Menu
Menu
Menu
Menu
Pregnancy and birth outcomes have been found to be sensitive to meteorological variation, yet few studies explore this relationship in sub-Saharan Africa where infant mortality rates are the highest in the world. We address this research gap by examining the association between meteorological factors and birth weight in a rural population in southwestern Uganda. Our study included hospital birth records (n = 3197) from 2012 to 2015, for which we extracted meteorological exposure data for the three trimesters preceding each birth. We used linear regression, controlling for key covariates, to estimate the timing, strength, and direction of meteorological effects on birth weight. Our results indicated that precipitation during the third trimester had a positive association with birth weight, with more frequent days of precipitation associated with higher birth weight: we observed a 3.1g (95% CI: 1.0-5.3g) increase in birth weight per additional day of exposure to rainfall over 5mm. Increases in average daily temperature during the third trimester were also associated with birth weight, with an increase of 41.8g (95% CI: 0.6-82.9g) per additional degree Celsius. When the sample was stratified by season of birth, only infants born between June and November experienced a significant associated between meteorological exposures and birth weight. The association of meteorological variation with foetal growth seemed to differ by ethnicity; effect sizes of meteorological were greater among an Indigenous subset of the population, in particular for variation in temperature. Effects in all populations in this study are higher than estimates of the African continental average, highlighting the heterogeneity in the vulnerability of infant health to meteorological variation in different contexts. Our results indicate that while there is an association between meteorological variation and birth weight, the magnitude of these associations may vary across ethnic groups with differential socioeconomic resources, with implications for interventions to reduce these gradients and offset the health impacts predicted under climate change.
Our study sample consists of 3691 women who gave birth at Bwindi Community Hospital in Kanungu District, Uganda, between June 2012 and June 2015. Kanungu District is situated in southwestern Uganda, near the borders of the Democratic Republic of the Congo and Rwanda (Fig 1). The study area is situated in the region surrounding Bwindi Impenetrable National Park, which is inhabited predominantly by the Bakiga ethnic group as well as approximately 900 members (1%) of the Indigenous Batwa population [28, 29]. Both the Bakiga and Batwa have high burden of ill health relative to the national average, with particularly high inequality and burden of ill health among the Indigenous Batwa [28]. Evicted from their forest homes when the National Park was created in the early 1990s, the Batwa were forced to resettle in agrarian communities [28, 30, 31]. The existing burden of ill health among the Batwa has been characterized across a variety of metrics, including reduced life expectancy (28 years for the Batwa compared to the Ugandan average of 53 [28]), higher prevalence of malaria (9.4% for the Batwa compared to 4.5% in the Bakiga population [32]), acute gastrointestinal illness (compared to East Africa [33]), and extreme food insecurity [34]. The prevalence of HIV among the Batwa population is, however, lower than that in the Bakiga population [35]. Community-based health surveys in Kanungu District (in the same communities from which our hospital sample was derived) have established a significant correlation between ethnicity and indicators of socio-economic status (SES) (Table 1). Donnelly, Berrang-Ford [32] found significantly lower levels of education and asset ownership among Indigenous Batwa compared to the non-Indigenous population. Donnelly, Berrang-Ford [32] also show that ethnicity and SES had both independent and collinear effects on malaria infection, implying that ethnicity may act as a partial proxy for gradients in SES in this population. There is also a higher burden of malnutrition among Batwa women than among Bakiga women [36]. The study area is located northwest of Bwindi Impenetrable National Park, hemmed in by the border of the Democratic Republic of the Congo (DRC). *Prevalence in July 2013 and April 2014—survey of all Batwa adults, sample of Bakiga adults **Only asked of people that had access to hand washing facility, for example for the Batwa, 32 or 94% of the households that had access to handwashing had access to soap Research has documented high vulnerability to the health impacts of climate change within both the Batwa and Bakiga populations, and both groups have identified malaria, food insecurity, and gastrointestinal illnesses as climate-sensitive health concerns [28, 29]. Lack of access to health care and the strenuous nature of subsistence farming labour also give rise to substantial disparities in perinatal health in Kanungu District. Approximately 40% of births in the region occur in health facilities [37] compared to 57% country-wide [38]. This disparity is paralleled by similar metrics examining the presence of a skilled healthcare provider at delivery: 59% of all Ugandan infants are delivered by a skilled provider, compared to 42% of infants in the Southwest Region [37]. Bwindi Community Hospital (BCH) was established in 2003 as an outreach clinic for the Batwa, but has expanded into a full inpatient hospital serving 100 000 people across three sub-counties in Kanungu district [39]. The hospital is located in Buhoma trading centre, and also operates several satellite clinics in more remote settlements. The hospital’s antenatal clinic sees approximately 250 mothers per month and performs over 1000 deliveries each year [40]. BCH also