WPS8487 Policy Research Working Paper 8487 Transport Connectivity, Medical Supplies, and People’s Health Care Access Evidence from Madagascar Atsushi Iimi Voahirana Hanitriniala Rajoela Transport and Digital Development Practice June 2018 Policy Research Working Paper 8487 Abstract Health care access is a challenge in rural areas in Africa. people’s demand for health care services is affected not On the demand side, rural people are often poor, and only by local transport connectivity, but also availability transport connectivity is typically bad in rural and of medical supplies at the health facility level, which is remote areas. Because of limited transport connectivity, also determined by primary and secondary road network the quality of health care services provided is also often connectivity. This in turn further suppresses people’s compromised. In Madagascar, the poor condition of the demand in rural areas. The results also indicate that it is road network has long hampered the sustainability of the important to ensure financial affordability among the poor, medical supply chain in rural areas. The paper shows that which is found to be one of the most crucial constraints. This paper is a product of the Transport and Digital Development Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/research. The authors may be contacted at aiimi@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Transport Connectivity, Medical Supplies, and People’s Health Care Access: Evidence from Madagascar Atsushi Iimi ¶ 1, Voahirana Hanitriniala Rajoela 2 1 Transport & Digital Development Global Practice, The World Bank Group 2 Health, Nutrition & Population Global Practice, The World Bank Group Key words: rural roads; health care access; transport connectivity; three-stage least squares estimation. JEL classification: C25, C26, I15, Q42, R40. ¶ Corresponding author. -2- I. Introduction Health care access is an important challenge in rural areas, particularly in developing countries. It is generally difficult for people who live in remote areas to access good-quality health services (e.g., Gamm et al. 2003; Institute of Medicine 2005; Bourke et al. 2012). Even in the United States, rural residents of Washington State were less likely to visit the emergency department than their urban counterparts (Lishner et al. 2000). According to Laditka et al. (2009), rural residents are more likely to be hospitalized for ambulatory care sensitive conditions in the United States. This could be prevented if people had better access to primary care, but because they do not, their condition tends to deteriorate. Among others, transport connectivity is a particularly important constraint in rural areas. Rural dwellers’ travel distance to health facilities is much longer than that of urban dwellers (Gamm et al. 2003). In Zambia, about 75 percent of urban residents, but only 30 percent of rural residents, live within 2 kilometers of the nearest health facility (Zambia Central Statistical Office 2012). Remoteness and unavailability of transport services increase costs further for rural residents, limiting them from accessing health care services (e.g., Klemick et al. 2009). Limited transport connectivity not only reduces people’s accessibility, but also compromises the quality of health care services provided. Bronstein et al. (1997) show that rural health services are cheaper but that fewer ancillary services are provided. Sufficient doctors and nurses may not be available in rural areas as well. This may be a more generic disadvantage of rural areas or remoteness. For whatever reason, doctors are less available in rural areas. In the United States, only 9 percent of physicians practice in rural areas where about 20 percent of the total population lives (Rosenblatt and Hart 2000). As a result, rural physicians often have to see more patients than urban doctors do, therefore, potentially compromising the level of services they provide. Medical supplies may also be compromised in rural and remote areas. How to ensure supply chain sustainability is an important research issue (e.g., Uysal and Tosun, 2014). -3- In particular in Africa, these constraints often tend to be severe. In the region, about 70 percent of the total rural population do not have access to any all-weather road (Roberts et al., 2006). In Liberia, about 40 percent of rural residents must spend more than 2 hours to walk to the nearest health facility (Kruk et al. 2010). Moreover, the road condition is generally poor in rural Africa. In Uganda, 79 percent of paved roads are in good condition, but 75 percent