live in the vicinity of a health center, and who own a radio appear to be more likely to buy insurance. Other important factors also influence households' probability of enrolliing such as their risk aversion, their exposure to effective information campaigns on prepayment schemes, as well as trust in the scheme management, which is related to households' willingness to see CBHI as an investment, and which supports the argument that enrollment in health insurance is not necessarily driven by economic conditions such as household income. The following reasons were identified in different surveys (focus group, household and patient exit interview survey) to be important in households' enrollment decisions: · Both districts Byumba and Kabgayi, had intensive awareness and information campaigns on PPS during the first year, supported by the district authorities and prepayment federation, resulting in steady monthly enrollment increases. · The prepayment schemes' features, including benefit package, premium level, enrollment categories, copayments, and waiting period, were designed, discussed, and agreed upon (by voting) in a series of about 30 workshops in the three districts. These workshops were attended by the local population. As a result, the health insurance schemes were "tailored" as desired by, and in response to, the needs of the local people. · The main determinant of PPS participation is trust, which might be captured by the time variable. People living near the facilities are more likely to enroll, because they know the health center personnel, as well as the prepayment scheme management team, and have been exposed to regular information campaigns on prepayment. · The participatory approach and the democratic management of PPS lead to sentiments of "ownership" and increased trust among the poor, which are basic conditions for poor households to engage in any investment. · Households that did not have the 2,500 francs (about US$8) to pay the one-year PPS enrollment fee have joined a "tontine." During five weeks, each tontine household paid 500 francs as an 12 installment toward the 2500 francs total fee to the "tontine-caisse." Households are enrolled as full members once they had contributed 2,500 francs. · Local initiatives (churches and members who attended the PPS general assemblies) helped to pay enrollment fees for indigents, widows, orphans, and poor high-risk patients such as HIV positive individuals. This shows that poor households will enroll in well-designed health insurance schemes that improve their access to health care. At the same time, these solidarity groups contribute to positive social capital in a society that is recovering from a civil war. Therefore, community-based health insurance becomes a form of social cohesion and provides a link between the poor and the health facilities. The following section answers the second question, whether health insurance membership improves financial accessibility to care without increasing members' financial burden. C. MODEL 2: EQUITY IN FINANCIAL ACCESSIBILITY TO PROFESSIONAL CARE12 Evaluation of the first year of prepayment schemes in Rwanda has been based on extensive data collection. The analysis of monthly health service utilization data in health centers and hospitals has revealed that the overall use of curative services for adults and children, and preventive health services for children and women, was up to five times higher for PPS members than for nonmembers13. Results in Table 9 reveal that the insured report considerably better access to the modern health care system with a visit probability of 0.45 compared to the uninsured in the pilot districts (0.15 visit probability). Given the relatively moderate visit probability for sick PPS members of 0.45, it can be assumed that prepayment enrollment was not driven by adverse selection, and there was no frivolous use of health care caused by members' moral hazard behavior. Probability of visit by members does not vary by patients' gender, age, and income quartile; it is determined by patients' geographical access to the health facility (time distance), and health status, with the sick and very sick being three times more likely to seek care than those who are healthier. Interestingly, CBHI members who said they were not very sick or were sick, reported higher visit likelihood (0.22 and 0.64, respectively) than the average uninsured (0.15), indicating that CBHI membership causes sick individuals to seek care at the onset of illness. Access to professional care is lowest for the uninsured in the lowest income quartile, who are about four times less likely to seek care than the same income group who is insured. Table 9: Probability of using a professional provider by insurance status Probability of Visit Pilot Districts Members (n 376) Nonmembers (n 3459) 12Professional care = public and church owned health centers, district hospitals, and dispensaries. This excludes sick individuals who sought care at traditional healers and others (e.g. ,drug vendors, pharmacy). 