provides antenatal care and family planning services. In 2008, BCH opened a ‘Waiting Mothers Hostel,’ where women who live at greater distances from the hospital can stay while they are waiting to give birth [41]. The climate in Kanungu includes two rainy seasons: the ‘short rains’ from October to December and the ‘long rains’ from March to May [42]. Average temperatures in the region are relatively cool (typically below 20°C) in comparison to the rest of the country, though there has been an increase in mean annual temperature of 1.3°C over the last fifty years [42]. Global climate models for the region predict increases in both mean annual temperature and heavy rain events [43, 44] along with increases in severe dry conditions in August and September [45] and trends of increasing drought [46]. Southwestern Uganda has been reported as the fastest warming region in the country [47]. Dominant livelihood activities in the region include agriculture, industrial tea and coffee production, and tourism [28, 29]. Birth outcomes were ascertained from the hospital records that were completed by nurses during labour and after birth. They include information on the mother’s medical history, intranatal measures and interventions, and assessments of the baby at birth. Birth weight in grams is our primary outcome. In the available birth record data from Bwindi Community Hospital (2012–2015), there were 3691 births, 3343 of which were singleton births. Of these, there were 3197 complete records with sufficient information to estimate gestational age. Gestational age (GA) was estimated for all singleton birth records (n = 3197) to determine the exposure windows of the meteorological variables and to include as a key control variable in our ‘base’ regression models for the primary outcome analysis. In the case of 591 observations (17.9%), gestational age was recorded based on ultrasound dating, the most reliable measurement standard for GA. The date of the ultrasounds were not collected, but we note that ultrasounds conducted later in pregnancy have a greater margin of error than those conducted within the first trimester, and that there is variation in the timing of initial presentation of pregnant women to the hospital or antenatal clinic. Where GA based on ultrasound was not available, last menstrual period (LMP, determined by maternal recall, n = 1742, 52.7%) and adjusted fundal height (n = 971, 29.4%) were used. Fundal height measures (estimated by midwives by hand measurements, typically without a measuring tape) upon presentation for delivery were compared with ultrasound expected delivery dates and adjusted to control for fundal height underestimation of gestation length by an average of two weeks. If the gestational age based on LMP exceeded 330 days [48], fundal height was used to estimate GA. Meteorological predictor selection was based on the literature and variation in weather in the study area. We selected the number of days of precipitation (as per Grace et al. (2015)) above 5mm during the exposure window (described below) as our meteorological exposure for the effects of precipitation in our models. Given the limited range of temperature fluctuations in the region, the mean of daily temperature (°C) during each exposure period was selected to examine the effects of temperature on the outcome variables. Birth season was included in all models as a fixed effect to control for any unaccounted effects of seasonality [49]. Meteorological data for rainfall and temperature were extracted and matched to each birth for multiple exposure periods reflecting the entire pregnancy period and each trimester. Daily rainfall data were estimated based on satellite observations using the Rainfall Estimator, version 2.0 (RFE2) algorithm [50, 51]. In this algorithm, rainfall amounts estimated from geostationary satellite infrared images at high spatial and temporal resolutions are calibrated against ground rain gauge data as well as satellite microwave measurements. These data were then interpolated to the Buhoma region to form a daily rainfall time series for the study period. We processed daily temperature data from the ERA-Interim (ERAi) reanalysis dataset of European Center of Medium-range Weather Forecasts [52]. This dataset is generated by assimilating meteorological measurements from various observational sources into a global numerical model and then forecast at high temporal resolution. We interpolated gridded 3-hourly ERAi temperature data at 0.75 degree resolution to Buhoma site. The daily mean/maximum/minimum temperatures were then analyzed from the 3-hourly temperature time series at the site. We validated our extracted daily rainfall and temperature data against local meteorological measurements (approximately 12 months) for a weather station in Buhoma Town. Both rainfall and temperature variables were highly correlated with the station measurements at higher than a 99% confidence level. We controlled for variables known to influence birth weight (Table 2). We included infant sex as a control variable as male infants are typically larger than female infants [53]. To estimate effects of meteorological exposures on foetal growth in utero, we controlled for gestational age. Though foetal growth is approximated as gestational age-specific birth weight, we included a categorized gestational age as a covariate to reduce measurement errors due to heterogeneous measures of gestational age in the study sample. Based on available data and known determinants of birth weight [54], we included maternal age, ethnicity, parity, and maternal marital status as control variables. HIV status was also included as a control as infants born to HIV-positive mothers are more likely to be classified as low birth weight (LBW) [55]. We considered delivery type as a possible control variable; while delivery type would not be a