of unpaved roads are in poor condition. In Kenya, about 40 percent of tertiary roads are in bad condition (World Bank, 2016). Demand-side constraints further aggravate rural people’s health access already limited by the above-mentioned supply-side difficulties. In Africa, rural residents are generally much poorer than urban residents. Thus, holding everything else constant, the rural demand for health services is lower in rural areas. Jovanovic et al. (2003) show that subscription to health insurance is lower in rural areas. Because of the lack of proper insurance, rural residents are less likely to use emergency medical services (Fan et al., 2011). This is a particularly important concern in Africa where poverty is concentrated in rural areas. Many studies are supportive of the importance of health insurance to improve people’s health care access in developing counties (see, e.g., Spaan et al., 2012). This means that financing is a critical issue for poor people. This paper casts light on the nexus of transport connectivity, medical supply sustainability and people’s health care service access. As discussed, these factors are interdependent on one another. To estimate the system of equations and solve the endogeneity problem, the paper relies on the three-stage least squares (3SLS) model (Zellner and Theil, 1962). One equation examines people’s demand for health care services, and another represents the quality of the services determined by transport connectivity and other factors. The data come from Madagascar where rural transport connectivity is extremely limited, because of low road density as well as the poor condition of the rural road network. Detailed health facility databases are combined with spatial road condition data. -4- The following sections are organized as follows: Section II describes the data and methodologies. Section III presents the main estimation results and discusses policy implications. Then, Section IV concludes. II. Methodology and data To examine people’s demand for health care services, two particular impacts of transport connectivity are considered. On the demand side, health access is considered to be affected by the degree of transport impediment between health facilities and households. The less connected, the smaller number of patients. On the supply side, the quality of health care services offered is also dependent on transport connectivity. If a health facility is not well connected, medical supplies may not be sustained. Thus, people cannot expect quality services, which would in turn affect people’s demand. To assess both aspects simultaneously, the following system of equations is considered: (1) (2) Note that our unit of observation is the health facility. In Madagascar, more than 3,000 health facilities, including hospitals and basic health centers (CSB), exist all over the country (Figure 1). The empirical analysis will only use data for about 900 facilities in eight regions (Alaotra- Mangoro, Analamanga, Bongolava, Itasy, Vakinankaratra, Amoron'i mania, Vatovavy Fitovinany, and Haute matsiatra), where detailed facility-level data are available. -5- Figure 1. Health facility network in Madagascar Source: Ministry of Health, Madagascar The first equation aims at capturing the direct transport impacts on people’s demand, for which the dependent variable is the share of patients out of the total local population. N is the number of patients who newly visited and were treated at each facility. To control for the size effect, this is divided by the total number of local population (POP) in villages or fokontany covered by each facility. T denotes a set of connectivity measurements. In Madagascar, many rural people are disconnected from the road network and health facilities: About half of fokontany where people live are more than 10 km away from the nearest basic health center. Even by simple statistics, it is clear that more people would likely visit health centers if they are closer to them (Figure 2). For Equation (1), local connectivity is measured by two variables: (i) weighted average distance between a fokontany and the nearest basic health center, and (ii) rural access index (RAI), which is available at the district level. The latter is a traditional transport indicator (Figure 3). -6- Figure 2. People’s access and distance to health facilities Source: World Bank estimate. Figure 3. Rural Access Index Source: World Bank estimate. X includes other explanatory variables that potentially affect the demand. As discussed above, the level of household affordability is a matter of particular