Association between proximity to a health center and early childhood mortality in madagascar

Objective: To evaluate the association between proximity to a health center and early childhood mortality in Madagascar, and to assess the influence of household wealth, maternal educational attainment, and maternal health on the effects of distance. Methods: From birth records of subjects in the Demographic and Health Survey, we identified 12565 singleton births from January 2004 to August 2009. After excluding 220 births that lacked global positioning system information for exposure assessment, odds ratios (ORs) and their 95% confidence intervals (CIs) for neonatal mortality and infant mortality were estimated using multilevel logistic regression models, with 12345 subjects (level 1), nested within 584 village locations (level 2), and in turn nested within 22 regions (level 3). We additionally stratified the subjects by the birth order. We estimated predicted probabilities of each outcome by a three-level model including cross-level interactions between proximity to a health center and household wealth, maternal educational attainment, and maternal anemia. Results: Compared with those who lived >1.5-3.0 km from a health center, the risks for neonatal mortality and infant mortality tended to increase among those who lived further than 5.0 km from a health center; the adjusted ORs for neonatal mortality and infant mortality for those who lived >5.0-10.0 km away from a health center were 1.36 (95% CI: 0.92-2.01) and 1.42 (95% CI: 1.06-1.90), respectively. The positive associations were more pronounced among the second or later child. The distance effects were not modified by household wealth status, maternal educational attainment, or maternal health status. Conclusions: Our study suggests that distance from a health center is a risk factor for early childhood mortality (primarily, infant mortality) in Madagascar by using a large-scale nationally representative dataset. The accessibility to health care in remote areas would be a key factor to achieve better infant health. © 2012 Kashima et al. The study was based on 48464 birth records from the 2008–2009 Demographic and Health Survey (quatrième Enquête Démographique et de Santé réalisée à Madagascar: EDSMD-IV), which is a nationally representative sample survey. The 2008–2009 EDSMD-IV fieldwork was carried out from November 2008 to mid-August 2009 in selected areas in 22 regions in Madagascar (Figure 1). The EDSMD-IV samples were selected using a stratified two-stage design. First, each of the 22 regions was divided into 45 strata designated as urban, rural, and the city of Antananarivo. Then, 600 clusters were identified with a probability proportional to the size of the 22 regions. In the second stage, 32 households were randomly selected from each of the 600 clusters [9]. Each household was then surveyed. DHS, Demographic and Health Survey; EDSMD, quatrième Enquête Démographique et de Santé réalisée à Madagascar. The EDSMD-IV collected demographic, socioeconomic, and health information from each household using three questionnaires: the Household Questionnaire, the Women’s Questionnaire for ever-married women aged 15–49 years, and the Men’s Questionnaire for currently married men aged 15–54 years. In total, 18177 ever-married women were identified in the target area, and complete interviews were obtained with 17375 (96%). On the basis of the Women’s Questionnaire, 48464 birth records were created in the EDSMD-IV. In the present study, we extracted details of 12565 singleton births between January 2004 and August 2009 from the birth records, and excluded 220 births that lacked global positioning system (GPS) information for exposure assessment (Figure 2). We thus included 12345 births in the study. GPS, global positioning system; EDSMD, quatrième Enquête Démographique et de Santé réalisée à Madagascar. According to the Plan de Développement Secteur Santé 2007–2011, the health delivery system in Madagascar is composed of four types of health centers: basic health centers (Centre de Santé de Base: CSB) I and II; district hospitals (Centre Hospitalier de District: CHD) I and II; regional hospitals (Centre Hospitalier Regional: CHR); and university hospitals (Centres Hospitaliers Universitaires: CHU). Among these, only CSB I is not assigned a full-time medical doctor. These health facilities are composed of a four-step pyramidal referral system (Figure 3). The majority of these facilities is in imbalances in the distribution of medical staff [10] and concentrated in Antananarivo and other major cities. Indeed, around 28% of doctors serve 75% of the population living in the rural areas [10]. In these public health centers, the basic physical examination was provided without any charge. Also the Madagascar government introduced an equity funds system for achieving universal access to health care. The system funds the basic care or essential drugs to residents if they are certified as being poor by communities (fokontany and commune) [11]. Although these programs were provided by the government, Madagascar ranks the third lowest among the 44 African countries in term of hospital care availability [10]. We measured the distance from each household to the nearest health center to assess the subjects’ accessibility to health-care facilities. We included 3309 public health centers from all levels of the pyramidal system (Figure 3), excluding 98 centers because of a lack of location information (latitude and longitude), or absence of type of the health center. A location of these health centers was obtained based on the health care mapping software (Cart Sanitaire Madagascar) from the Service of Statistic Health of the Ministry of Health, which was updated in February 2011 and was supported by the Japan International Cooperation