determinant of LBW, it can be an indicator of complications, and was thus included in models. There were three classifications for delivery type: spontaneous vaginal birth, assisted vaginal birth (i.e. via vacuum extraction or with forceps), or Caesarean section. Whether or not a woman had an ultrasound scan during her pregnancy was included as a proxy for access and quality of antenatal care, as only women who attended antenatal care underwent ultrasounds and ultrasounds were generally only available in the better-equipped hospital-based clinics (as opposed to satellite clinics in more remote areas). Women who attended antenatal care at a facility with ultrasound services are meant to undergo two scans at different points in their pregnancy, but often undergo only one. The timing of the first scan is dependent upon the date of the first antenatal care visit, which occurs in the second trimester for the majority of women attending antenatal care at Bwindi Community Hospital [56]. The ultrasound costs 1000 Ugandan shillings, a cost that the BCH sonographer expressed may be prohibitive to some mothers with limited financial means [57]. Women experiencing complications or who are primigravida may undergo ultrasound scans sooner and/or more often than women whose pregnancies are progressing normally [56]. We included ethnicity to compare Indigenous vs. non-Indigenous birth outcomes in the population, which in this context serves as a partial proxy for socio-economic status (see Section 2.1). We hypothesized that ethnicity would act not only as a control variable, but also an effect modifier for meteorological effects on birth weight. This hypothesis is consistent with climate change adaptation literature, which theorizes that climatic and/or meteorological effects will manifest through, and be modified by, existing gradients in health [2, 4, 6, 7]. Our ethnicity variable had low variation; the small subset of Indigenous Batwa limited the statistical power of analyses using data from this subset. Our analytical sample included live singleton births with information on birth weight, gestational age, and all known covariates (n = 3197). Linear regression was used to estimate mean differences in birth weight associated with meteorological exposures. We constructed a ‘base’ model that includes infant sex and gestational age to examine sex- and gestational age-specific birth weight differences by exposure and potential confounders. Gestational age was included as a categorical control variable due to the potential relatively high measurement errors inherent to LMP- and fundal height-based measurements and a non-linear relation to birth weight suggested by the lowess-smoothed scatterplot. We then built models that included all control variables and meteorological exposures, constructing four models: a model with meteorological exposures and all control variables for each of the three trimesters, and a fourth model with meteorological exposures and all control variables for the entire pregnancy period. We stratified the sample by ethnicity and tested the same models in each group to investigate ethnicity as an effect modifier. In a separate analysis, we stratified the sample by season of birth and tested the models for each exposure period. We conducted multiple sensitivity analyses to examine robustness of associations: (1) re-analyzing the sample after excluding preterm births (<37 weeks), (2) re-analyzing the sample after excluding all cases for which GA was not determined by ultrasound, (3) examining associations using different thresholds for temperature (e.g., maximum and minimum temperature exposures) and precipitation (e.g., exposure to number of days with rainfall over 1mm, or number of days with rainfall over 10mm), (4) testing for interaction effects between meteorological exposures and infant sex, and (5) testing models with month of conception instead of season of birth to evaluate any non-random seasonal fertility selection. Further sensitivity analyses evaluated LBW as an outcome using logistic regression. This analysis allowed us to examine the lower 5–10% of birth weights separately, since as Grace, Davenport [17] state, this lowest distribution of birth weights is not accurately captured as a continuous dependent variable. To assess model fit, we evaluated collinearity by examining the Pearson correlation coefficient matrix for all predictor variables and control variables. We conducted post-estimation tests to assess heteroscedasticity (Breysch-Pagan test) and the normality of the residuals (Q-Q plots and Shapiro-Wilk tests) in our models. We checked that the outcome variable was normally distributed (in both the overall samples and those stratified by ethnicity). We also evaluated the linearity of the meteorological exposure variables and other control variables using scatterplots, as well as lowess smoothing to visually assess trends in these plots. Data were analysed using Stata v.13 (StataCorp, USA). This study was approved by the McGill University and the University of Guelph Research Ethics Boards (REB File #461–0414). Additionally, the research team has a Memorandum of Understanding with, and received approval for this study from, Bwindi Community Hospital (BCH). The study design conforms to the Canadian Tri-council Policies and follows the requirements of the Ethical Conduct of Research Involving Human Subjects; it is also in compliance with Ugandan laws and regulations for foreign researchers. As per McGill, Tri-Council, and BCH ethical research policies, informed consent by individual patients specific to this study was not required. This study included retrospective analysis of de-identified hospital records, and consent to use the data for hospital and hospital-approved research analysis is provided by the patient at the time of admission to the hospital or maternity ward.
N/A
