Linking climate change and health outcomes: Examining the relationship between temperature, precipitation and birth weight in Africa

Global Environmental Change 35 (2015) 125–137

Contents lists available at ScienceDirect

Global Environmental Change journal homepage: www.elsevier.com/locate/gloenvcha

Linking climate change and health outcomes: Examining the relationship between temperature, precipitation and birth weight in Africa Kathryn Grace a,*, Frank Davenport b, Heidi Hanson c, Christopher Funk b,d, Shraddhanand Shukla b,d a

Department of Geography, University of Utah, United States Climate Hazards Group, University of California, Santa Barbara, United States c Family and Preventative Medicine, University of Utah, United States d US Geological Survey (USGS), United States b

A R T I C L E I N F O

A B S T R A C T

Article history: Received 14 November 2014 Received in revised form 26 June 2015 Accepted 29 June 2015 Available online

This paper examined the relationship between birth weight, precipitation, and temperature in 19 African countries. We matched recorded birth weights from Demographic and Health Surveys covering 1986 through 2010 with gridded monthly precipitation and temperature data derived from satellite and ground-based weather stations. Observed weather patterns during various stages of pregnancy were also used to examine the effect of temperature and precipitation on birth weight outcomes. In our empirical model we allowed the effect of weather factors to vary by the dominant food production strategy (livelihood zone) in a given region as well as by household wealth, mother’s education and birth season. This allowed us to determine if certain populations are more or less vulnerable to unexpected weather changes after adjusting for known covariates. Finally we measured effect size by observing differences in birth weight outcomes in women who have one low birth weight experience and at least one healthy birth weight baby. The results indicated that climate does indeed impact birth weight and at a level comparable, in some cases, to the impact of increasing women’s education or household electricity status. ß 2015 Elsevier Ltd. All rights reserved.

Keywords: Africa Health Climate Agriculture

1. Introduction Climate change has the potential, and has likely already begun, to adversely affect the health of millions through changes in agricultural production and food insecurity (Jones and Thornton, 2003; Juli and Duchin, 2007; Fraser, 2006). However, the link between weather and health remains controversial and somewhat convoluted, in part because of the complications inherent in analyzing data with different temporal and spatial scales. In this study, we explore the link between birth weight, and weather in 19 African countries – some of the poorest and most climatically exposed areas in the world. Low birth weight may adversely affect the health and economic trajectories of an individual throughout the life course (Walker et al., 2007; Victora et al., 2008). The in utero period is a critical

* Corresponding author. E-mail address: [email protected] (K. Grace). http://dx.doi.org/10.1016/j.gloenvcha.2015.06.010 0959-3780/ß 2015 Elsevier Ltd. All rights reserved.

period of human development. Dramatic changes in temperature or precipitation during pregnancy may adversely affect fetal growth or initiate spontaneous preterm births. We focus on birth weight because low birth weight is costly for individuals, households and communities. Prior research suggests strong links between birth weight and future health problems, educational attainment, and income (Walker et al., 2007; Victora et al., 2008). Increasing our understanding of the link between birth weight, low birth weight and variation in climate is essential to improving health in climate sensitive communities. The relationship between heat stress, heat-waves, precipitation and birth weight is complex (Wells and Cole, 2002; McGrath et al., 2005; Strand et al., 2011). To date, however, no study of African households using fine resolution precipitation and temperature data exists. Additionally, because weather may be linked to local food production, we argue that in countries with a high incidence of food insecurity, it is important to also consider variability in livelihood strategies. This consideration is important for the countries in our study because they are subsistence-based

K. Grace et al. / Global Environmental Change 35 (2015) 125–137

126

economies where food producers rely on precipitation to grow their food crops. Our primary goal is to examine the linkages between community-level precipitation and temperature variations and birth weight. We seek to answer the following question: Do increasing temperature and decreasing precipitation result in decreased birth weight – even after adjusting for variations in food production strategy, and household- and individual-level factors? The data for this study comes from retrospective birth records included in the Demographic and Health Survey (DHS). The DHS data are combined with satellite observed precipitation, mean satellite observed land surface temperatures, and mean satelliteobserved brightness infrared temperatures to create a 0.18 grid that covers the entire region (Funk et al., 2012). Recorded live birth weights are temporally and spatially matched to community-level precipitation and temperature data. To account for the variation in birth weight that may be associated with food production we incorporate growing season information and use livelihood zone data collected by the US Agency for International Development’s Famine Early Warning System (FEWS NET) program. The data information on the 19 countries included in the study is found in Table 1 and the map in Fig. 1 presents the 17 of the 19 countries with specific data from FEWS NET. 1.1. Background 1.1.1. Low birth weight Biological, behavioral, nutritional, and socio-environmental factors act in concert to impact the weight of a baby at birth (Mwabu, 2008; Abu-Saad and Frasier, 2010; also see Kramer, 1987, for an extensive discussion). Given changes in the global climate that may impact local weather patterns, particularly precipitation and temperature, interest in the relationship between ambient air temperature, precipitation, and birth outcomes has increased. These patterns may linearly affect birth weight through growth restriction and lead to increased rates of preterm birth. Further understanding of the nature of the relationship will allow for public health programs in the developing world to develop appropriate programs and underscore the adverse health effects of a changing climate.

About 15% of all babies born in the continent of Africa weigh less than 2500 g which is the standard cut-off to establish low birth weight (LBW) (Wardlaw and Tessa, 2004). This level is twice the US share and almost three times the share of LBW babies born in Europe (United Nations Children’s Fund and World Health Organization, 2004). African LBW babies are unlikely to receive the additional medical care available in Europe or the US. When LBW babies are born, their risk of mortality increases as compared to healthy weight babies (Walker et al., 2007). LBW babies who do survive are more likely to develop disabilities (vision, hearing, cognitive functioning), are less likely to attain the level of education and income of their healthy weight counterparts and, in some cases, have behavioral problems (Victora et al., 2008; Bhutta et al., 2002; Darlow et al., 2005; Farooqi et al., 2006; Hack et al., 2005; Hille et al., 2007; Marlow, 2004). Additionally, LBW girls may be more likely to birth their own LBW infants when they begin childbearing (Victora et al., 2008). Because LBW infants may die, or have special needs and require specialized care, the entire family supporting the infant may also experience negative economic and psychological effects of LBW (Alderman and Behrman, 2006; Singer et al., 1999). In many African countries where there is often limited physical, emotional, and institutional support for LBW infants and their families, coping with the realities of an LBW infant may cause significant challenges for the families and the communities. Taking a broader perspective, because LBW has a negative impact on overall educational attainment and earnings potential, low-income countries with high rates of LBW will likely face development challenges well into the future. Therefore, understanding the mechanisms that result in LBW may aid in reducing the rate of LBW, ultimately improving the lives of infants and their families and improving the human development trajectories of the poorest countries in the world. Previous research, focused almost exclusively on the developed world, has identified the importance of birth seasonality on birth weight with inconclusive results (Cooperstock and Wolfe, 1986; Keller and Nugent, 1983; Rayco-Solon et al., 2005). From these studies, LBW appears to occur more often during the autumn and winter months – a time period linked to seasonal food availability (Hort, 1987; Strand et al., 2011). Food availability and seasonal

Table 1 Count of households by country and time period. 1990 Burkina Faso Central African Republic Ethiopia Guinea Kenya Lesotho Liberia Madagascar Malawi Mali Niger Nigeria Rwanda Senegal Sierra Leone Tanzania Uganda Zambia Zimbabwe Total

1992

1993

1994

1996

1997

1998

1530

1999

2000

2001

1010

2003

2004

2005

2006

2007

2008

2009

2010

2558

5098 1169

1169

615

599 2020 1670

1900 1731

1292 541

1037

3879 3584

992

3580

6249

1948

1550

3254

900

758

628 1193

2753 1838 3621

1706

1545 1016 1016

1957

1550

2723

1169

992

2743

900

3667

2947 5215

1948

4856

5311

11,025

4796

2498

1214 2020 3570 3023 541 4916 13,413 6194 2450 4139 1838 6520 1545 1016 2558 1957 4588

1542

1641 758

Total

10,077

1292

6249

67,769

K. Grace et al. / Global Environmental Change 35 (2015) 125–137

127

Fig. 1. Map of select countries and sampling locations.

