Critical Windows of Maternal Exposure to Biothermal Stress and Birth Weight for Gestational Age in Western Australia

Background: There is limited and inconsistent evidence on the risk of ambient temperature on small for gestational age (SGA) and there are no known related studies for large for gestational age (LGA). In addition, previous studies used temperature rather than a biothermal metric. Objectives: Our aim was to examine the associations and critical susceptible windows of maternal exposure to a biothermal metric [Universal Thermal Climate Index (UTCI)] and the hazards of SGA and LGA. Methods: We linked 385,337 singleton term births between 1 January 2000 and 31 December 2015 in Western Australia to daily spatiotemporal UTCI. Distributed lag nonlinear models with Cox regression and multiple models were used to investigate maternal exposure to UTCI from 12 weeks preconception to birth and the adjusted hazard ratios (HRs) of SGA and LGA. Results: Relative to the median exposure, weekly and monthly specific exposures showed potential critical windows of susceptibility for SGA and LGA at extreme exposures, especially during late gestational periods. Monthly exposure showed strong positive associations from the 6th to the 10th gestational months with the highest hazard of 13% for SGA (HR=1.13; 95% CI: 1.10, 1.14) and 7% for LGA (HR=1.07; 95% CI: 1.03, 1.11) at the 10th month for the 1st UTCI centile. Entire pregnancy exposures showed the strongest hazards of 11% for SGA (HR=1.11; 95% CI: 1.04, 1.18) and 3% for LGA (HR=1.03; 95% CI: 0.95, 1.11) at the 99th UTCI centile. By trimesters, the highest hazards were found during the second and first trimesters for SGA and LGA, respectively, at the 99th UTCI centile. Based on estimated interaction effects, male births, mothers who were non-Caucasian, smokers, ≥35 years of age, and rural residents were most vulnerable. Conclusions: Both weekly and monthly specific extreme biothermal stress exposures showed potential critical susceptible windows of SGA and LGA during late gestational periods with disproportionate sociodemographic vulnerabilities. https://doi.org/10.1289/EHP12660


Table of Contents
Table S1.The exposure-response association between maternal weekly-specific UTCI exposures from twelve weeks before conception through to gestational weeks at birth and the hazard ratios HR (95% CI) of SGA at various percentiles of the exposure in Western Australia, 2000Australia, -2015. .Table S2.The exposure-response association between maternal weekly-specific UTCI exposures from twelve weeks before conception through to gestational weeks at birth and the hazard ratios HR (95% CI) of LGA at various percentiles of the exposure in Western Australia, 2000Australia, -2015. .Table S3.The exposure-response association between maternal monthly-specific UTCI exposures from three months before conception through to gestational weeks at birth and the hazard ratios HR (95% CI) of SGA at various percentiles of the exposure in Western Australia, 2000Australia, -2015. .Table S4.The exposure-response association between maternal monthly-specific UTCI exposures from three months before conception through to gestational weeks at birth and the hazard ratios HR (95% CI) of LGA at various percentiles of the exposure in Western Australia, 2000-2015.
Table S5.The exposure-response association between maternal cumulative UTCI exposures over twelve weeks preconception through to pregnancy and trimester-specific periods with reference to median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA at various percentiles of the exposure in Western Australia, 2000Australia, -2015. .Table S6.The exposure-response association between maternal cumulative UTCI exposures over twelve weeks preconception through to pregnancy by subgroups with reference to the median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA at various percentiles of the exposure in Western Australia, 2000-2015.

Figure S2.
The exposure-response association between maternal cumulative UTCI exposures over twelve weeks preconception, preconception through to birth, and entire pregnancy with reference to median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA.Solid blue lines represent point estimates, and the whiskers represent 95% confidence intervals.Numeric data can be found in Table 3 and Table S5.DLNM Cox proportional hazard models were adjusted for infant sex, maternal age, race or ethnicity, marital status, smoking status, parity, remoteness, socioeconomic status, and year and month of conception.