13Schneider, et al. (2001b). District averages are 1.5 curative consultations per member per year in Kabutare and Kabgayi, and 1.1 curative consultation per member in Byumba, whereas nonmembers' curative care consultation level scores around 0.2 consultation per nonmember per year. 13 Sick Individuals (n 3,835) 0.45*** 0.15 Patient gender Female 0.42 0.14 Male 0.50 0.16 Patient age 6 years and older 0.45 0.13 0-5 years 0.46 0.19 Time from household to h- facility More than 30min 0.33 0.12 Less than 30min 0.60 0.19 Income (Expenditure) groups Quartile 1 0.40 0.06 Quartile 2 0.35 0.13 Quartile 3 0.49 0.14 Quartile 4 0.54 0.26 Self-perceived health status Not very sick 0.22 0.05 Sick 0.64 0.15 Very sick 0.61 0.30 Note: Probability of sick individuals with at least one visit with a professional provider during the two weeks prior to the interview. T-tests were performed to compare the rates of the insured with the uninsured in the pilot districts. *** Significant at 1 percent level of significance. The following logit regression estimates the probability of a professional health care visit for members and nonmembers, and controls for skewness in the data distribution that could have influenced the access result presented in Table 10. The logit regression coefficient estimates were translated into odds ratios to facilitate interpretation. Findings show that health insurance has tremendously improved the financial accessibility of its members to the modern health care system, particularly for women, children, and the poor. Access to care is determined by prepayment membership, patient age, pregnancy, patients' health status, distance to the health facility, and households' income group. Most important, prepayment members are 559 percent more likely to enter the modern health care system when sick compared to nonmembers. Health-related indicators significantly influenced health-seeking behavior, with children under 5 years of age being 92 percent more likely to report a visit than older patients. Also, pregnant women report 65 percent higher probability for seeking care, and the sick individuals who spent four and more days in bed were 96 percent more likely to go to a modern health care provider than people who were not that long in bed. People who live close to the health facility are significantly more likely to seek care (61 percent) than those who live farther away. Patients in the lowest income quartile are far less likely to seek care than those in the highest income quartile. This means that while the prepayment scheme has significantly increased access to health care for members including those who are poor, the impact at the level of the district in increased access to health care for the poor remains an issue. The solution is to find mechanisms to increase enrollment of the poor households in the prepayment schemes. Table 10: Logit regression results for probability of 14 at least one professional provider visit for members and nonmembers Probability of visit Explanatory Variable Reference Category Variable Odds Ratio S.E. Sign Prepayment members Nonmembers 6.59*** 0.263 0.000 Male patient Female patient 1.21 0.140 0.170 Patient age 0-5 years Patient age 6 years and older 1.92*** 0.158 0.000 Pregnant in past year No pregnancy in past year 1.65*** 0.248 0.043 Patient spent 4 and more days in bed Less than 4 days in bed 1.96*** 0.139 0.000 Less than 30 min. from hh cell to h-facility More than 30 min to h-facility 1.61*** 0.137 0.000 Household with 5 and more members Small HH size, less than 5 1.17 0.142 0.277 Household head attended school HH head, illiterate 0.91 0.141 0.519 Household with cattle No cattle 1.26 0.162 0.162 Household with radio No radio 1.33 0.143 0.050 First Income Quartile Fourth Income Quartile 0.18*** 0.230 0.000 Second Income Quartile 0.44*** 0.174 0.000 Third Income Quartile 0.46*** 0.172 0.000 Ancillary statistics: N 1,502 - 2 Log likelihood 1434.941 Goodness fit (chi-squared test) 211.744 Degree of freedom 13 R Square 19.3% Note: Z-tests were performed to test the probability of enrollment for each characteristic in a logit model. *** Significant at 1 percent level of significance. ** 5 percent level of significance. D. MODEL 3: AVERAGE OUT-OF-POCKET HEALTH EXPENDITURES PER EPISODE OF ILLNESS14 Based on detailed provider and insurance data, the evaluation of the prepayment pilot phase has shown that health insurance has a substantial impact on members' financial contribution to health care, as well as on providers' cost and financial situation15. Findings in PHR report No. 61 have shown that the PPS members' annual per capita contribution to the modern health care system are up to five times higher than those of nonmembers16. However, the out-of-pocket expenditures per episode of illness are significantly lower for members as compared to nonmembers. This means, that by paying insurance, scheme members face lower fees at the time of illness and have greater access at times of need. Further, in Rwanda, with very low per capita use rates, the higher utilization by members should not be interpreted as an effect of moral hazard, but rather as improved access to essential basic health services. 