concern. The district-level poverty data (POV) are used to control for this. The latest available official poverty data are at the regional level. To increase granularity, the poverty equation is calibrated using the latest household survey in 2010, generating the district-level poverty rates (Figure 4). While it is not official, the estimates are broadly consistent with the general understanding of the geographic distribution of poverty in the country (Figure 5). -7- Figure 4. District-level poverty rate Figure 5. Distribution of poverty estimates Source: World Bank estimate. Source: World Bank estimate. Finally, our main hypothesis in this paper is that the people’s health care demand depends on the quality of health services, denoted by M. Of course, there are a number of aspects in this regard. For data availability and empirical tractability purposes, the following analysis uses the average availability of medical supplies at each health facility: ∑ (3) where m is the availability rate of medicine j at each facility. It is defined by the number of days when the stock is sufficient, divided by 365 days. In Madagascar, detailed supply and stock data are available for 15 medicines. Because of unreliable transport connectivity, particularly in remote areas during the rainy season, availability of medical supplies varies significantly from one location to another (Figure 6). Even by simple statistics, it is clear that more people would likely more often visit health facilities (Figure 7). Needless to say, it is determined by not only medicine availability but also other factors. That is why we have other variables, such T and X, in Equation (1). -8- Figure 6. Availability of medical supplies (%) Source: Ministry of Health, Madagascar. Figure 7. People’s access to health care services and availability of medical supplies Source: World Bank estimate. Importantly, the availability of medical supplies is crucially dependent on transport connectivity in rural Madagascar. About 25 percent of the health facilities are disconnected more than 5 km (approximately 1 hour walking distance) from the official road network. It is particularly challenging to sustain it in rural and remote areas, especially during the rainy season when rural roads are often flooded and impassable. There is broad negative correlation between people’s tendency to access health care services and remoteness of basic health centers (Figure 8). The latter is measured by distance from the district capital, which is an important part of the medical supply network in Madagascar. -9- Figure 8. People’s access to health care services and remoteness of health centers Source: World Bank estimate. Needless to say, there are other factors that potentially affect the availability or unavailability of medical supplies, as modeled in Equation (2). For T2, two variables are used, which represent connectivity at the higher level of transport network than T1: (i) distance from the road network and (ii) average travel time from each health center to the district capital. All the variables that are used in our empirical analysis are included in the summary statistics in Table 1. To solve the endogeneity problem between the two equations and improve efficiency in estimation, the paper relies on the three-stage least squares (3SLS) model (Zellner and Theil, 1962). To instrument the endogenous variable M, three variables are used. First, the straight- line distance from each facility to Antananarivo, the capital city, is used. This is an artificial measurement, as often referred to as the “as-the-crow-flies” distance, and is unlikely to be directly related to people’s access to health care services. It should really depend on the actual road alignment and conditions on the ground. On the other hand, this is considered to be somewhat relevant to the availability of medical supplies because they are generally distributed from the center of the economy, Antananarivo. Following Dinkelman (2011), geographic conditions are also used. Land gradient (denoted by SLOP) and elevation (ELEV) at each locality may affect the technical feasibility of developing road infrastructure, which is likely to impact the medicine availability. Of course, these geographic conditions may also possibly influence people’s accessibility to health - 10 - facilities, violating the validity of the instruments. This will ex post be examined by the conventional test of overidentifying restrictions. Table 1. Summary statistics Variable Abb. Obs Mean Std. Dev. Min Max Number of new patients who visited each CSB in 2016 N 892 2378 2510 0 