Figure S3
. The exposure-response association between maternal trimester-average cumulative UTCI exposures with reference to median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA. Solid blue lines represent point estimates, and the whiskers represent 95% confidence intervals.DLNM Cox proportional hazard models were adjusted for infant sex, maternal age, race or ethnicity, marital status, smoking status, parity, remoteness, socioeconomic status, and year and month of conception.Numeric data can be found in Table 3 and Table S5.

Figure S4.
The exposure-response association between maternal weekly-specific average UTCI exposures over twelve weeks preconception through to gestational weeks at birth and the hazard ratios HR (95% CI) of SGA and LGA at various thresholds of exposure.Mean (14.5 ˚C) instead of the median (14.2 ˚C) UTCI was used as a reference.Numeric data can be found in Excel Tables S1 and S2.

Figure S5.
The exposure-response association between maternal weekly-specific average UTCI exposures over twelve weeks preconception through to gestational weeks at birth with reference to median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA at various thresholds of exposure.Cross-basis was constructed by increasing the degrees of freedom by one (that is 7 and 4 for exposure and exposure period, respectively).Numeric data can be found in Excel Tables S3  and S4.

Figure S6.
The exposure-response association between maternal weekly-specific average UTCI exposures over twelve weeks preconception through to gestational weeks at birth with reference to median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA at various thresholds of exposure.Maternal age was adjusted as categorical instead of a natural spline of the continuous variable.Numeric data can be found in Excel Tables S5 and S6.

Figure S7.
The exposure-response association between maternal weekly-specific average UTCI exposures over twelve weeks preconception through to gestational weeks at birth with reference to median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA at various thresholds of exposure.Season of conception was adjusted as categorical (summer, spring, winter, and autumn) variable instead of calendar month index.Numeric data can be found in Excel Tables S7 and S8.

Figure S8.
The exposure-response association between maternal weekly-specific average UTCI exposures over twelve weeks preconception through to gestational weeks at birth with reference to median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA at various thresholds of exposure.All eligible births with 22-42 gestational weeks were analysed instead of only term births.Numeric data can be found in Excel Tables S9 and S10.

Figure S9.
The exposure-response association between maternal weekly-specific average UTCI exposures from conception through to gestational weeks at birth with reference to median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA at various thresholds of exposure.Included only weekly exposures from conception to birth (that is, preconception exposures were excluded).Numeric data can be found in Excel Tables S11 and S12.

Figure S10.
The exposure-response association between maternal trimester-average cumulative UTCI exposures with reference to median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA.Three separate models were conducted for each trimester-average exposure instead of concurrent analysis in a single model.DLNM Cox proportional hazard models were adjusted for infant sex, maternal age, race or ethnicity, marital status, smoking status, parity, remoteness, socioeconomic status, and year and month of conception.Numeric data can be found in Excel Table S13.

Figure S11.
The exposure-response association between maternal weekly-specific average UTCI exposures over twelve weeks preconception through to gestational weeks at birth with reference to median 14.2 ˚C and the odds ratios OR (95% CI) of SGA and LGA at various thresholds of exposure.DLNM logistic regression instead of DLNM Cox proportional hazard regression was used.Numeric data can be found in Excel Tables S14 and S15 in the excel spreadsheet.1.00 Note: Model was adjusted for infant sex, maternal age, race or ethnicity, marital status, parity, maternal smoking, remoteness, areal level socioeconomic status, and year and month of conception.UTCI, Universal Thermal Climate Index in degree Celsius; HR, hazard ratio; LCI and UCI, 95% lower and upper confidence intervals; LGA, large for gestational age; P1-P99, first to 99 th centiles of exposure.