14Professional care = public and church-owned health centers, district hospitals, and dispensaries. This excludes sick individuals who sought care at traditional healers and others (e.g., drug vendors, pharmacy). 15See PHR Technical Report No. 61: Utilization, Cost, and Financing of District Health Services in Rwanda. 16See PHR TR No. 61, Table 3.14: Members in Byumba reported 580 francs annual per capita contributions to health centers, whereas this amount is only 104 francs per year per capita for nonmembers, due to their lower health service use. 15 The third model in this paper documents insured and uninsured patients' out-of-pocket health expenditures per episode of illness. This information is first shown for all sick, independent of their care- seeking behavior, and second for those who reported at least one health facility visit. Table 11 presents total health-related out-of-pocket expenditures for each of the different health-related services paid during an episode of illness, which includes care received before visiting a provider, out-of- pocket spending at the first professional visit, and out-of-pocket spending for other professional providers. This total health expenditure information is shown for the insured and uninsured sick in pilot districts, and further broken down for each group by "with and without a professional provider visit." Within these categories, the different health expenditures are shown by patients' income quartiles. It is found that per episode of illness, sick members pay on average 294 francs for all health related expenditures. This amount is higher for nonmembers in pilot districts (450 francs). The differences are even stronger when comparing only the sick insured and uninsured with a professional care visit. Sick members with professional visit pay 497 francs for the full episode of illness, whereas nonmembers' out- of-pocket health expenditures per episode of illness with professional visit amount to 1,987 francs in pilot districts. Insurance membership has significantly decreased out-of-pocket spending for a full episode of illness for sick members with and without visit, and at the same time has substantially improved members' access to the modern health care system. Also, health insurance has changed patients' health care seeking behavior. A comparison of members' and nonmembers' average out-of-pocket spending for home and other care in Table 11 shows that the uninsured spend two and a half times more on home care and traditional remedies than the insured, who are more likely to seek quality care in the modern health system. Thus, prepayment schemes have not only reduced financial barriers in accessing better quality care and thus equity in accessing care, but insurance membership has also shifted the demand for health care toward more efficient care. Members who seek care pay 100 francs copayment per episode of illness at the health center. Sick individuals from richer households spend more on home care and on professional care compared to lower income groups, and this holds for the insured and the uninsured. The fact that the richer insured pay up to 10 times more per episode of illness than the poorest CBHI members supports the assumption made in the previous model. That is, prepayment schemes favor cross-subsidies from richer to poorer members, if there is a uniform premium per family, if poorer families number more individuals than richer families, and if members care-seeking behavior is independent of their income status. So far this argument holds. However, the richer insured may still pay more because they are willing to pay additional amounts for care not covered by the insurance scheme such as drugs excluded from the MOH essential drug list. Table 11: Average health expenditure (RWF17) per sick individual without or with visit to a professional provider, by health insurance status and expenditure quartile, in pilot districts Out-of-pocket Income Pilot Districts spending quartile Nonmembers Members RWF No prof. 1+ prof. Total No prof. 1+ prof. Total visit visit visit visit 1 90 85 90 276 0 165 17Nominal Exchange Rate: USD 1$ = RWF 370 (official period average in 2000). Source: National Bank of Rwanda. 