Elsevier

Health & Place

Volume 17, Issue 4, July 2011, Pages 929-936
Health & Place

Spatial modeling of geographic inequalities in infant and child mortality across Nepal

https://doi.org/10.1016/j.healthplace.2011.04.006Get rights and content

Abstract

A survival regression model that allows for spatially correlated random effects is used to predict the hazard of dying among 12,714 children born between 1996 and 2006 in Nepal. The maps of fitted hazard rates show that even after accounting for individual and community-level covariates, a residual spatial pattern in infant mortality remains, with higher mortality concentrated in parts of Nepal's Far-Western and Mid-Western development regions. Results suggest a need to consider health policies and programs that reach children in spatially concentrated high-mortality areas.

Introduction

Infant and child mortality measures are sensitive indicators of population health and wellbeing, as they reflect a combination of individual, mother, household, community, and environmental factors (Mosely and Chen, 1984). Underscoring the importance of the measure in reflecting population health and development, under-five mortality was adopted as one of the eight Millennium Development Goals by the UN member states in 2000 (United Nations, 2000). Improvements in child mortality reflect national and international health policy and program efforts to raise education levels, combat undernutrition and poverty, and implement immunization campaigns. In spite of increased attention to improve survival of young children, some 9–10 million children still die before their fifth birthday each year (Murray et al., 2007, Loaiza et al., 2008, UNICEF, 2009).

The determinants and correlates of child mortality have been widely studied, and are generally comprised of individual, maternal and household factors, and community or environmental determinants. Biological causes, as well as nutrition deficits and illness can contribute to a child's risk of dying, particularly at early ages, though these factors are not addressed in this study. Maternal factors such as age, parity, birth order, education, and socioeconomic status are associated with changing relative risks of child mortality (Hobcraft et al., 1985; Boerma and Bicego, 1992; Rutstein, 1984, Rutstein, 2005, Rutstein, 2008, Wang, 2003).

The determinants of mortality have been well-studied using nationally representative data from the MEASURE Demographic and Health Surveys (DHS) program (Mahy, 2003, Wang, 2003). Mother's age at birth typically exhibits a U-shaped relationship with the risks of child deaths. Children born to mothers at young and old ages tend to experience higher risks of dying. Young mothers are more often socioeconomically disadvantaged and less educated, which are associated with increases in the risk of child death, while older mothers may have experienced more pregnancies, increasing the risk of child death as parity increases (Mahy, 2003). Both very young and older mothers may be more likely to have pregnancy and delivery complications, further increasing the risk of child death. Birth intervals of 36–47 months exhibit the lowest risks of neonatal, infant, and under-five mortality, but the benefits of longer intervals seem to diminish after 47 months (Rutstein, 2005, Rutstein, 2008).

Around 2001, DHS began to make available to the public latitude and longitude coordinates for the communities where survey respondents live. Some studies incorporated geographically derived variables into non-spatial models of child mortality or morbidity (Balk et al., 2004, Curtis and Hossain, 1998). Other research explicitly used spatial models to study child mortality with DHS data. Gemperli et al. (2004) investigated spatial patterns of malaria endemicity as well as socioeconomic risk factors on infant mortality in Mali in a Bayesian hierarchical geostatistical model. Aside from confirming the expected relationships between individual and maternal factors, the resulting residual spatial pattern of infant mortality showed a clear relation to well-known foci of malaria transmission. However, no effect of estimated parasite prevalence could be demonstrated, possibly due to confounding by unmeasured covariates and sparsity of the source malaria data. Kazembe et al. (2007) studied the influence of individual determinants, malaria endemicity, and group-specific environmental factors approximated by geographical location on child survival in Malawi using a spatial Cox model. The expected socioeconomic effects on infant and child mortality were confirmed; and while malaria endemicity was not associated with the risk of infant mortality, it was an important determinant of child mortality. Kandala and Ghilagaber (2006) used a geo-additive Bayesian child survival model for Malawi that showed district-level socioeconomic characteristics were important determinants of childhood mortality. Residual district-level clustering of childhood mortality suggested the remaining importance of underlying spatial effects.

Nepal has achieved remarkable reduction in child mortality over the past decades, yet the rates are still among the highest in the world. While health and development programs have contributed to the improvements, high rates of mortality persist in Nepal, and further reduction may depend on improved targeting of intervention programs. The individual and socioeconomic determinants of child mortality have been widely studied, yet the methods used in previous research on child mortality in Nepal have not accounted for spatial autocorrelation of community-level effects, nor estimated residual spatial inequalities in mortality (Thapa, 2008, Katz et al., 2003, Gubhaju et al., 1991). The objective of this study is to explain the spatial pattern of child mortality in Nepal as a function of socio-demographic and community characteristics while accounting for potential spatial autocorrelation in covariates at the community level. This study builds on existing methodologies by applying a flexible hierarchical survival model that allows for spatially correlated random effects. The spatial correlation may result from the spatial arrangement of the community locations, because communities in close proximity to each other may be more alike than communities farther away from each other. Model-predicted mortality is then mapped to display the remaining spatial pattern of mortality that remains after accounting for community-level spatial autocorrelation.