food issues are more likely to be relevant in low-income countries characterized by high rates of food insecurity. Only a few studies evaluate birth weight and food availability among low-income countries, and the results have been inconclusive (Rousham and Gracey, 1998; Rayco-Solon et al., 2005; see Grace et al., 2014 for a discussion of food security and birth weight). Other studies have directly evaluated the impact of ambient temperature on birth weight, supported by the idea that heat stress can have an impact on a developing neonate (Ruff, 1994). Pregnant women are hypothesized to be more sensitive to warm temperatures because of the additional physical and emotional strain caused by increased ambient temperature (Prentice et al., 1993; Wells and Cole, 2002; Berman et al., 1982; Okun et al., 2009). In a synthesis of the relevant studies of ambient temperature, Strand et al. (2011) suggest that extreme temperatures, either warm or cold, may be more important than just seasonal temperature variation, in terms of birth weight outcomes. Their finding is supported by the presence of statistically significant correlations between temperature and birth weight with the lack of consistent findings across study areas. In other words, they find that in some places increased risk of LBW occurred during the hot months in other places during the colder months. Furthermore, the period during pregnancy when the extreme temperatures occurred may also be related to birth weight because of the changing needs of the neonates and the pregnant women. Extreme temperatures may also lead to spontaneous preterm birth, and therefore LBW (Basu et al., 2010; Wang et al., 2013). Overall the research on birth weight and temperature has produced inconsistent results. This likely reflects differences in available data and analytic approaches, as well as the differing influences of regional geographies and climate. In some cases, temperature increases are associated with decreases in birth weight but not necessarily an increase in LBW (Lawlor et al., 2005; Descheˆnes et al., 2009; Flouris et al., 2009; Torche and Corvalan, 2010). In other contexts, usually cooler climates, an increase in

temperature was positively correlated to birth weight (Elter et al., 2004; Murray et al., 2000) and in one study of New Zealand births, there was no significant correlation (Tustin et al., 2004). The impact of precipitation, drought, and floods has also been studied and also produced inconsistent results (Hoddinott and Kinsey, 2001; Bantje and Niemeyer, 1984; Murray et al., 2000; Uddenfeldt Wort et al., 2004). Extreme levels of precipitation may increase the risk of LBW. Low precipitation may adversely affect birth outcomes through food insecurity and undernutrition. Borders et al. (2007) found that food insecure mothers were three times more likely to have a LBW child than food secure mothers. Restricted access to normal caloric and nutrient intake during pregnancy may also adversely affect the birth weight of the infant (Godfrey et al., 1997; Haggarty et al., 2009); however, the timing and type of nutrient deficiency are important considerations with the suggestion that even the period leading up to conception may be significant to the overall healthy physical development of the infant (Lunney, 1998). High precipitation may also lead to increased rates of LBW through placental malaria infections (Bantje, 1987; McGregor et al., 1983; Murray et al., 2000; Newman et al., 2003; Uddenfeldt Wort et al., 2004), particularly in women giving birth to their first child (Okoko et al., 2002). Several studies have shown an increase in malarial infections in pregnant women, especially during the first pregnancy (Uddenfeldt Wort et al., 2004). The final component of birth weight relevant to this study is the relationship between birth weight and maternal nutrition/food security. In terms of defining food security, we rely on the widely used definition which states that: ‘‘Food security exists when all people, at all times, have physical and economic access to sufficient, safe and nutritious food that meets their dietary needs and food preferences for an active and healthy life’’ (World Food Summit, 1996). In this research, any time that one of the underlying pillars of food security – availability, access, utilization or stability – is absent the individual, household or community is

128

K. Grace et al. / Global Environmental Change 35 (2015) 125–137

considered to be food insecure. In communities characterized by a high incidence of food insecurity, maternal weight gain is directly linked to food availability and accessibility (FAO, 2006). Because birth weight may reflect nutritional deficiencies of the mother during her pregnancy (Young, 2001), an analysis of the impact of temperature and precipitation on birth weight must also consider the factors related to food insecurity. We incorporate growing season and livelihood information, in this study ‘‘livelihood’’ refers to the strategy by which communities produce food or earn income, because different livelihood strategies can be differentially impacted by temperature and/or precipitation changes. For example, agriculturalists may benefit from increased temperatures during some periods of the growing season while these increases may be extremely problematic for early stages of crop growth. Including the livelihood information assists in disentangling the nutritional relationship of temperature and precipitation with birth weight. Our analysis examines nearly 70,000 births over a 20-year period and adjusts for spatially and temporally specific weather impacts. We aim to expand this area of research and ultimately improve understanding of the factors impacting birth weight and LBW among low-income countries. 2. Data and measures To identify and isolate the impact of temperature and precipitation variation we incorporate factors related to an infant’s biology, the socioeconomic characteristics of her or his mother and household, and factors related to food access. Therefore, to examine LBW we rely on two primary types of data – health and environmental/geographical. The descriptions of the data and the measures calculated from each data source follow. 2.1. Health data and measures DHS data provide detailed health and population information for the poorest countries around the world and aim to capture a representative sample of the study population. These crosssectional data are spatially referenced data at the communitylevel because DHS provides the latitude and longitude values of their sampling clusters. The sampling cluster is a geographic location that represents the centroid of the residential area where the child lives. To preserve anonymity, these clusters are randomly spatially displaced up to 10 km from their actual locations. We accommodate this random displacement by assuming that the cluster can be located anywhere within a 10 km radius of the provided latitude/longitude. At the individual-level, the data contain detailed information about maternal characteristics and provide retrospective information about the health of infants, including their birth weights. The DHS provides the micro-level health and socioeconomic information used in the analysis. All African countries with geo-referenced DHS data and residence information of the respondents are included in the study. Because many countries have multiple time periods of data collection, we often use data from several time periods within one country. Table 1 lists the surveys by country and date used in this study and Fig. 1 presents the locations of the clusters across Africa for 17 of the 19 countries – only those countries with FEWS NET livelihood information as discussed below. Births recorded in the 5 years prior to the DHS interviews are included in the analysis. Each singleton infant, with a recorded birth weight provided by the mother, is classified as LBW or nonLBW using the 2500 g WHO cutoff. Both the recorded birth weight and the categorical LBW measure will serve as the dependent variables. We account for recall bias by including a variable that indicates whether the mother had the weight of a child on a birth

information card or whether she simply recalled the birth weight (Boerma et al., 1996). DHS data also provide information on length of time the mother has lived at her current residence. We use this information to restrict the sample to only those households where the pregnancies occurred in the current area of residence. This restriction allows us to link past climate information with the relevant pregnancy. Unfortunately, DHS does not contain information on prior residence making it impossible to construct a full residential history or allow us to link women with information related to their prior place of residence. Additionally, the DHS has recently discontinued this question and surveys collected after 2010 generally do not have any information on time at current residence. We select a suite of explanatory variables from the DHS to control for variation in birth weight. The sex of the infant may be related to birth weight so each child’s sex is included in the analysis. We also adjust for birth order, and birth interval, with the assumption that children of higher birth order and those who were born within a short period of time following the preceding birth may be smaller than their counterparts (Seidman et al., 1988; Rutstein, 2008). There may also be fewer food resources per person in a household with more children and the mother’s time for personal care may be limited by the increase in childrearing responsibilities that come with more young children and children spaced close in age. Mothers’ characteristics are also included as control factors. Mother’s height is included because it reflects each mother’s experience with food insecurity – women who were stunted as children are more likely to give birth to LBW infants (Young, 2001). Maternal age is related to infant size as babies born to older women are more likely to weigh more than those born to younger women (Swamy et al., 2012; Seidman et al., 1988). Mother’s education is also included. More educated mothers may have greater access to health care, may have a greater understanding of nutritional requirements during pregnancy, and may also have greater social capital which may help to ensure a healthier and less stressful pregnancy (Caldwell, 1979, 1994; Hobcraft, 1993). With respect to household characteristics, floor material and electricity status serve as measures of household wealth (see Rutstein, 2008 for discussion of measuring wealth and different DHS variables). We assume that the poorest households, those with unfinished flooring and no electricity, will have fewer options available for alleviating weather induced stresses. We include type of place of residence – urban or rural. We assume that urban households will likely have greater access to food markets and prenatal health care, which are both likely to impact the weight of a baby at birth. Finally, we incorporate birth month and country, owing to the potential for macro-level political or environmental events that may impact health outcomes that we have not accounted for. Table 2 contains summary information on the variables included in the analysis. 2.1.1. Environmental/geographical data and measures The livelihood zone data come from FEWS NET’s efforts to characterize the dominant livelihood strategies in a number of low-income countries. With the use of local weather patterns, market information, and local expert knowledge, FEWS NET has constructed zones that characterize the dominant strategy used to produce food in a general area (see fews.net) for each country. The livelihood zone data designates different areas based on the dominant subsistence pattern and is carefully constructed to reflect differences within each country. These categories reflect within-country nuances and have been used in related research to explore differential risks and vulnerabilities associated with area of residence (Grace et al., 2012, 2014). The livelihood zone information serves as a measure of a community’s dependence on