16 Home & other 2 124 178 131 68 23 53 care 3 171 230 180 28 20 24 4 277 322 288 79 133 108 Total 160 245 172 93 52 74 1 11 693 49 10 112 51 1st professional 2 10 1,356 180 20 178 75 provider 3 43 1,445 246 9 220 112 4 36 2,228 600 105 966 572 Total 24 1,693 269 32 418 207 1 0 262 15 0 0 0 Other 2 0 27 3 0 0 0 professional 3 1 42 7 0 1 1 providers 4 9 22 13 0 91 50 Total 2 50 9 0 28 12 1 101 1,041 154 286 112 216 Total illness 2 134 1,561 314 88 201 128 related 3 215 1,717 433 37 242 137 expenditure 4 322 2,573 901 184 1,190 730 Total 186 1,987 450 126 497 294 The following log-linear regression estimates sick individuals' average health expenditures for the insured and uninsured in pilot districts. Findings show that prepayment schemes have significantly decreased out-of-pocket spending for the entire episode of illness for sick individuals who are members. Individuals' out-of-pocket health expenditures are positively influenced by the patient's gender, with men paying more than women, by household size, and by use of professional care. Patients classified in the three lower income quartiles report significantly lower out-of-pocket spending for an episode of illness, with or without a visit, than those in fourth income quartiles. Also, out-of-pocket spending per episode of illness is significantly influenced negatively if patients live in the health center's vicinity and if they own cattle (which can be interpreted as a sign of wealth). Table 12: Log-linear regression results: Estimated total health related expenditures per episode of illness for sick individuals with and without visit Explanatory Variable Reference Category Variable Out-of-pocket all sick Coeff. S.E. P>t Sick insured members Sick uninsured in pilot districts -0.604*** 0.141 0.000 Male patient Female patient 0.056*** 0.061 0.000 Patient age 0-5 years Patient age 6 years and older -0.006 0.075 0.362 Pregnant in past year No pregnancy in past year -0.227 0.120 0.933 Patient spent 4 and more days in bed Less than 4 days in bed 0.228 0.060 0.057 Less than 30 min. from hh cell to h-facility More than 30 min to h-facility -0.125*** 0.062 0.000 17 Household with 5 and more members Small HH size, less than 5 0.111*** 0.063 0.045 Household head attended school HH head, illiterate 0.262 0.061 0.075 Household with cattle No cattle -0.090*** 0.075 0.000 Household with radio No radio 0.258 0.067 0.234 First Income Quartile Fourth Income Quartile -0.544*** 0.091 0.000 Second Income Quartile -0.290*** 0.085 0.000 Third Income Quartile -0.183*** 0.085 0.001 All sick w/ 1+ professional care visit All sick without visit 1.645*** 0.077 0.030 (Constant) 1.048*** 0.101 0.000 Ancillary Statistics: N 1,596 F 52.686 Degree of freedom (14 1,582) Prob > F 0.000 R square 0.318 Note: Includes total health related out-of-pocket spending for sick with and without visit. T-tests were performed to test significant difference for each characteristic. *** Significant at 1 percent level of significance; ** 5 percent level of significance. Table 13 presents log-linear regression results on out-of-pocket spending per episode of illness for the 336 patients who reported a professional health care visit. As can be expected, CBHI membership has significantly decreased health care costs for the sick members with visit. Patients classified in the first income quartile report significantly lower out-of-pocket spending per episode of illness if they had a visit, than patients in fourth income quartile. However, those in the second and third quartile report lower out of pocket spending per episode but are not statistically significant. Male patients with visits spend significantly more per episode of illness than female patients. Table 13: Log-linear regression results: Estimated out-of-pocket health expenditures per episode of illness for sick individuals, who reported a professional visit in past 2 weeks Log of total health expenditures for with visit Out-of-pocket w/visit Explanatory Variable Reference Category Variable Coeff. S.E. P>t Sick insured members Sick uninsured in pilot districts -0.808*** 0.145 0.000 Male patient Female patient 0.274*** 0.099 0.006 Patient age 0-5 years Patient age 6 years and older 0.068 0.107 0.523 Pregnant in past year No pregnancy in past year -0.127 0.172 0.460 Patient spent 4 and more days in bed Less than 4 days in bed 0.088 0.101 0.386 Less than 30 min. from hh cell to h-facilityMore than 30 min to h-facility -0.121 0.098 0.215 Household with 5 and more members Small HH size, less than 5 0.061 0.103 0.556 18 Household head attended school HH head, illiterate 0.133 0.106 0.213 Household