22563 Population in fokotany served by each (CSB) POP 892 11220 15135 787 298597 Share of local people who visited CSB in 2016 (%) N/POP 892 28.5 45.9 0.0 1039.8 Distance from fokontany to CSB (km) KM_FKT 892 6.3 4.3 0.0 65.4 Rural Access index (%) RAI 892 14.0 14.7 0.0 100.0 Poverty rate at the district level (%) POV 892 76.6 16.4 16.3 97.6 Distance to the official road network (km) KM_RD 892 3.3 5.6 0.0 45.5 Travel time from CSB to District capital (hours) TIME_DIST 892 0.51 0.52 0.00 4.41 Medicine availability (% of days a year): Aminophilline m1 892 93.9 17.9 0.0 100 Amoxicilline m2 892 96.4 11.3 8.6 100 Benzyl Pénicilline m3 892 94.3 15.2 0.0 100 Captopril m4 892 77.9 33.3 0.0 100 Chlorphéniramine m5 892 95.8 11.6 0.0 100 Cotrimoxazole m6 892 96.9 10.2 0.0 100 Fer acide folique m7 892 93.1 15.6 0.0 100 Gentamicine m8 892 95.5 13.4 0.0 100 Hydrochlorothiazide m9 892 64.0 39.8 0.0 100 Ibuprofène m10 892 96.1 11.7 0.0 100 Métoclopramide m11 892 86.1 24.1 0.0 100 Métronidazole m12 892 97.2 9.1 0.0 100 Paracétamol m13 892 97.1 9.3 0.0 100 Phénobarbital m14 892 91.7 18.1 0.0 100 SRO m15 892 87.9 23.5 0.0 100 Average availability of all medicines M 892 90.9 11.2 13.3 100 Instrumental variables: Straight-line distance to Antananarivo (km) KM_TANA 892 164.0 108.6 0.2 394.3 Slope at the location of each CSB (degree) SLOP 892 1.45 1.41 0.02 11.22 Elevation at the location of each CSB (m) ELEV 892 1034.7 485.2 0.0 2398.0 III. Main estimation results Although the results may be biased and inefficient, the ordinary least squares (OLS) regression is first performed: The results are shown in Table 2. The estimated coefficients are generally consistent to our prior expectation. Clearly, transport connectivity is likely to - 11 - matter to both people’s local accessibility to health facilities and availability of medical supplies. The coefficient of the medical supply availability in question is insignificant in both OLS and instrumental variable (IV) estimation. The coefficient is estimated at 0.007 and - 2.333, respectively. Again, the results may be biased and the estimation efficiency can be improved by estimating the two equations together. Table 2. OLS and IV estimation results Dependent variable: N/POP N/POP M Estimation method: OLS IV OLS Coef. Std.Err. Coef. Std.Err. Coef. Std.Err. KM_FKT -1.184 (0.452) *** -1.406 (0.691) ** RAI -0.617 (0.230) *** -0.440 (0.212) ** POV -0.765 (0.278) *** -0.783 (0.297) ** -0.037 (0.019) ** M 0.007 (0.072) -2.333 (2.577) KM_RD -0.195 (0.130) TIME_DIST -1.757 (0.817) ** Constant 102.504 (26.476) *** 315.673 (252.608) 95.321 (1.399) *** Obs. 892 892 892 R-squared 0.063 … 0.026 F statistic 2.39 4.78 Wald chi2 9.03 Note: Robust standard errors are shown in parentheses. *, ** and *** indicate statistical significance at the 10, 5 and 1 percent level, respectively. The 3SLS model is applied (Table 3). Two specifications are considered: with and without region-specific dummy variables. The results are broadly similar; however, our preferred model is the one without regional dummy variables, shown in the first column of the table. The Hansen-Sargan test statistic for overidentifying restrictions is estimated at 6.088, which is insignificant, suggesting the validity of our instruments. With regional dummy variables, the test statistic is 37.645, which is significant at the 1 percent level. In this specification, our instruments are found invalid. Based on the result without regional dummies, the largest determinant in terms of elasticity is poverty, of which the coefficient is estimated at -0.696 with a standard error of 0.129. The implied elasticity is -1.87 with a standard error of 0.34, which means that if poverty is - 12 - reduced by 1 percent, the demand for health care service would increase by 1.87 percent (Figure 9). The availability of medical supplies also has a significant impact. The coefficient is 1.995, which can be translated into an elasticity of 0.42 with a standard error of 0.24.1 Thus, if the availability of one particular medicine improves by 1 percent, the health care demand would increase by 0.24 percent. An unexpected result is that rural accessibility has a negative coefficient. Rural accessibility is highly correlated with poverty as usual, and it seems that the poverty aspect dominates the possible impact of transport disconnectedness. Local transport connectivity (KM_FKT) has a negative and significant coefficient, which is consistent with our prior expectation. In the medicine supply equation, the results show that the availability of medicines, M, decreases with distance to the road network