K. Grace et al. / Global Environmental Change 35 (2015) 125–137

129

Table 2 Dependent and independent variables included in the analysis. Mean

Count Independent variables Child’s sex Child’s birth order Mother’s age (years) Mother’s height (mm) Marital status Mother’s education Prenatal care Malaria drugs (during pregnancy) Birth weight recall Urban residence Electricity Floor material

Born during the rainy season Temperature (days above 100 F) Temperature (days above 105 F) Rainfall (mm)

girl 33,230; boy 34,539 67,769 67,769 49,221 married/partnered 59,796; no partner 7973 none 42,375; primary 19,810; secondary and beyond 5584 none 1069; not recorded 14,517; at least one 52,183 no 12,953; not recorded 29,602; unsure 326; yes 24,888 card 26,190; recall 34,842 rural 39,957; urban 27,812 none 49,497; yes 18,272 finished 34,093; natural 32,854; not recorded 104; other 200; unfinished wood 518 agricultural 29,404; agropastoral 15,324; fishing 2371; irrigated 2024; pastoral 1483; urban 7239 Jan. 5603; Feb. 5165; March 5826; April 5548; May 5552; June 5526; July 5321; Aug. 5744; Sept. 6164; Oct. 5902; Nov. 5600; Dec. 5818 yes 26,332; no 41,437 67,769 67,769 67,769

Dependent variables Birth weight (grams, continuous) Birth weight (categorical)

67,769 LBW 6694; Healthy birth weight 61,075

Livelihood zone Birth month

precipitation for food production, which may be related to birth weight through the mother’s food intake during pregnancy. We aggregate these into dominant groups – ‘pastoralist’, ‘agropastoralist’, and ‘agriculturalist’ (see Table 2) within each country. We also separate out urban residents in the livelihood classifications, owing to the potential for different food sources and different dependencies on precipitation and temperature. In addition to livelihood zone information, FEWS NET also provides a growing season calendar. We include this information in the analysis to identify typical hunger periods and growing season periods. Instead of using calendar month of birth alone, we can identify when the birth occurred with respect to known periods of hunger or intense labor. This additional data provides further nuance and understanding into the seasonal aspects of birth weight. The nutritional needs of pregnant women may vary depending on the stage of pregnancy. Given the sensitivity to timing, we incorporate seasonality, as well as the temperature and precipitation information as described later, at the scale of pregnancy trimesters (trimesters 1–3). Additionally, because of the potential significance of the preconception period, we also include a ‘trimester 0’ – corresponding to the three month period leading to conception. To explicitly control for the effects of seasonality, birth month was included in the analyses as a fixed effect. This allows us to control for all of the unobservable characteristics associated with month of birth. In addition to controlling for birth month, we control specifically for births during the growing season and livelihood zone because seasonality of birth weight has been linked to maternal nutrition and food availability. Our primary independent variables of interest are temperature and precipitation. Gridded data used to calculate the temperature and precipitation metrics used in the study are derived from combined satellite and ground station observations. Each of the datasets is described in more detail below. We link the environmental data to a child’s in-utero experience by using the child’s birthdate and the location of the DHS sampling cluster. Because of the spatial displacement mentioned previously, we link

3.58 28.88 1593

7.82 2.23 188

3217

Std. dev.

Min.

Max.

2.4 6.86 66.93

1 15 1400

18 49 2059

17.5 8.5 120.5

707.09

0 0 2.3

200

92 88 821.1

6000

each child’s birth weight record to environmental data in a 10 km radius around the sampling cluster for a period one year prior to, and including, the child’s birthdate – trimesters 0–3. This study uses the Climate Hazards Group InfraRed Precipitation with Station (CHIRPS) dataset (Funk et al., 2014a). The CHIRPS data set, developed recently by USGS scientists in collaboration with Climate Hazards Group at University of California at Santa Barbara, combines a high resolution (0.058) climatology (Funk et al., 2012, 2015) with time-varying station data and observations from geostationary weather satellites. CHIRPS period of record, 1981-present, compares reasonably well with in situ rain gauge observations (Tote´ et al., 2015) in Africa. While CHIRPS tends to underestimate sub-seasonal precipitation extremes, it does a good job capturing seasonal year-to-year variations (Tote´ et al., 2015). CHIRPS is currently in use by USAID supported projects for monitoring and forecasting drought conditions in Africa (Funk et al., 2014b). The temperature dataset used in this analysis was developed by Sheffield et al. (2006). This dataset has been widely used in various global studies (Sheffield and Wood, 2008a,b; Shukla et al., 2013) and also supports Princeton University’s Africa Food and Drought Monitor. Further details about this monitoring system can be found in Sheffield et al. (2013). This temperature data is primarily derived from the National Center for Environmental Prediction’s reanalysis (Kalnay et al., 1996; Kistler et al., 2001). The reanalysis is a long record of global analyses of atmospheric and surface fields, such as air temperature, radiations, wind speed, sea-level pressure and sea-surface temperature. It is generated by assimilating a longterm quality controlled database of land, oceanic, and atmospheric observations to a global climate model simulation that was kept constant throughout the period of analysis. To generate temperature data as used in this analysis, Sheffield et al. first bilinearly interpolated reanalysis air temperature from its native resolution of approximately 1.98 latitude  1.8758 longitude to a 2.08 regular grid while allowing for elevation related changes (Sheffield et al., 2006). These gridded data were then spatially downscaled to an 0.58 regular grid using bilinear

130

K. Grace et al. / Global Environmental Change 35 (2015) 125–137

interpolation with adjustments for change in the elevation. Specifically, the temperature data was lapsed at a rate of 6.5 8C km1 following the methods of Cosgrove et al. (2003). Next, the monthly averaged temperature data was corrected to match the monthly temperature values of the Climate Research Unit’s 0.58 gridded temperature data (New et al., 1999, 2000). This step was undertaken to remove any bias in the temperature data that is inherent in reanalysis products. Finally, the daily values from the uncorrected temperature dataset were shifted so that the monthly values match the corrected monthly averages. The temperature and precipitation data used here are presented in Fig. 2. 3. Methods 3.1. Temperature and precipitation 3.1.1. Precipitation The precipitation data is observed monthly at 0.058 spatial resolution (5 km). Each grid cell records the total precipitation (in mm) for a given month at that location. Precipitation records for each child are generated by taking the average value of grid cells that lie within a 10 km radius of the child’s sampling cluster. To construct the growing season precipitation variable, we use an area-weighted average, meaning some grid cells get cropped and only the portion within the radius contributes to the average, to ensure that the same total area is used for each calculation. This is done separately for every month one year before the child’s birth date. Monthly precipitation values are then summed over trimesters to create a measure of total precipitation during each three month period of the year before the child’s birth, including the trimester 0. Using the FEWS NET growing calendar information, we sum the rain per trimester that falls within the growing season. For example, if the first trimester of the pregnancy contains only one month of the prime growing season for that region then only the precipitation during that growing season month is counted. In this way, we can examine relationship between rain and birth weight with attention to the needs of subsistence farmers reliant on precipitation. It is important to note that some of the births may be pre-term. This information is rarely available in DHS but as many as 26.1% of LBW births in Africa may be pre-term (Marchant et al., 2012). In the case of pre-term birth, the trimester information will be incorrect. Without gestational age, there is not adequate information to support an adjusted analysis. However, exploratory research of pre-term births using monthly proxy measure of precipitation and temperature for a subset of the sample, did not give any indication that a finer temporal scale was necessary or that the relationship between the environmental variables and LBW would change. 3.1.2. Temperature Our objective is to measure heat stress occurring prior to and during the pregnancy. Our primary constructed measure of temperature, referred to as hot days, is the count of days in each trimester where the maximum daytime temperature exceeds 105 F, and a lower temperature of 100 F. 105 F represents the approximate heat stress temperature threshold for very arid environments as indicated by NOAA (see Harlan et al., 2006). We use these two different temperature thresholds, one for very hot environments and one for less hot environments, because an individual’s response to heat is expected to vary dependent on if they have acclimated to a hot environment (Harlan et al., 2006). To obtain this measure we first calculate maximum daily temperature for a given calendar day and grid cell. The values are then joined to the cluster locations based on the 12 months before the child’s birth and a 10 km radius surrounding the sample cluster location. We then count the number of days in each birth month where the

