Environmental heat stress on maternal physiology and fetal blood flow in pregnant subsistence farmers in The Gambia, west Africa: an observational cohort study

Background: Anthropogenic climate change has caused extreme temperatures worldwide, with data showing that sub-Saharan Africa is especially vulnerable to these changes. In sub-Saharan Africa, women comprise 50% of the agricultural workforce, often working throughout pregnancy despite heat exposure increasing the risk of adverse birth outcomes. In this study, we aimed to improve understanding of the pathophysiological mechanisms responsible for the adverse health outcomes resulting from environmental heat stress in pregnant subsistence farmers. We also aimed to provide data to establish whether environmental heat stress also has physiological effects on the fetus. Methods: We conducted an observational cohort study in West Kiang, The Gambia, at the field station for the Medical Research Council Unit The Gambia at London School of Hygiene & Tropical Medicine (named the MRC Keneba field station). Pregnant women who were aged 16 years or older and who were at <36 weeks’ gestation of any gravida or parity were invited to participate in the study. Participants were eligible if they were involved in agricultural or related manual daily tasks of living. Participants were ineligible if they refused to provide consent, had multiple pregnancies (eg, if they had twins), were acutely unwell, or were diagnosed with pre-eclampsia or eclampsia. Heat stress was measured by wet bulb globe temperature (WBGT) and by using the universal thermal climate index (UTCI), and maternal heat strain was directly measured by modified physiological strain index calculated from heart rate and skin temperature. Outcome measures of fetal heart rate (FHR) and fetal strain (defined as a FHR >160 beats per min [bpm] or <115 bpm, or increase in umbilical artery resistance index) were measured at rest and during the working period. Multivariable repeated measure models (linear regression for FHR, and logistic regression for fetal strain) were used to evaluate the association of heat stress and heat strain with acute fetal strain. Findings: Between Aug 26, 2019, and March 27, 2020, 92 eligible participants were recruited to the study. Extreme heat exposure was frequent, with average exposures of WBGT of 27·2°C (SD 3·6°C) and UTCI equivalent temperature of 34·0°C (SD 3·7°C). The total effect of UTCI on fetal strain resulted in an odds ratio (OR) of 1·17 (95% CI 1·09–1·29; p<0·0001), with an adjusted direct effect of OR of 1·12 (1·03–1·21; p=0·010) with each 1°C increase in UTCI. The adjusted OR of maternal heat strain on fetal strain was 1·20 (1·01–1·43; p=0·038), using the UTCI model, with each unit increase. Interpretation: Data from our study show that decreasing maternal exposure to heat stress and heat strain is likely to reduce fetal strain, with the potential to reduce adverse birth outcomes. Further work that explores the association between heat stress and pregnancy outcomes in a variety of settings and populations is urgently needed to develop effective interventions. Funding: The Wellcome Trust. This observational cohort study was set in West Kiang, a rural area that is approximately 150 km from the capital of The Gambia, where the field station for the Medical Research Council Unit The Gambia at London School of Hygiene & Tropical Medicine is located (named the MRC Keneba field station). West Kiang is populated by 34 villages, with populations ranging from 200 to 2000 people. Antenatal care is provided free of charge at the MRC Keneba field station, where women attend antenatal appointments every month, are routinely supplemented with iron and folic acid, and around 30% of whom attend for delivery.17 At the time of the study, most villages did not have access to gridded electricity or piped running water. Pregnant women who were aged 16 years or older and who were at fewer than 36 weeks’ gestation of any gravida or parity were invited to participate in the study. Participants were eligible if they were involved in agricultural or related manual daily tasks of living (including water collection, gardening, rice planting, harvesting, and washing clothes). Participants were ineligible if they refused to provide consent, had multiple pregnancies (eg, if they were pregnant with twins), were acutely unwell, or were diagnosed with pre-eclampsia or eclampsia (appendix p 3). All participants gave written informed consent to participate in the study. The study was approved by the Gambian Government and the Medical Research Council Unit The Gambia joint ethics committee and the London School of Hygiene & Tropical Medicine ethics advisory board in accordance with the Declaration of Helsinki (2013). The published study protocol gives full details of the methods.18 Figure 1 provides a summary of the physiological pathways investigated and relevant measurements taken. Maternal physiological responses to thermal heat stress Measurements taken at each field visit are shown in grey. PSI=physiological strain index. RI=resistance index of umbilical artery. Baseline data for previous medical conditions, obstetric history, and current pregnancy were collected for this study. Gestational age was ascertained through last menstrual period (if known) or biparietal diameter, which was measured by ultrasound scan. An ultrasound scan at each visit was conducted to measure fetal growth and to identify anomalies. The study team