Agency. In the survey, location information of the village or settlement of the household was gathered by a trained interviewer using a GPS. Among the 594 GPS points available in the EDSMD-IV (version 2, updated in April 2011), 585 points could be assigned to the households in the present study. One GPS point represented 21 households on average (maximum, 50; minimum, 5; standard deviation, 8.7). To protect the privacy of respondents, offsets were employed by the EDSMD-IV. The offsets ranged from 0–2 km in urban areas and 0–5 km in rural areas. Furthermore, in rural areas, a 10 km offset was applied to every 100th village. All geographic variables were analyzed using the geographic information systems software ArcGIS (ESRI Japan Inc., version 9.3). As health outcomes of interest, we used neonatal mortality (<28 days) and infant mortality (<1 year). These were ascertained from the Woman’s Questionnaire in the EDSMD-IV. In line with previous studies [7], [12]–[14], our covariate sets included child, maternal as well as household characteristics. As the child characteristics, we adjusted for the birth order. The maternal characteristics included maternal age at birth, current maternal smoking status (smoker or not), and birth spacing (<2 years or ≥2 years). Further, to assess current maternal health status, we used maternal anemia (hemoglobin <11.0 g/dL), maternal height, and maternal body mass index. Note that of the 12345 eligible births, the information on maternal health status was available for only 6041 births. The household characteristics included household wealth, maternal education level (no education, primary school, secondary school or higher), religion, type of the nearest health center (CSB I vs. all other), and existence of referral hospital within 30 km, which is assumed to be 1 hour’s travel by car. Household wealth was defined in terms of characteristics related to wealth status, including household ownership of a number of consumer items such as a television and a car, dwelling characteristics such as flooring materials, drinking water source, and toilet facilities [15]. Then, the index was divided into tertiles. As referral hospitals, 102 centers including CHD II, CHR, and CHU in the higher or highest levels of the pyramidal system were adopted (Figure 3). We categorized the distance into five groups: first, we adopted two cut-off points (≤5 km, >5–10 km, and >10 km), as recommended by the World Health Organization (WHO) to monitor the health status in developing countries; then, we additionally divided the distance ≤5 km into three categories, approximately by tertile (≤1.5 km, >1.5–3.0 km, and >3.0–5.0 km). We also evaluated linearity of the crude relationship between a distance to a health center and the risk for each outcome in a graphic examination by using linear models (LM). By using the lm function in R version 2.14.1 (R development Core Team 2011), we utilized natural splines with five degrees of freedom for the distance. The data had a three-level structure of 12345 births at level 1, nested within 584 village locations at level 2, in turn nested within 22 regions at level 3. We thus used three-level logistic regression models with a random intercept. In other words, we allowed the intercept to vary across geographic localities since our data covers the whole nation with wide variations in terms of regional characteristics, e.g., malaria endemicity and different intervention programs for vaccinations. Note that a distance to a health center, type of the nearest health center, and existence of referral hospitals within 30 km were treated as level 2 variables in the models. After examining the crude association between the proximity to a health center and each health outcome, we adjusted for child characteristics, maternal characteristics, and household characteristics (model 1). Then, we additionally adjusted for maternal health status (model 2). The fixed and random parameter estimates (along with their standard errors) for the multilevel binomial logit link model were calibrated using a marginal quasi-likelihood procedure with first order Taylor series expansion, as implemented within the MLwiN 2.24 [16]. We used the second nearest group (>1.5–3 km) as the reference category, because the nearest group (≤1.5 km) was likely to be comprised of an unrepresentative subjects with high wealth status and maternal education attainment who lived in urban areas. We calculated the odds ratios (ORs) and 95% confidence intervals (CIs) for each health outcome. A P-value <0.05 (two-sided test) was considered statistically significant. Although we adjusted for the birth order in the main analysis, a previous study has implied that the association between birth order and childhood mortality might be J- or U-shaped [10]. Thus, we additionally stratified the subjects according to the birth order, and examined the associations between proximity to a health center and each health outcome. In the stratified analyses, the birth order was divided into three categories (first order, second or third order, and fourth or more). As a sensitivity analysis, we also conducted the analyses by using the nearest group (≤1.5 km) as a reference category. Subsequently, we estimated mean predicted probabilities by a three-level model, including cross-level interaction, between proximity to a health center and household wealth, maternal educational attainment, and maternal health. In this analysis, we modeled the distance as a continuous variable (per increase of 1 km), and we calculated mean predicted probabilities for each health outcome, adjusting for maternal characteristics, maternal health status, and household characteristics. As a supplementary analysis, we described behavioral characteristics of the subjects and their mothers by the distance to a health center.