grid cell values exceed 105 F and 100 F as the maximum daily temperature. These counts are then summed over trimesters to get a count of ‘‘hot’’ days that exceed one of these two threshold values during each trimester. As with precipitation, we also sum the number of hot days during the trimesters that correspond to the growing season to construct a growing season specific measure. 3.2. Statistical analysis As noted in the background section, other studies of low birth weight treat the outcome as either continuous or categorical. We examine both outcomes in this study because regression models of birth weight as a continuous variable will tend toward the mean weight of the sample. LBW usually represents the lower 5–10% of the distribution of birth weights and will not be properly represented as a continuous dependent variable. We use generalized linear models to analyze birth weight as a categorical variable (<2500 g is low birth weight and >=2500 g is healthy birth weight) in our study. We also use standard linear regression for the analysis of birth weight where birth weight is kept as a continuous outcome variable. We use cluster-adjusted standard errors (Moulton, 1986; Wooldridge, 2003) to reflect the assumption that the birth weights of babies born within the same sampling cluster are not independent. Within-cluster correlation could result from locally relevant factors related to road-networks, disease outbreak, health-care facilities, community cultures related to pregnancy and other factors that are unmeasured. This estimation procedure allows the error terms to be heteroskedastic and correlated within sampling clusters but assumes that, after controlling for observed variables, the errors are independent across differing countries and sample years. This corrects for unobserved factors at the cluster-level that may influence the birth weight of children in that cluster. Statistical controls have been used in an attempt to isolate the effect of weather on birth weight, however un-measurable characteristics, such as genetic predisposition and in-utero environment, still exist. To control for unobserved factors shared between siblings that contribute to LBW, including shared genetics, we use mother fixed-effect model and control only for characteristics that differ between siblings. In this approach we assume that unobserved heterogeneity is family-specific and discordant siblings from the same family are compared. DHS data allow us to leverage this method because some women report multiple births over the 5-year period. We use conditional logistic regression to compare multiple births from the same woman. In summary, we estimate four categories of models: (1) baseline model with only the household and individual characteristics (2) models that include precipitation and the count of hot days in addition to the factors in the baseline model (3) models that use livelihood zone as fixed-effects in addition to the baseline, precipitation and hot days factors (4) mother fixed effect models to control for unobserved heterogeneity. All factors that do not vary by mother are conditioned out of the equation, therefore the independent variables are the individual-level factors that differ for children born to the same mother, including the count of hot days and precipitation. Models were run for all sibling sets with discordant outcomes and then run separately by livelihood zone. All of the models are estimated for the categorical LBW outcome as well as for the continuous LBW outcome. Technical empirical model specifications can be found in the Appendix.

K. Grace et al. / Global Environmental Change 35 (2015) 125–137

131

Fig. 2. Temperature and precipitation variation.

4. Results As described above we constructed temperature and precipitation variables based on the specific growing seasons of each

country – the total precipitation during the growing season by trimester and total number of days above a specified temperature during the growing season by trimester. Focusing on the growing season precipitation and temperature characteristics helps to

K. Grace et al. / Global Environmental Change 35 (2015) 125–137

132

determine the growing season conditions as they relate to food production and food availability. However, because precipitation and temperature may have an impact on birth weight apart from their impacts on food production, we also included measures of average precipitation and hot days regardless of the overlap with the growing season calendar. We examined both types of climate variables in the regression analyses. Ultimately, because we already adjust for birth month as well as birth seasonality there was little statistical difference in the regression results when using the growing season climate variables versus the non-growing season variables. Because precipitation during the growing season is heavily associated with food production (Grace et al., 2012) and heat stress/high temperatures can occur at any point, here we present only the growing season results for precipitation and the number of hot days, regardless of season, for the temperature variables. We do note the potential for correlation between the temperature and precipitation variables and paid close attention to bivariate correlations between temperature and precipitation. All bivariate comparisons of correlation were at or below the level of 0.2. We also monitored standard errors for any unexpected or large changes, as this would indicate potential correlation. Ultimately, in the models presented here, no notable signs of correlation between the precipitation and temperature variables appeared. Beginning with the continuous regression (see Table 3), we see the anticipated results – a positive relationship between birth weight and growing season precipitation and negative relationship between birth weight and hot days. Higher amounts of precipitation indicate larger birth weights, and this result is significant for all trimesters On average, a 10 mm increase in precipitation during a particular trimester corresponds to an approximate increase in birth weight of around 0.3–.5 g (using Model 2 results). When the number of hot days is included in the model (Model 3 results) an

increase in the number of days above 100 F during any trimester, corresponds to a decrease in birth weight. One extra day with a temperature above 100 F in the trimester 2, for example, corresponds to a 0.9 g decrease in weight. Although not shown here, this result holds with a larger effect when we increase the temperature threshold for the count of hot days to 105 F. When included together in a regression model (Model 4 results), growing season precipitation and the count of hot days remain significant with essentially the same effect size as when they are included in a model alone. Precipitation during trimesters 0 and 1 and hot days during trimester 3, however, are no longer significant in Model 4. While the overall climate effects may seem modest when taken in isolation, if precipitation decreases in all trimesters and is coupled with a rise in the number of hot days, we would observe notable changes in birth weight outcomes. We consider the combined effects of these warming and drying scenarios in contrast to changes in socio-economic characteristics in the final section of the paper. Next, we compare separate models based on livelihood zone. The motivation behind the separate models is as follows: the dependence on precipitation for food production, and the resulting variation in vulnerability to food security will vary according to livelihood practice. For example, agriculturalists and agropastoralists will generally be more dependent on precipitation for food production than those who live in fishing areas. Further, our large sample size supports the development of livelihood-specific models and allows us to avoid analyses reliant on complex interaction effects. The results of these analyses suggest that increase in the number of hot days is an important correlate of birth weight outcomes. When evaluating the number of hot days and growing season precipitation with regard to livelihood zone (Table 3a), the number of hot days is significant and only for the

Table 3 Regression examining the effects of temperature and precipitation during the year preceding birth on birth weight (continuous). Model 1 Coef. Independent variable Child’s sex Mother’s age Birth order Marital status (married/partnered) Mother’s education (none) Primary Secondary + Birthweight recall (yes) Urban residence (yes) Electricity (yes) Floor material (finished) Unfinished Not recorded Other Unfinished wood Climate variables Growing season precipitation tri-0 tri-1 tri-2 tri-3 Number of days above 100 F tri-0 tri-1 tri-2 tri-3 Born during growing season (yes) N R2 Adjusted R2

117.63*** 0.91** 22.66** 53.24***

Model 2 SE

Coef. 5.329 0.642 1.901 8.503

117.61*** 0.87 22.61*** 53.98***

Model 3 SE

Coef. 5.328 0.642 1.901 8.503

117.65*** 0.97 22.69*** 53.96***

Model 4 SE

Coef. 5.327 0.642 1.901 8.502

117.64*** 0.87 22.63*** 53.60***

SE 5.327 0.645 1.900 8.502

23.48+ 12.55 51.07*** 0.42 33.04**

6.837 12.058 6.092 7.330 8.312

22.80*** 11.21 50.37*** 0.82 33.63*

6.835 12.060 6.100 7.329 8.312

21.50** 10.54 50.34*** 1.17 32.17***

6.839 12.058 6.096 7.332 8.311

22.74*** 11.00 50.33*** 1.72 33.30***

6.839 12.059 6.100 7.333 8.311

17.56* 23.95 68.41 54.45+

6.996 68.134 49.566 31.954

17.29* 24.89 70.25 64.14*

6.995 68.127 49.562 31.945

15.53* 27.70 73.42 54.43+

7.002 68.125 49.553 31.926

17.49* 24.58 69.08 63.46*

7.002 68.123 49.553 31.940

0.02 0.02 0.03+ 0.04* 0.45+ 0.71** 0.85*** 0.24 9.234 67,768 0.04 0.04

0.016 0.016 0.016 0.017 0.221 0.221 0.225 0.221

0.03+ 0.03** 0.04** 0.05**

18.28* 67,768 0.04 0.04

8.830 67,768 0.04 0.04

0.016 0.016 0.016 0.017

10.99+

0.53* 0.63** 0.90*** 0.44+ 6.325 67,768 0.04 0.04

0.218 0.212 0.224 0.212 15.87+

Note: Birth month and country are included in the analysis as fixed effects. The coefficients are not included here because of space constraints. Significance is indicated as follows: *** for p < 0.001, ** for p < 0.01, * for p < 0.05 and ‘+’ for p < 0.10. SE indicates standard error.