Nepal is a small country of 29 million people, 80 percent of whom live in rural areas. The annual per capita income is $1530, among the lowest in the world. The geography of the country is highly diverse, ranging from the Himalayan mountain range to the flat Terai lowlands along the Indian border. Although under-five mortality rates have declined significantly (48 percent) since 1991, the country still ranks in the bottom third of nations globally (UNICEF, 2009, Ministry of Health and Population [Nepal] et al., 2007). Children under five experience a mortality rate of 61 deaths per 1000 live births (Ministry of Health and Population [Nepal] et al., 2007). Results from the most recent Nepal Demographic and Health Survey (NDHS) show that child mortality varies substantially across the country (Table 1). The Mid-Western development region experiences under-five mortality rates of almost twice that of the lowest region in Nepal. Across the three ecological zones, the Mountain zone experiences under-five mortality rates of about twice that of the Hill zone. Infant and child mortality is also higher in the Mountain zone, and in the Mid- and Far-Western development regions. Maps of infant and child mortality rates (Fig. 1, Fig. 2) display the same pattern of high rates in the West, and lower rates in the East. Yet it is likely that these aggregate estimates mask substantial heterogeneity within these regions.

Section snippets

Population

Data from the Nepal Demographic and Health Surveys carried out in 2001 and 2006 are used for the analysis. These nationally representative surveys were carried out under the USAID-funded MEASURE DHS project in collaboration with the Nepal Ministry of Health and New ERA (Ministry of Health [Nepal] et al., 2002, Ministry of Health and Population [Nepal] et al., 2007). The NDHS samples are two-stage cluster sample surveys, designed to be representative at the national level and for both urban and

Methods

Time until death was modeled using a proportional hazards model with spatially correlated random effects. This model generalizes the usual proportional hazards model in two ways. First, it allows for a multilevel structure, as in the data, where participants are nested into PSUs, via a frailty term (Vaupel et al., 1979), which is the survival data analog to random effects in regression models (Gutierrez, 2002). In the usual proportional hazards model, the hazard of the event at a given time has

Results

The association between infant mortality and mother and household characteristics is presented in Table 4. Among children under one year of age, boys experienced 16 percent lower hazard of death while multiple birth children experienced over nine times the hazard of dying compared to single births. Second and higher order births experienced lower hazards of dying compared to the first birth children. Children born to mothers over the age of 35 experienced over four times the hazard rate of

Conclusion

The death of a child is a relatively rare event, even in Nepal, which has a relatively high mortality rate. Mother and household level covariates are significant predictors of both infant and child mortality. However, a substantially significant spatial trend remains for infant mortality after holding these mother and household level predictors constant. The findings show increased mortality in the Far-Western and Mid-Western development regions after accounting for mother and household

Acknowledgments

The authors would like to thank Xavi Puig for his technical assistance in the modeling process and Pradeep Adhikari for providing the health facilities GIS data.

References (36)

  • A. Gemperli et al.

    Spatial patterns of infant mortality in Mali: the effect of malaria endemicity

    American Journal of Epidemiology

    (2004)
  • B. Gubhaju et al.

    Socioeconomic, demographic and environmental determinants of infant mortality in Nepal

    Journal of Biosocial Science

    (1991)
  • R.G. Gutierrez

    Parametric frailty and shared frailty survival models

    Stata Journal

    (2002)
  • J.N. Hobcraft et al.

    Demographic determinants of infant and early child mortality: a comparative analysis

    Population Studies

    (1985)
  • N.-B. Kandala et al.

    A geo-additive Bayesian discrete-time survival model and its application to spatial analysis of childhood mortality in Malawi

    Quality and Quantity

    (2006)
  • J. Katz et al.

    Risk factors for early infant mortality in Sarlahi district, Nepal

    Bulletin of the World Health Organization

    (2003)
  • L.N. Kazembe et al.

    Spatial analysis of the relationship between early childhood mortality and malaria endemicity in Malawi

    Geospatial Health

    (2007)
  • D.J. Lunn et al.

    WinBUGS—a Bayesian modelling framework: concepts, structure, and extensibility

    Statistics and Computing

    (2000)
  • Cited by (33)

    • General equilibrium effects of spatial structure: Health outcomes and health behaviours in Scotland

      2014, Regional Science and Urban Economics
      Citation Excerpt :

      If these underlying spatial patterns are not accounted for this is likely to bias research findings and thus invalidate conclusions (Bech and Lauridsen, 2008). Chin et al. (2011) and Barufi et al. (2012) found evidence of spatial correlation in Nepal and Brazil respectively, after controlling for individual and community-level covariates. Such spatial patterns in unexplained variation indicate that place based targeting of health policies could be effective.

    • Estimation of district-level under-5 mortality in Zambia using birth history data, 1980-2010

      2014, Spatial and Spatio-temporal Epidemiology
      Citation Excerpt :

      Small area models are statistical models which address the issues raised by small sample sizes by explicitly accounting for the large sampling variance and exploiting spatial and temporal relatedness to increase predictive power. These types of models have been occasionally applied to birth history data on mortality in the past, but the focus has usually been on the relationship between socioeconomic or environmental factors and the risk of death, not prediction of under-5 mortality or other similar childhood mortality indicators (Gemperli et al., 2004; Adebayo et al., 2004; Kandala and Ghilagaber, 2006; Kazembe et al., 2007; Kazembe and Mpeketula, 2010; Asiimwe et al., 2011; Chin et al., 2011). This is likely because mortality data pose several challenges in the context of small area methods.

    • Altitude and early child growth in 47 countries

      2021, Population and Environment
    View all citing articles on Scopus
    View full text