were scheduled to attend field visits with participants every 2 months until the time of delivery. During field visits, participants were encouraged to carry out their usual daily tasks. Hourly air temperature, relative humidity, black globe temperature, wet bulb globe temperature (WBGT), and wet bulb and dew point data were measured throughout the participants’ outdoor working shift using the HT200: Heat Stress WBGT Meter (Extech, Nashua, NH, USA). Wind speed was measured using a portable anemometer (Extech AN100 thermo-anemometer, Nashua, NH, USA). From these measurements, the universal thermal climate index (UTCI) was calculated using the alfrisci/rBiometeo package.19 Both average conditions and peak conditions were calculated for WBGT, UTCI, air temperature, and relative humidity for each participant during each visit. WBGT is widely used in both occupational health regulations and in research. Our decision to measure WBGT was to aid the comparability and potential generalisability of our findings. WBGT is a continuous measure but has been divided into five categories, from no risk (0) of heat illness in occupational settings to extreme risk (5) of heat illness in occupational settings. The UTCI, a more recent heat stress index, is based on a physiological model and is potentially better than WBGT at expressing biothermal conditions. The UTCI is divided into ten categories that range from extreme cold stress to extreme heat stress.20 To the best of our knowledge, there are no modifications that have been specifically developed for pregnancy in any of the heat indexes and therefore both were evaluated. Participants were fitted with an Equivital LifeMonitor (Cambridge, UK), which is a wearable device with inbuilt fabric sensors. This device continuously recorded maternal heart rate, skin temperature, and triaxis accelerometer data. Duplicate tympanic temperature measurements from both membranes were taken using a Braun ThermoScan 7 tympanic thermometer (Braun, Kronberg, Germany) and the highest measurement was recorded at each timepoint.21, 22 The physiological strain index (PSI) has been widely accepted as an accurate measurement of heat strain.23 This model has been validated for men and women, but there are no data for its use during pregnancy. PSI requires continuous core temperature measurements, but these were unavailable owing to an absence of data for the safety of telemetry pills in pregnancy and therefore we used continuous chest skin temperature measurements to give a modified PSI (PSIMOD), using the following equation:23, 24 wherein skin temperature at time t (TSKT), tympanic temperature at baseline (TTYMP0), heart rate at time t (HRt), and heart rate at rest (HR0) are represented. Use of triaxial accelerometer data in pregnancy to estimate energy expenditure is problematic.25 We increased the robustness of our estimates by calculating the physical activity energy expenditure (PAEE) with two methods. First, PAEE was calculated from the LifeMonitor data using the following validated equation (method A), with the standard 6 min walk test used to calibrate the device:26 wherein heart rate above sleeping or resting heart rate (HRaS), average heart rate above resting heart rate at 3 min into the calibration test (HRaS3min), root mean square of successive difference of inter-beat intervals during calibration test (RMSSD), and sleeping or resting heart rate (SHR) are represented. Second, energy expenditure was determined on the basis of direct activity observations of each participant matched to historic indirect calorimetry data for energy expenditure by task in pregnant women in West Kiang, determined by Lawrence and colleagues (method B),27 which was given in metabolic equivalents (METs). Where estimates from the LifeMonitor were outside the expected range (2–15 METs), historical data were used. Measures of hydration, body composition, haematology, and biochemistry were measured at the beginning and end of each field visit. The TANITA BC-418MA segmental body composition analyser uses bioimpedance to measure free fat mass, fat percentage, total body water, and basal metabolic rate. Maternal symptoms of heat illness were collected at the midpoint and endpoint of field visits. According to Centers for Disease Control and Prevention guidance and other published literature,11, 28 self-reported symptoms included nausea, vomiting, headache, dizziness, weakness, muscle ache, fatigue, and dry mouth. Ultrasound for biometry, fetal heart rate (FHR), placental position, and assessment of liquor volume was conducted before every field visit by a trained midwife. Umbilical artery doppler is an established non-invasive tool for measuring placental blood flow and gives information on the maternal–fetal circulation.29 The UmbiFlow device is a portable, low-cost, continuous wave doppler that was developed in South Africa and is validated for use in measuring umbilical artery blood flow. In eligible participants (those at ≥28 weeks’ gestation), a baseline recording of the resistance index (defined as [systolic velocity minus diastolic velocity] divided by systolic velocity) was recorded at this time. FHR and resistance index measures that were taken at this time were taken as the baseline measurements. During field visits, FHR and (if eligible) the umbilical artery resistance index were recorded at the midpoint and the endpoint of the working shift. The umbilical artery resistance index was recorded twice at each timepoint and assessed for