with cattle No cattle -0.068 0.112 0.547 Household with radio No radio 0.007 0.103 0.948 First Income Quartile Fourth Income Quartile -0.464*** 0.181 0.011 Second Income Quartile -0.152 0.124 0.221 Third Income Quartile -0.143 0.124 0.250 (Constant) 2.850*** 0.155 0.000 Ancillary Statistics: N 336 F 5.286 Degree of freedom (13 323) Prob > F 0.000 R square 0.175 Note: Includes total health related out-of-pocket spending for the full episode of illness for those who were sick and went to seek professional care. T-tests were performed to test significant difference for each characteristic. *** Significant at 1 percent level of significance. Combining these findings with the results presented in the first and second model show that community- based health insurance in Rwanda has been successfully used as a tool to improve financial accessibility to care for the poor who enroll in the scheme while, at the same time, their out-of-pocket health care expenditures could be reduced per episode of illness. 19 V. DISCUSSION AND CONCLUSION Findings confirm that prepayment schemes in Rwanda successfully target the rural poor, with their members having an annual average per capita income of approximately US$100. Logit regression results have shown that the probability of enrolling in prepayment schemes is equal among all income groups and is determined by factors such as households' distance to the health center, exposure to radio awareness campaigns, age, gender, and education status of the household head. Other factors have influenced enrollment such as precautionary behavior for one's family's health and foresight and the possibility of seeing the insurance premium as a trustworthy investment. Survey findings also allude to the fact that social capital such as trust and sentiments of ownership are important determinants of participation in prepayment schemes. Prepayment schemes are forms of solidarity and social cohesion that help strengthen the link between the poor and the health facilities where the local people would like to seek care. Health insurance has significantly improved equity in financial accessibility to care for members by increasing their probability of a visit while at the same time the financial burden per episode of illness could be reduced. This argument holds for all income groups in the insurance pool, although richer members still pay up to 10 times more out-of-pocket, supporting the argument of possible cross-subsidies to poorer patients. Health insurance has helped to eliminate financial barriers in access to care for the poorest among the insured members, whereas the uninsured in lowest expenditure quartiles continue to report significantly worse access to care than the richer insureds. In addition to improved access, faster access to care of the insured patients has contributed to a shift of demand for care from the traditional to the modern health sector and has improved the efficient use of limited medical resources such as drugs and staff in district health facilities. The analysis of the financial impact of prepayment schemes on health care providers has shown that community-managed health plans, combined with provider capitation payment, have built up expertise and capacity among insurance members about their rights and obligations and, as a result, have empowered consumers in discussions for better quality care with health center managers during the schemes' general assemblies. Although data collection during the pilot phase was extensive and included patient exit interview and focus group information as well as routine provider insurance and household data, it is too early to conclude that better access to care due to prepayment membership has caused members' health to improve. However, findings from different sources suggest that conclusion. Still, about 90 percent of the population in the three districts has not enrolled with prepayment schemes and continues to report dismal health care utilization patterns. Although the large majority of nonmembers interviewed in the household survey said they would like to become members, three fourths of them had serious doubts that they will have the 2,500 francs available to pay the annual fee for their family, raising concerns about how successful the schemes are in improving equity of access to care. During the pilot year, church groups have facilitated enrollment by financing membership for widows, orphans, indigents, and HIV-infected poor individuals. This targeted demand-side subsidy has contributed to a welfare gain if they benefited the indigents' insurance enrollment without decreasing benefits for patients who did not enroll. The church groups' experience with financing health insurance membership has caused prepayment to become a promising tool to subsidize targeted access to care for the vulnerable. 