K. Grace et al. / Global Environmental Change 35 (2015) 125–137

133

Table 3a Regression examining the effects by livelihood of temperature and precipitation during the preceding year on birth weight (continuous). Agriculturalists

Coef.

Agropastoralists, Pastoralists, Urban, Fishers and Irrigated*

Agriculturalists

Agropastoralists, Pastoralists, Urban, Fishers and Irrigated*

SE

Coef.

SE

Coef.

SE

Coef.

SE

Growing season total precipitation 0.01 Trimester 0 Trimester 1 0.01 Trimester 2 0.004 Trimester 3 0.03

0.025 0.025 0.026 0.027

0.01 0.01 0.01 0.01

0.029 0.029 0.029 0.031

0.01 0.01 0.01 0.02

0.025 0.025 0.025 0.027

0.02 0.004 0.002 0.01

0.028 0.029 0.029 0.031

Number of days above 100 F Trimester 0 0.31 Trimester 1 0.16 Trimester 2 0.02 Trimester 3 0.32

0.416 0.418 0.425 0.415

0.44 0.73* 0.81* 0.58+

0.339 0.337 0.343 0.341

105 F 0.87 0.24 0.49 0.13

0.931 0.969 0.917 0.906

0.76 1.30* 1.46** 1.70**

0.549 0.527 0.523 0.535

Note: The results are based on models that include all of the covariates as found in Model 4 from Table 3. Significance of the other covariates remains as presented in the previous table. SE indicates standard errors. * In the regression model categories for different livelihood zones (excluding agriculturalists) are included.

non-agriculturalist livelihood groups. More days above 100 F during the trimesters 1–3 are correlated to a lower birth weight. This result has a larger effect when we look at the number of days above 105 F. We observe no statistically significant impact of growing season precipitation on birth weight for any of the livelihood groups. We also explored interactions between precipitation and the number of hot days because the effects of drought usually result from a combination of both high temperature and low precipitation. However, none of these interactions were statistically significant. In Table 4, we present the results for the categorical birth weight analysis. Here, LBW is coded as 0 for weights less than 2500 g and healthy birth weights are coded as 1. After adjusting for the control variables, the growing season precipitation during all

trimesters is positive and significant. In terms of temperature, an increase in the number of hot days above 100 F corresponds to an increase in the likelihood of delivering an LBW baby. Again, these results hold when using the 105 F threshold. When the count of hot days and precipitation are included together, the effect sizes and significance values change somewhat, with early trimester (0 and 1) precipitation maintaining significance and trimester 1 and 2 hot days maintaining significance. The results suggest that the combined impact of low precipitation and high numbers of hot days together can significantly increase the likelihood of delivering an LBW infant. In Table 4a, we evaluate growing season precipitation and the number of hot days with attention to livelihood zone for the

Table 4 Regression examining the effects of temperature and precipitation during the year preceding birth on birth weight (categorical). Model 1

Independent variable Child’s sex (male) Mother’s age Birth order Marital status (married/partnered) Mother’s education (none) Primary Secondary + Birthweight recall (yes) Urban residence (yes) Electricity (yes) Floor material (finished) Unfinished Not recorded Other Unfinished wood Climate variables Growing season precipitation tri-0 tri-1 tri-2 Number of days above 100 F tri-0 tri-1 tri-2 tri-3 Born during growing season (yes) AIC N

Model 2

Model 3

Model 4

Coef.

SE

Coef.

SE

Coef.

SE

Coef.

SE

0.31*** 0.001 0.06** 0.25***

0.026 0.003 0.010 0.039

0.31*** 0.001 0.06** 0.24***

0.026 0.003 0.010 0.039

0.31*** 0.001 0.06** 0.25***

0.026 0.003 0.010 0.039

0.31*** 0.001 0.06** 0.25***

0.026 0.003 0.010 0.039

0.16*** 0.29*** 0.30*** 0.09* 0.05

0.034 0.065 0.030 0.036 0.041

0.16*** 0.28*** 0.30*** 0.09* 0.06

0.034 0.065 0.030 0.036 0.041

0.15*** 0.28*** 0.30*** 0.08* 0.05

0.034 0.065 0.030 0.036 0.041

0.16*** 0.28*** 0.30*** 0.08* 0.05

0.034 0.065 0.030 0.036 0.041

0.13*** 0.35 0.15 0.08

0.034 0.282 0.290 0.148

0.13*** 0.35 0.16 0.11

0.037 0.282 0.290 0.149

0.12*** 0.32 0.17 0.08

0.037 0.283 0.290 0.148

0.12*** 0.32 0.16 0.11

0.037 0.283 0.290 0.149

0.02** 0.02* 0.01*

0.008 0.008 0.008

0.02+ 0.02** 0.01 0.17 0.20+ 0.33** 0.17 0.02 42,818 67,768

0.008 0.008 0.008 0.096 0.098 0.099 0.098 0.045

42,850 67,768

0.06 42,838 67,768

0.045

0.18+ 0.30** 0.37*** 0.18+ 0.03 42,822 67,768

0.099 0.100 0.102 0.100 0.047

Note: Birth month and country are included in the analysis as fixed effects. The coefficients are not included here because of space constraints. Significance is indicated as follows: *** for p < 0.001, ** for p < 0.01, * for p < 0.05 and ‘+’ for p < 0.10. SE indicates standard error. SE and coefficients for precipitation and temperature are multiplied by 100.

K. Grace et al. / Global Environmental Change 35 (2015) 125–137

134

Table 4a Regression examining the effects by livelihood of temperature and precipitation during the preceding year on birth weight (categorical). Agriculturalists

Coef.

SE

Agropastoralists, Pastoralists, Urban, Fishers and Irrigated*

Agriculturalists

Coef.

Coef.

SE

Growing season total precipitation 0.02+ Trimester 0 Trimester 1 0.02 Trimester 2 0.01 Trimester 3 0.01

0.013 0.012 0.013 0.013

0.01 0.04* 0.01 0.01

0.014 0.014 0.014 0.015

Number of Days above 100 F Trimester 0 0.11 Trimester 1 0.06 Trimester 2 0.15 Trimester 3 0.12

0.192 0.191 0.195 0.192

0.19 0.18 0.19 0.30*

0.144 0.144 0.147 0.146

0.02+ 0.02 0.01 0.01 105 F 0.39 0.04 0.31 0.19

Agropastoralists, Pastoralists, Urban, Fishers and Irrigated* SE

Coef.

SE

0.013 0.012 0.013 0.013

0.003 0.03* 0.005 0.005

0.013 0.014 0.014 0.014

0.431 0.439 0.405 0.402

0.36 0.02 0.38+ 0.49*

0.215 0.214 0.21 0.212

Note: The results are based on models that include all of the covariates as found in Model 4 from Table 4. Significance of the other covariates remains as presented in the previous table. SE indicates standard errors. SE and coefficients for precipitation and temperature are multiplied by 100. * In the regression model categories for different livelihood zones (excluding agriculturalists) are included.