quality. An outcome of fetal strain was defined as a FHR greater than or equal to 160 beats per min (bpm), less than or equal to 115 bpm for 5 min, or an increase in umbilical artery risk category from baseline, based on the definition of fetal distress and in keeping with clinical practice and current evidence.30 These measurements were used as surrogate markers for impact on maternal–fetal circulation. Data for gestational age, birthweight (within 72 h of delivery), infant sex, mode of delivery, place, level of training of the attendant at delivery, and infant or maternal mortality were collected. The primary outcome of the study was presence of fetal strain, which was assessed in all participants at each field visit. The theoretical framework of this work has been described previously.18 As fetal heat strain has never been defined previously and would be extremely challenging to measure, especially in the field, we use surrogate measures (fetal heart rate and umbilical artery resistance index) to determine the effect of heat stress on blood flow to the fetus, defined below as fetal strain. This study was not powered to assess the effect of heat strain on birth outcomes. The initial sample size estimation in our study protocol was based on 35% of workers being exposed to heat stress (WBGT >24·8°C).11, 18 An updated sample size calculation that included an increased exposure level and that assumed an unexposed incidence risk of fetal strain to be 5% estimated that a sample size of 74 would be needed to detect an exposure incidence risk of 30%, with an α of 0·05 and a power of 80%. All analyses were conducted in R, version 4.1.0. Normally distributed continuous variables are presented as a mean with SD, non-parametric data as median and IQR, and categorical variables as counts. Heat stress exposure variables were analysed as continuous data. Outcome measures of fetal strain were analysed by both FHR as a continuous variable and fetal strain as a binary variable as defined in the Procedures and outcomes section. Initial data exploration assessed changes in mean temperature and heart rate from baseline to working state using Wilcoxon signed-rank tests. The correlation between multiple similar variables (eg, WBGT, UTCI, and air temperature) were evaluated using Pearson’s correlation. Univariable analysis of maternal heat strain (PSIMOD), fetal strain by FHR (continuous variable and linear regression), and fetal strain (binary variable and logistic regression) were conducted to explore risk factors. Final datasets in multivariable analyses were complete. The association between heat stress and maternal heat strain (by PSIMOD) was explored using linear and non-linear models.31 Non-linear models with natural and logarithmic splines with knots placed at the 50th and 90th centiles were evaluated, in keeping with other studies on the association between temperature and health outcomes. The linear model had the lowest Akaike information criterion and was used in the subsequent repeated measures multivariable models. To explore the effect of heat stress and maternal heat strain on the fetus, we used two multivariable repeated measures models with FHR (model A, linear regression) and fetal strain (model B, logistic regression) as outcomes. All variables were decided a priori on the basis of biological plausibility and directed acyclic graphs (appendix pp 13–14). Model 1 shows the total effect of heat stress on fetal strain or FHR: wherein fetal strain or fetal heart rate for individual i at time j (Y) and heat stress exposure for individual i at time j are represented. Model 1 gives the estimate of effect (model A) and the odds ratio (OR; model B) for each 1°C increase in heat stress on fetal strain to give the total effect of heat stress. Model 2 gives the direct effect of heat stress on fetal strain while controlling for maternal heat strain: wherein FHR or fetal strain for individual i at time j (Y), UTCI or WBGT for individual i at time j (heat stress), PSIMOD of individual i at time j (heat strain), estimation of cardiovascular fitness determined by distance travelled in standardised 6 min walk test for individual i at time j (fitness), and measurement of body fat (as BMI is not useful in pregnancy) for individual i at time j (% fat mass) are represented. Model 2 gives the estimate of effect (model A) and the OR (model B) for each 1°C increase in heat stress on fetal strain, controlling for maternal heat strain. This model estimates the effects of heat stress on fetal strain due to other mechanisms outside of maternal heat strain. Model 3 gives the direct effect of maternal heat strain on fetal strain while controlling for heat stress, cardiac fitness, percentage of fat mass, and gestational age: wherein FHR or fetal strain for individual i at time j (Y), UTCI or WBGT for individual i at time j (heat stress), PSIMOD of individual i at time j (heat strain), estimation of cardiovascular fitness determined by distance travelled in standardised 6 min walk test for individual i at time j (fitness), and measurement of body fat (as BMI is not useful in pregnancy) for individual i at time j (% fat mass) are represented. Full details of all models and analysis code can be found in the appendix (pp 19–22). The final models were assessed for violation of model assumptions by assessing the linearity of residuals, homoscedasticity by Levene’s Test, and the normal distribution of residuals. The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.