and connectivity to the district center, which is measured by travel time. Both are statistically significant, indicating that the primary and secondary road network connecting health facilities and district capitals is crucial to ensure the medical supply chain in Madagascar. Basic health centers should be better connected to the official road network and the nearest district centers where medical supplies are generally distributed. This impact would in turn affect people’s demand for health care services. Recall that in the demand equation, the coefficient of M is significantly positive. Therefore, if a health facility is not well connected to the district center, its medical supplies are more likely to be insufficient, and as a result, people are less motivated to visit that health center. 1 It is divided by 15 since our dependent variable is the average of stock availability of 15 medicines. - 13 - Table 3. Three-stage least squares estimation result Without regional dummies With regional dummies Coef. Std.Err. Coef. Std.Err. N/POP equation: KM_FKT -1.002 (0.364) *** -0.849 (0.361) ** RAI -0.676 (0.155) *** -0.896 (0.153) *** POV -0.696 (0.129) *** -0.626 (0.200) *** M 1.995 (1.117) * 2.411 (0.648) *** Constant -83.770 (102.575) -110.666 (65.623) * M equation: KM_RD -0.217 (0.063) *** -0.139 (0.058) ** TIME_DIST -1.644 (0.678) ** -1.499 (0.632) ** POV -0.037 (0.023) -0.127 (0.028) *** Constant 95.318 (1.767) *** 98.014 (1.819) *** Obs. 892 892 R-squared: N/POP -0.169 -0.228 M 0.026 0.126 Overidentification statistic: 6.088 37.645 *** Note: Robust standard errors are shown in parentheses. *, ** and *** indicate statistical significance at the 10, 5 and 1 percent level, respectively. Figure 9. Implied elasticity of people’s health care access Note: * Divided by the number of medical supplies, M, for presentation purposes. One might be concerned about the validity of our instruments, although it is confirmed by the traditional overidentification test as discussed above. Particularly, the two geographic variables (i.e., slope and elevation at each health facility) may also have some implication for people’s accessibility to health facilities: People might have more difficulty visiting a health facility in a mountainous area. To check this possibility, the 3SLS is performed with these - 14 - two variables included as explanatory variables in X (Table 4). It was found that they have little explanatory power in the equations. Thus, they are not considered to belong to the explanatory equations. The estimation results are similar to the above: People’s health care service demand declines with the distance between villages and health facilities as well as poverty prevalence. The availability of medical supplies also affects people’s demand. To improve medicine availability, health facilities need to be connected to the road network and the district capital, which has an important role in the medical supply distribution system. Table 4. Three-stage least squares estimation results with two instrumental variables included Without regional dummies With regional dummies Coef. Std.Err. Coef. Std.Err. N/POP equation: KM_FKT -1.010 (0.364) *** -0.880 (0.360) ** RAI -0.671 (0.157) *** -0.891 (0.154) *** POV -0.675 (0.132) *** -0.702 (0.241) *** AVG_M 2.015 (1.134) * 1.719 (1.107) SLOP -1.390 (1.238) -1.021 (1.198) ELEV 0.0005 (0.0037) 0.018 (0.014) Constant -85.697 (104.125) -66.532 (125.669) M equation: KM_RD -0.221 (0.064) *** -0.169 (0.062) *** TIME_DIST -1.739 (0.700) ** -1.693 (0.671) ** POV -0.043 (0.024) * -0.130 (0.028) *** SLOP 0.234 (0.276) 0.089 (0.261) ELEV -0.0006 (0.0008) -0.011 (0.002) *** Constant 96.057 (2.215) *** 112.046 (3.030) *** Obs. 892 892 R-squared: N/POP -0.172 -0.071 M 0.027 0.157 Overidentification statistic: 4.286 5.993 Note: Robust standard errors are shown in parentheses. *, ** and *** indicate statistical significance at the 10, 5 and 1 percent level, respectively. - 15 - IV. Conclusion Health care access is an important challenge in rural areas, particularly in developing countries. There are various reasons on both the demand and supply sides. On the demand side, rural people tend to be poorer than urban residents for whatever reason. In rural areas, transport connectivity is also more limited, which constrains people’s demand for health care services but also compromises the quality of health care services provided. In Madagascar, the poor condition of the road network has long hampered the sustainability of the medical supply chain to rural and remote areas, where rural roads are often impassable during the rainy