categorized birth weight. In these results, an increase in growing season precipitation during trimester 0 reduces the likelihood of a baby being born as LBW for agriculturalists. An increase in growing season precipitation during the first trimester reduces the likelihood of low birth weight for the other livelihood groups. The number of hot days, in general has a negative relationship with birth weight outcomes. An increase in the number of days over 105 F during trimesters 0, 2 and 4 for non-agriculturalists resulted in an increase in risk of LBW. The LBW status of infants born in agricultural zones seems to be little impacted by increases in the number of hot days (either 100 F or 105 F). High temperatures for extended periods during the pregnancy may physically stress the mother, and therefore the fetus. To test this hypothesis, exposure was defined as 45 or more days of in utero exposure to extreme heat (>=1058F). Pearson Chi-square tests were used as a simple test for an association between exposure to heat stress and birth weight for the overall sample as well as stratified by livelihood zone. A mother fixed effect design was used to further investigate the relationship between ambient temperature and low birth weight. This design is particularly powerful because all confounding shared by the siblings, environmental and genetic, is controlled for by making comparisons across siblings, within mothers. Exposed individuals were matched to an unexposed sibling and models controlled for mother’s age at the time of the birth, birth order, precipitation, malaria medication during pregnancy, and ever having a prenatal visit. All models were run for the full sample and by livelihood zone. Overall, we found no association between the number of hot days and birth weight in the full-sample or when stratified by livelihood zone. The exception to this result was for those mothers living in agricultural areas exposed to prolonged heat during trimester 0 had decreased odds of giving birth to a LBW baby. While we found little association between long periods of exposure and LBW, it is still possible that acute exposures to heat lead to spontaneous preterm birth. Malaria is an important health challenge for many countries and has been linked to precipitation patterns and to infant health issues, including low birth weight. DHS data are not consistent in their reporting of malaria information, however, where malaria data and pre-natal care data were included we conducted separate regression analysis (based on the models in Tables 3 and 4) to evaluate the relationship between malaria status and low birth weight. Because use of pre-natal care may also indicate the potential for treatment of malaria, we evaluated the LBW outcomes of mothers who have at least some prenatal care, as

opposed to those who had none. Using either of these variables in a regression model reduced the sample size dramatically, yet the relationship between precipitation or temperature and LBW was unaffected by either the malaria variable or the pre-natal care variable. We do note that because of the large selection of independent variables, some of which may capture some aspects of variability found in other present independent variables, that there may be some autocorrelation. Given the potential for correlation among variables, close attention to descriptive analyses and standard errors, variance inflation factors and covariance was paid during the modeling stage. No evidence of strong collinearity that would dramatically impact the results was found using these methods.

5. Discussion Given patterns of global climate change, interest in the relationship between weather and health outcomes has increased. One health outcome that may be modified by climate patterns and that has important short- and long-term implications for lowincome countries is low birth weight. In this study we combined data from 19 African countries with fine scale precipitation and temperature data to explore the relationship between weather and birth weight. We focused the analysis on some of the poorest and most climatically vulnerable countries in the world. The results of the research suggest that temperature and precipitation are significant factors in birth weight and LBW outcomes. While not consistent across all trimesters and in all situations, our results suggest that increases in the number of hot days and decreasing precipitation correlate to lower birth weights and higher rates of LBW in sub-Saharan Africa. The relationship across multiple trimesters and for continuous and categorical measures of birth weight suggests that the mechanisms driving the association may be related to intrauterine growth and not just heat related spontaneous preterm birth. In terms of LBW specifically, growing season precipitation and the number of hot days are both significant factors. During early periods of pregnancy, including the 3 months prior to conception (trimester 0) growing season precipitation decreases the risk of LBW increases (Table 4, Model 2). This outcome occurs regardless of household wealth indicators, the season of birth, or the country of residence. The impact of the number of hot days at any point during pregnancy leads to an increase in the risk of LBW births (Table 4, Model 3).

K. Grace et al. / Global Environmental Change 35 (2015) 125–137

The relationship between precipitation and birth weight seems primarily to be related to food production, where precipitation influences the amount of food that is or will be available at some point during the pregnancy. The countries involved in our study depend heavily on rainfed agriculture because sophisticated irrigation systems are uncommon. Therefore high amounts of precipitation during the early stages of pregnancy will likely result in more food, more farm-related incomes, or lower prices during the later stages of pregnancy. The relationship with temperature may be more related to heat stress in general and not specifically to food production. Heat stress and even smaller temperature increases has been shown to have long-term impacts on the health of children The mother fixed effects models provide a different perspective on a smaller sample of our data. However, because the sample used in this analysis is very select – at least two births in the five years prior to the survey and residence in the same community for the past five years – there are limitations in how the results can be compared to the larger sample used in this study. In this comparison of siblings, weather variation is not a consistent correlate of birth weight outcomes. Combining this result with the larger sample’s regression results suggest that precipitation and livelihood are important components of LBW. In other words, arid environments likely produce more LBW infants, but it may not simply be the lack of precipitation that creates these birth outcomes. Instead, it may be that women who live in arid environments are more likely to give birth to LBW children for other reasons – culture, family background, prolonged exposure to inadequate nutrition or other factors. Because the mother fixed effect analysis is based on an extremely small sample size, further research, especially life-course oriented studies, of food insecurity and reproduction in the arid environments of Africa is required to provide greater insight into food insecurity and birth outcomes. 5.1. What do the results of this research indicate in terms of climate change and climate scenarios? Much of the contemporary research of climate and health outcomes is motivated by an interest in quantifying the human impacts of climate change. In this case our results support two important conclusions: (1) climate change – increasing temperatures and decreasing precipitation – will likely have an impact (albeit small) on birth weight outcomes, even possibly causing an increase in LBW and (2) the impact of climate change on birth weight and LBW will depend on where a pregnant woman lives. In other words, the relationship between climate and birth weight or LBW is not consistent for all mothers but likely reflects the local food production, storage strategies and, possibly, heat coping strategies. While the overall variance explained by our models was low, and sensitivity tests (Table 5) indicate only modest weatherrelated changes in birth weight – a 4.3% decrease in average birth weight under a scenario of combined warming and drying. The magnitude of these influences is similar, or greater, as compared to important socio-economic indicators, like electricity status or education level. Given the relationships shown here and since wide-spread warming is consistently predicted by the most recent generation of climate change models (ref IPCC WGII report) temperature influences may pose the greater overall threat. In summary, if temperature increases, precipitation decreases, and food availability is unstable, our results suggest that we will likely see an increase in the number of LBW babies. This effect has the potential to reverberate across generations because LBW girls are more likely to give birth to their own LBW infants, and because LBW has negative consequences for short- and long-term health and economic development. Our findings suggest the potential for long-term climate-related health challenges for the poorest countries in the world. In a future paper, we will further explore

135

Table 5 Scenarios analysis.

Temperature increaseb Precipitation decreasea Precipitation decrease  temperature increase Increase education and add electricity Increase education, add electricity, temperature increase  precipitation decrease

Mean predicted birth weight (using model 3) (g)

Mean predicted birth weight (using scenario) (g)

Percent change in birth weight

3217.37 3217.37 3217.37

3082.49 3211.87 3076.99

4.19 .17 4.36

3217.37

3236.18

.58

3217.37

3095.80

3.78

a Represented by a 100 mm decline in total average rainy season precipitation during term. b Represented by a 50% increase the count of days during term greater than or equal to 100 8F.

how weather induced impacts to child health may vary across specific countries under a variety of climate change and development scenarios.

Appendix Our specification for the continuous outcome model is as follows. This model without precipitation and hot days is the baseline model. The logit form of the model is used when estimating the categorical ‘‘low birth weight’’ outcome:

birthweightðiÞ ¼ b0 þ b1 IðcÞ þ b2 IðmÞ þ b3 IðrsÞ þ X u ðkÞ þ RaðtriÞ þ T g ðtriÞ þ eðiÞ ; where birthweight(i) is the birthweight (in grams) of child i. The n T 1 vectors Ic, Im, and Irs, are dummy variables indicating the survey country, month of birth, and if the child was born in the rainy season. The n T k matrix X includes controls for characteristics of the mother, child, and household (described in Section 2.1). The n  4 matrix R contains monthly precipitation for the four trimesters prior to the child’s birth. The n  4 matrix T stores the counts of days over 100 F during the same four trimesters. The main parameters of interest are the 4  1 vectors atri and gtri which contain the estimated coefficients for precipitation and hot days variables. Finally, the n  1 vector ei is an unobserved heteroskedastic error term. We also fit the same model using livelihood zone as a dummy coded variable. Our specification for the mother fixed effects model is as follows:

logitðnormal BW ji Þ ¼ ai þ b1 I jiðrsÞ þ b2 X jiðkÞ þ b3 R jiðtriÞ þ b4 T jiðtriÞ ; where j = 1, 2, . . ., m denotes the mother and i = 1, 2, . . ., Ij denotes the individual for the jth mother. The normal BWjt is the indicator of normal birth weight for individual i mother j; aj is the mother specific intercept which is conditioned out of the likelihood equation; Iji(rs) is a dummy variable indicating if the child was born in the rainy season; the n  k matrix X includes controls for malaria medication during pregnancy, and prenatal care. The n  4 matrix R contains monthly precipitation for the four trimesters prior to the child’s birth. The n  4 matrix T stores the counts of days over 100 F during the same four trimesters. These models are estimated for the entire sample and separately for each livelihood zone.