20 VI. REFERENCES HFS and PHR Technical Documents: Diop, François, P. Schneider and D. Butera. 2001a. Summary of Results: Prepayment Schemes in the Rwandan Districts of Kabgayi, Byumba, and Kabutare. Technical Report No. 59. Partnerships for Health Reform. Bethesda, MD. Diop, Francois, and P. Schneider. 2001b. Household Survey in 5 Rwandan Districts. Technical Report (forthcoming). Partnerships for Health Reform. Bethesda, MD. Schneider, Pia, F. Diop, C. Leighton. 2001a: Pilot Testing Prepayment for Health Services in Rwanda: Results and Recommendations for Policy Direction and Implementation. Technical Report No. 63. Partnerships for Health Reform. Bethesda, MD. Schneider, Pia, F. Diop, D. Maceira, and D. Butera. 2001b: Utilization, Cost and Financing of District Health Services in Rwanda. Technical Report No. 61. Partnerships for Health Reform. Bethesda, MD. Schneider, Pia., F. Diop, and S. Bucyana. 2000: Developing and Implementing Prepayment Schemes in Rwanda. Technical Report No. 45. Partnerships for Health Reform. Bethesda, MD. Other Sources: Bennett, S., A. Creese, and R. Monasch. (1998). Health insurance schemes for people outside formal sector employment. WHO Current Concerns WHO/ARA/CC/98.1. Geneva: World Health Organization, Division of Analysis, Research and Assessment. Colletta, Nat J., and M. L. Cullen. (2000). Violent conflict and the transformation of social capital. Lessons from Cambodia, Rwanda, Guatemala, and Somalia. The World Bank. Washington, D.C. Deaton, A. (1997). The analysis of household surveys. A microeconomic approach to development policy. The World Bank. Washington, D.C. Diop, F., A. Yazbeck, and R. Bitran. (1995). The impact of alternative cost recovery schemes on access and equity in Niger. Health Policy and Planning; 10(3):223-240. Manning, W. G., J. P. Newhouse, N. Duan, E. B. Keeler, A. Leibowitz, and S. M. Marquis. (1987). Health insurance and the demand for medical care: evidence from a randomized experiment. American Economic Review 77(3): 251-77. Mathiyazhagan, K. (1998). Willingness to pay for rural health insurance through community participation in India. Int. J Health Plann. Mgmt. 13: 47-67. Muller, C. (1997). Transient seasonal and chronic poverty of peasants: Evidence form Rwanda. WPS/97- 8. Centre for the Study of African Economics. Institute of Economics and Statistics. University of Oxford. Oxford. Mukherjee, Chandan, H. White, and M. Wuyts. (1998). Econometrics and data analysis for developing countries. Routledge. London. 21 Republic of Rwanda, Office National de la Population. (1999). "ETUDE SUR LES CONNAISSANCES ET ATTITUDES SUR LE SYSTEME DE PRE-PAIEMENT ET D'ASSURANCE--MALADIE POUR LA SANTE AU RWANDA (Focus Group Survey on Prepayment Schemes in Rwanda)." In collaboration with Partnerships for Health Reform. Kigali, Rwanda. Schoemaker, P.J.H., (1982). The expected utility model: Its variants, purposes, evidence and limitations. Journal of Economic Literature. Vol. XX June: 529-563. The World Bank. 1998. Rwanda Poverty Note. Rebuilding an Equitable Society: Poverty and Poverty Reduction after the Genocide. Macroeconomics 3. Africa Region. Washington. Wilkinson, Richard G., (1996). Unhealthy societies: the afflictions of inequality. Routledge, London. Yip, Winnie, and Peter Berman. (2001). Targeted health insurance in a low-income country and its impact on access and equity in access: Egypt's school health insurance. Health Econ. 10: 207-220. 22 About this series... This series is produced by the Health, Nutrition, and Population Family (HNP) of the World Bank's Human Development Network. The papers in this series aim to provide a vehicle for publishing preliminary and unpolished results on HNP topics to encourage discussion and debate. The findings, interpretations, and conclusions expressed in this paper are entirely those of the author(s) and should not be attributed in any manner to the World Bank, to its affiliated organizations or to members of its Board of Executive Directors or the countries they represent. Citation and the use of material presented in this series should take into account this provisional character. For free copies of papers in this series please contact the individual authors whose name appears on the paper. Enquiries about the series and submissions should be made directly to the Editor in Chief Alexander S. Preker (apreker@worldbank.org) or HNP Advisory Service (healthpop@worldbank.org, tel 202 473-2256, fax 202 522-3234). For more information, see also www.worldbank.org/hnppublications. THE WORLD BANK 1818 H Street, NW Washington, DC USA 20433 Telephone: 202 477 1234 Facsimile: 202 477 6391 Internet: www.worldbank.org E-mail: feedback@worldbank.org