season. Without sufficient medical supplies, people’s demand to visit health facilities would be suppressed. The paper addresses this nexus of transport connectivity, medical supplies and people’s health demand, by estimating a system of equations. The results indicate that transport connectivity is essential on both sides: While proximity to health facilities would increase the demand for health care services, the medical supply availability is also affected by transport connectivity, especially to the official road network and the nearest district center. The supply availability in turn influences people’s demand for health care services. Therefore, the poor transport connectivity suppresses the demand further in rural areas. Policy implications are straightforward: It is important to develop and maintain transport infrastructure to better connect all health facilities and sustain necessary medical supplies. The estimation results also highlight the importance of not only infrastructure but also institutional aspects. Poverty is one of the most significant factors affecting people’s health demand. It is important to ensure financial affordability in the health sector, possibly providing targeted subsidies, expanding insurance mechanisms and/or implementing free health programs, especially in rural areas. - 16 - References Bourke, Lisa, John Humphreys, John Wakerman, and Judy Taylor. 2012. “Understanding Rural and Remote Health: A Framework for Analysis in Australia.” Health and Place 18: 496–503. Bronstein, J.M., V.A. Johnson, and C.A. Fargason, Jr. 1997. “How Rural Physicians Compare on Cost and Quality Measures for Medicaid Ambulatory Care Episodes.” Journal of Rural Health 13 (2): 126–35. Central Statistical Office, Zambia. 2012. Living Conditions Monitoring Survey Report 2006 and 2010. Zambia: Central Statistical Office. Fan, Lin, Manish Shah, Peter Veazie, and Bruce Friedman. 2011. “Factors Associated with Emergency Department Use among the Rural Elderly.” Journal of Rural Health 27 (1): 39–49. Gamm, Larry, Graciela Castillo, and Stephanie Pittman. 2003. “Access to Quality Health Services in Rural Areas: Primary Care: A Literature Review.” In Rural Healthy People 2010, edited by Larry Gamm, Linnae Hutchison, Betty Dabney, and Alicia Dorsey. http://sph.tamhsc.edu/centers/rhp2010/publications.htm Institute of Medicine. 2005. “Quality Through Collaboration: The Future of Rural Health.” http://www.iom.edu/Reports/2004/Quality-Through-Collaboration-The-Future-of-Rural- Health.aspx#sthash.nqB3dnEa.dpuf Jovanovic, Zorana, Chyongchiou Lin, and Chung-Chou Chang. 2003. “Uninsured vs. Insured Population: Variations among Nonelderly Americans.” Journal of Health and Social Policy 17 (3): 71–85. Klemick, Heather, Kenneth Leonard, and Melkiory Masatu. 2009. “Defining Access to Health Care: Evidence on the Importance of Quality and Distance in Rural Tanzania.” American Journal of Agricultural Economics 91 (2): 347–58. Kruk, Margaret, Peter Rockers, Elizabeth Williams, Tornorlah Varpilah, Rose Macauley, Geetor Saydee, and Sandro Galea. 2010. “Availability of Essential Health Services in Post-Conflict Liberia. Bulletin of the World Health Organization 88: 527–34. Laditka, James, Sarah Laditka, and Janice Probst. 2009. “Health Care Access in Rural Areas: Evidence that Hospitalization for Ambulatory Care-Sensitive Conditions in the United States May Increase with the Level of Rurality.” Health and Place 15: 761–70. Lishner, Denise, Roger Rosenblatt, Laura-Mae Baldwin, and Gary Hart. 2000. “Emergency Department Use by the Rural Elderly.” Journal of Emergency Medicine 18 (3): 289–97. Roberts, Peter, K. C. Shyam, and Cordula Rastogi. 2006. Rural Access Index: A key development indicator. Transport Papers TP-10. The World Bank Group, Washington, DC. - 17 - Rosenblatt, Roger, and Gary Hart. 2000. “Physicians and Rural America.” Western Journal of Medicine 173 (5): 348–51. Spaan, Ernst, Judith Mathijssen, Noor Tromp, Florence McBain, Arthur ten Have, and Rob Baltussen. 2012. The impact of health insurance in Africa and Asia: A systematic review. Bulletin of the World Health Organization, Vol. 90(9), pp. 685-692. Uysal, Fahriye, and Omur Tosun. 2014. Selection of sustainable warehouse location in supply chain using the grey approach, International Journal of Information and Decision Sciences, Vol. 6(4), pp. 338-353. World Bank. 2016. Measuring Rural Access: Using New Technologies. The World Bank Group. Zellner, Arnold and H. Theil Zellner. 1962. Three-stage least squares: Simultaneous estimation of simultaneous equations. Econometrica, Vol. 30(1), pp. 54-78