136

K. Grace et al. / Global Environmental Change 35 (2015) 125–137

References Abu-Saad, K., Frasier, D., 2010. Maternal nutrition and birth outcomes. Epidemiol. Rev. 32, 5–25. Alderman, H., Behrman, J.R., 2006. Reducing the incidence of low birth weight in low-income countries has substantial economic benefits. World Bank Res. Obs. 21 (1), 25–48. Bantje, H., 1987. Seasonality of births and birthweights in Tanzania. Soc. Sci. Med. 24 (9), 733–739. Bantje, H., Niemeyer, R., 1984. Rainfall and birthweight distribution in rural Tanzania. J. Biosoc. Sci. 16 (3), 375–384. Basu, R., Malig, B., Ostro, B., 2010. High ambient temperature and the risk of preterm delivery. Am. J. Epidemiol. 172 (10), 1108–1117. Berman, M.R., et al., 1982. Temperature dependence of CH radical reactions with O 2, NO, CO and CO2. In: Symposium (International) on Combustion, 19 (1). Elsevier. Bhutta, A.T., et al., 2002. Cognitive and behavioral outcomes of school-aged children who were born preterm: a meta-analysis. JAMA 288 (6), 728–737. Boerma, J.T., Weinstein, K.I., Rutstein, S.O., Sommerfelt, A.E., 1996. Data on birth weight in developing countries: can surveys help? Bull. World Health Organ. 74 (2), 209. Borders, A.E.B., Grobman, W.A., Amsden, L.B., Holl, J.L., 2007. Chronic stress and low birth weight neonates in a low-income population of women. Obstet. Gynecol. 109 (2 Part 1), 331–338. Caldwell, J.C., 1979. Education as a factor in mortality decline an examination of Nigerian data. Popul. Stud. 395–413. Caldwell, J.C., 1994. How is greater maternal education translated into lower child mortality? Health Transit. Rev. 224–229. Cooperstock, M., Wolfe, R.A., 1986. Seasonality of preterm birth in the Collaborative Perinatal Project: demographic factors. Am. J. Epidemiol. 124 (2), 234–241. Cosgrove, B.A., Lohmann, D., Mitchell, K.E., Houser, P.R., Wood, E.F., Schaake, J.C., Robock, A., Marshall, C., Sheffield, J., Duan, Q., Luo, L., Higgins, R.W., Pinker, R.T., Tarpley, J.D., Meng, J., 2003. Real-time and retrospective forcing in the North American Land Data Assimilation System (NLDAS) project. J. Geophys. Res. 108 (D22), 8842, http://dx.doi.org/10.1029/2002JD003118. Darlow, B.A., et al., 2005. Prenatal risk factors for severe retinopathy of prematurity among very preterm infants of the Australian and New Zealand Neonatal Network. Pediatrics 115 (4), 990–996. Descheˆnes, O., Greenstone, M., Guryan, J., 2009. Climate change and birth weight. Am. Econ. Rev. 211–217. Elter, K., et al., 2004. Exposure to low outdoor temperature in the midtrimester is associated with low birth weight. Aust. N. Z. J. Obstet. Gynaecol. 44 (6), 553–557. FAO, 2006. World Agriculture: Towards 2030/2050 – Interim Report. Global Perspective Studies Unit. FAO, Rome, Italy. Farooqi, A., et al., 2006. Growth in 10-to 12-year-old children born at 23 to 25 weeks’ gestation in the 1990s: a Swedish national prospective follow-up study. Pediatrics 118 (5), e1452–e1465. Flouris, A.D., et al., 2009. Short report effect of seasonal programming on fetal development and longevity: links with environmental temperature. Am. J. Hum. Biol. 21, 214–216. Fraser, E.D.G., 2006. Food system vulnerability: using past famines to help understand how food systems may adapt to climate change. Ecol. Complex. 3, 328–335. Funk, C., Michaelsen, J., Marshall, M., 2012. Mapping recent decadal climate variations in Eastern Africa and the Sahel. In: Wardlow, B., Anderson, M., Verdin, J. (Eds.), Remote Sensing of Drought: Innovative Monitoring Approaches. Funk, C., Peterson, P., Landsfeld, M., Pedreros, D., Verdin, J., Rowland, J., Romero, B., Husak, G., Michaelsen, J., Verdin, A., 2014a. A Quasi-global Precipitation Time Series for Drought Monitoring. USGS Data Series 832, http://pubs.usgs. gov/ds/832/pdf/ds832.pdf. Funk, C., Hoell, A., Shukla, S., Blade´, I., Liebmann, B., Roberts, J.B., Husak, G., 2014b. Predicting East African spring droughts using Pacific and Indian Ocean sea surface temperature indices. Hydrol. Earth Syst. Sci. Discuss. 11 (3), 3111–3136. Funk, C., Verdin, A., Michaelsen, J., Peterson, P., Pedreros, D., Husak, G., 2015. A global satellite assisted precipitation climatology. Earth Syst. Sci. Data. Godfrey, K.M., Barker, D.J.P., Robinson, S., Osmond, C., 1997. Maternal birthweight and diet in pregnancy in relation to the infant’s thinness at birth. BJOG: Int. J. Obstet. Gynaecol. 104 (6), 663–667. Grace, K., Davenport, F., Funk, C., Lerner, A., 2012. Child malnutrition and climate conditions in Kenya. Appl. Geogr. 35 (1), 405–413. Grace, K., Brown, M., McNally, A., 2014. Examining the link between food price and food insecurity: a multi-level analysis of maize price and birth weight in Kenya. Food Policy 46, 56–65. Hack, M., et al., 2005. Chronic conditions, functional limitations, and special health care needs of school-aged children born with extremely low-birthweight in the 1990s. JAMA 294 (3), 318–325. Haggarty, P., Campbell, D.M., Duthie, S., Andrews, K., Hoad, G., Piyathilake, C., McNeill, G., 2009. Diet and deprivation in pregnancy. Br. J. Nutr. 102 (10), 1487–1497. Harlan, S.L., Brazel, A.J., Prashad, L., Stefanov, W.L., Larsen, L., 2006. Neighborhood microclimates and vulnerability to heat stress. Soc. Sci. Med. 63 (11), 2847–2863.

Hille, E.T.M., et al., 2007. Functional outcomes and participation in young adulthood for very preterm and very low birth weight infants: the Dutch Project on Preterm and Small for Gestational Age Infants at 19 years of age. Pediatrics 120 (3), e587–e595. Hobcraft, J.N., 1993. Women’s education, child welfare and child survival: a review of the evidence. Health Transit. Rev. 3, 159–175. Hoddinott, J., Kinsey, B., 2001. Child growth in the time of drought. Oxf. Bull. Econ. Stat. 63 (4), 409–436. Hort, K.P., 1987. Seasonal variation of birthweight in Bangladesh. Ann. Trop. Paediatr. 7 (1), 66–71. Jones, P.G., Thornton, P.K., 2003. The potential impacts of climate change on maize production in Africa and Latin America in 2055. Glob. Environ. Change 13 (1), 51–59. Juli, R., Duchin, F., 2007. World trade as the adjustment mechanism of agriculture to climate change. Clim. Change 82, 393–409. Kalnay, E., Kanamitsu, M., Kistler, R., Collins, W., Deaven, D., Gandin, L., Iredell, M., Saha, S., White, G., Woollen, J., Zhu, Y., Leetmaa, A., Reynolds, R., Chelliah, M., Ebisuzaki, W., Higgins, W., Janowiak, J., Mo, K.C., Ropelewski, C., Wang, J., Jenne, R., Joseph, D., 1996. The NCEP/NCAR 40-Year Reanalysis Project. Bull. Am. Meteorol. Soc. 77 (3), 437–471, http://dx.doi.org/10.1175/ 1520-0477(1996)077<0437:TNYRP>2.0.CO;2. Keller, C.A., Nugent, R.P., 1983. Seasonal patterns in perinatal mortality and preterm delivery. Am. J. Epidemiol. 118 (5), 689–698. Kistler, R., Collins, W., Saha, S., White, G., Woollen, J., Kalnay, E., Chelliah, M., Ebisuzaki, W., Kanamitsu, M., Kousky, V., van den Dool, H., Jenne, R., Fiorino, M., 2001. The NCEP–NCAR 50-year reanalysis: monthly means CD–ROM and documentation. Bull. Am. Meteorol. Soc. 82 (2), 247–267, http://dx.doi.org/ 10.1175/1520-0477(2001)082<0247:TNNYRM>2.3.CO;2. Kramer, M., 1987. Determinants of low birth weight: methodological assessment and meta-analysis. Bull. World Health Organ. 65 (5), 663–737. Lawlor, D.A., Leon, D.A., Smith, G.D., 2005. The association of ambient outdoor temperature throughout pregnancy and offspring birthweight: findings from the Aberdeen Children of the 1950s cohort. BJOG: Int. J. Obstet. Gynaecol. 112 (5), 647–657. Lunney, L.H., 1998. Compensatory placental growth after restricted maternal nutrition in early pregnancy. Placenta 19 (1), 105–111. Marchant, T., Willey, B., Katz, J., Clarke, S., Kariuki, S., Kuile Ft, et al., 2012. Neonatal mortality risk associated with preterm birth in East Africa, adjusted by weight for gestational age: individual participant level metaanalysis. PLoS Med. 9 (8), e1001292, http://dx.doi.org/10.1371/ journal.pmed.1001292. Marlow, N., 2004. Neurocognitive outcome after very preterm birth. Arch. Dis. Child. Fetal Neonatal Ed. 89 (3), F224–F228. McGrath, J.J., et al., 2005. The association between birth weight, season of birth and latitude. Ann. Hum. Biol. 32, 547–559. McGregor, I.A., Wilson, M.E., Billewicz, W.Z., 1983. Malaria infection of the placenta in The Gambia, West Africa; its incidence and relationship to stillbirth, birthweight and placental weight. Trans. R. Soc. Trop. Med. Hyg. 77 (2), 232–244. Moulton, B.R., 1986. Random group effects and the precision of regression estimates. J. Econom. 32 (3), 385–397. Murray, L.J., O’Reilly, D.P.J., Betts, N., Patterson, C.C., Smith, G.D., Evans, A.E., 2000. Season and outdoor ambient temperature: effects on birth weight. Obstet. Gynecol. 96 (5 Part 1), 689–695. Mwabu, G., 2008. The production of child health in Kenya: a structural model of birth weight. J. Afr. Econ. 18 (2), 212–260. New, M., Hulme, M., Jones, P., 1999. Representing twentieth-century space–time climate variability. Part I: Development of a 1961–90 mean monthly terrestrial climatology. J. Clim. 12 (3) . New, M., Hulme, M., Jones, P., 2000. Representing twentieth-century space–time climate variability. Part II: Development of 1901–96 monthly grids of terrestrial surface climate. J. Clim. 13 (13) . Newman, R.D., Hailemariam, A., Jimma, D., Degifie, A., Kebede, D., Rietveld, A.E.C., Nahlen, B.L., Barnwell, J.W., Steketee, R.W., Parise, M.E., 2003. Burden of malaria during pregnancy in areas of stable and unstable transmission in Ethiopia during a nonepidemic year. J. Infect. Dis. 187 (11), 1765–1772. Okoko, B.J., Ota, M.O., Yamuah, L.K., Idiong, D., Mkpanam, S.N., Avieka, A., Banya, W.A., Osinusi, K., 2002. Influence of placental malaria infection on foetal outcome in the Gambia: twenty years after Ian Mcgregor. J. Health Popul. Nutr. 20 (1), 4–11. Prentice, I.C., Sykes, M.T., Cramer, W., 1993. A simulation model for the transient effects of climate change on forest landscapes. Ecol. Model. 65 (1), 51–70. Rayco-Solon, P., Fulford, A.J., Prentice, A.M., 2005. Differential effects of seasonality on preterm birth and intrauterine growth restriction in rural Africans. Am. J. Clin. Nutr. 81 (1), 134–139. Rousham, E.K., Gracey, M., 1998. Seasonality of low birthweight in indigenous Australians: an increase in pre-term birth or intrauterine growth retardation? Aust. N. Z. J. Public Health 22 (6), 669. Ruff, A.J., 1994. Breastmilk, breastfeeding, and transmission of viruses to the neonate. Semin. Perinatol. 18 (6) . Rutstein, S.O., 2008. Further evidence of the effects of preceding birth intervals on neonatal, infant, and under-five-years mortality and nutritional status in developing countries: Evidence from the demographic health surveys. DHS Working Paper No. 41United States Agency for International Development, Washington, DC.

K. Grace et al. / Global Environmental Change 35 (2015) 125–137 Seidman, D.S., Ever-Hadani, P., Stevenson, D.K., et al., 1988. Birth order and birth weight reexamined. Obstet. Gynecol. 72, 158–162. Sheffield, J., Wood, E.F., 2008a. Global trends and variability in soil moisture and drought characteristics, 1950–2000, from observation-driven simulations of the terrestrial hydrologic cycle. J. Clim. 21 (3), 432–458. Sheffield, J., Wood, E.F., 2008b. Projected changes in drought occurrence under future global warming from multi-model, multi-scenario, IPCC AR4 simulations. Clim. Dyn. 31 (1), 79–105. Sheffield, J., Goteti, G., Wood, E.F., 2006. Development of a 50-year highresolution global dataset of meteorological forcings for land surface modeling. J. Clim. 19 (13), 3088–3111, http://dx.doi.org/10.1175/JCLI3790.1. Sheffield, J., Wood, E.F., Chaney, N., Guan, K., Sadri, S., Yuan, X., Olang, L., Amni, A., Ali, A., Demuth, S., 2013. A drought monitoring and forecasting system for Sub-Sahara African water resources and food security. Bull. Am. Met. Soc., http://dx.doi.org/10.1175/BAMS-D-12-00124.1. Shukla, S., Sheffield, J., Wood, E.F., Lettenmaier, D.P., 2013. On the sources of global land surface hydrologic predictability. Hydrol. Earth Syst. Sci. 17 (7), 2781–2796, http://dx.doi.org/10.5194/hess-17-2781-2013. Singer, L.T., Salvator, A., Guo, S., Collin, M., Lilien, L., Baley, J., 1999. Maternal psychological distress and parenting stress after the birth of a very lowbirth-weight infant. JAMA 281 (9), 799–805. Strand, L., Barnett, A., Tong, S., 2011. The influence of season and ambient temperature on birth outcomes: a review of the epidemiological literature. Environ. Res. 111, 451–462. Swamy, G.K., Edwards, S., Gelfand, A., James, S.A., Miranda, M.L., 2012. Maternal age, birth order, and race: differential effects on birth weight. J. Epidemiol. Community Health 66, 136–142. Torche, F., 2011. The effect of maternal stress on birth outcomes: exploiting a natural experiment. Demography 48, 1473–1491.

137

Tote´, C., et al., 2015. Evaluation of satellite rainfall estimates for drought and flood monitoring in mozambique. Remote Sens. 7 (2), 1758–1776. Tustin, K., Gross, J., Hayne, H., 2004. Maternal exposure to first trimester sunshine is associated with increased birth weight in human infants. Dev. Psychobiol. 45 (4), 221–230. Uddenfeldt Wort, U., Hastings, I.M., Carlstedt, A., Mutabingwa, T., Brabin, B.J., ˜ o and malaria on birthweight in two areas of 2004. Impact of El Nin Tanzania with different malaria transmission patterns. Int. J. Epidemiol. 33 (6), 1311–1319. United Nations Children’s Fund and World Health Organization, 2004. Low birthweight: Country, regional and global estimates. UNICEF, New York. Victora, C., Adair, L., et al., 2008. Maternal and child undernutrition: consequences for adult health and human capital. Lancet 371 (9609), 340–357. Walker, S., Wachs, T., et al., 2007. Child development: risk factors for adverse outcomes in developing countries. Lancet 369 (9556), 145–157. Wang, J., et al., 2013. Maternal exposure to heatwave and preterm birth in Brisbane, Australia. BJOG: Int. J. Obstet. Gynaecol. 120 (13), 1631–1641. Wardlaw, Tessa, M., 2004. Low Birthweight: Country, Regional and Global Estimates. UNICEF. Wells, J.C.L., Cole, T.J., 2002. Birth weight and environmental heat load: a between population analysis. Am. J. Phys. Anthropol. 119, 276–282. Wooldridge, J.M., 2003. Cluster-sample methods in applied econometrics. Am. Econ. Rev. 93 (2), 133–138. World Food Summit, 1996. Rome Declaration on World Food Security. Young, H., 2001. Nutrition and intervention strategies. In: Devereux, Maxwell, (Eds.), Food Security in Sub-Saharan Africa. ITDG Publishing, London, pp. 231–264.