Appendix 1. R syntax for DLNMs Cox regression. Additional File-Excel Document
Table S3.The exposure-response association between maternal monthly-specific UTCI exposures from three months before conception through to gestational weeks at birth and the hazard ratios HR (95% CI) of SGA at various percentiles of the exposure in Western Australia, 2000 1.03 (0.97, 1.10) 1.01 (0.94, 1.09) Note: DLNM Cox proportional hazard models were adjusted for infant sex, maternal age, race or ethnicity, marital status, parity, maternal smoking, remoteness, areal level socioeconomic status, year, and month of conception.P1-P99, first to 99 th centile of UTCI, Universal Thermal Climate Index in degree Celsius; HR, hazard ratio; CI, confidence interval; SGA, small for gestational age; LGA, large for gestational age; DLNM, Distributed lag nonlinear model.Table S6.The exposure-response association between maternal cumulative UTCI exposures over twelve weeks preconception through to pregnancy by subgroups with reference to the median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA at various percentiles of the exposure in Western Australia, 2000Australia, -2015 Excluded births with incorrect SA1 to assign exposure (n=5) Figure S2.The exposure-response association between maternal cumulative UTCI exposures over twelve weeks preconception, preconception through to birth, and entire pregnancy with reference to median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA.Solid blue lines represent point estimates, and the whiskers represent 95% confidence intervals.Numeric data can be found in Table 3 and Table S5.DLNM Cox proportional hazard models were adjusted for infant sex, maternal age, race or ethnicity, marital status, smoking status, parity, remoteness, socioeconomic status, and year and month of conception.Note: HR, hazard ratio; CI, confidence interval; SGA, small for gestational age; LGA, large for gestational age; DLNM, Distributed lag non-linear model.

SGA
LGA Figure S3.The exposure-response association between maternal trimester-average cumulative UTCI exposures with reference to median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA.Solid blue lines represent point estimates, and the whiskers represent 95% confidence intervals.DLNM Cox proportional hazard models were adjusted for infant sex, maternal age, race or ethnicity, marital status, smoking status, parity, remoteness, socioeconomic status, and year and month of conception.Numeric data can be found in Table 3 and Table S5.Note: HR, hazard ratio; CI, confidence interval; SGA, small for gestational age; LGA, large for gestational age; DLNM, Distributed lag non-linear model.

SGA
LGA Figure S4.The exposure-response association between maternal weekly-specific average UTCI exposures over twelve weeks preconception through to gestational weeks at birth and the hazard ratios HR (95% CI) of SGA and LGA at various thresholds of exposure.Mean (14.5 ˚C) instead of the median (14.2 ˚C) UTCI was used as a reference.Numeric data can be found in Excel Tables S1 and S2.Note: DLNM Cox proportional hazard models were adjusted for infant sex, maternal age, race or ethnicity, marital status, parity, maternal smoking, remoteness, areal level socioeconomic status, year, and month of conception.UTCI, Universal Thermal Climate Index in degree Celsius; HR, hazard ratio; CI, confidence interval; SGA, small for gestational age; LGA, large for gestational age; DLNM, Distributed lag non-linear model.
SGA LGA Figure S5.The exposure-response association between maternal weekly-specific average UTCI exposures over twelve weeks preconception through to gestational weeks at birth with reference to median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA at various thresholds of exposure.Cross-basis was constructed by increasing the degrees of freedom by one (that is 7 and 4 for exposure and exposure period, respectively).Numeric data can be found in Excel Tables S3 and S4.Note: DLNM Cox proportional hazard models were adjusted for infant sex, maternal age, race or ethnicity, marital status, parity, maternal smoking, remoteness, areal level socioeconomic status, year, and month of conception.UTCI, Universal Thermal Climate Index in degree Celsius; HR, hazard ratio; CI, confidence interval; SGA, small for gestational age; LGA, large for gestational age; DLNM, Distributed lag non-linear model.

SGA
LGA Figure S6.The exposure-response association between maternal weekly-specific average UTCI exposures over twelve weeks preconception through to gestational weeks at birth with reference to median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA at various thresholds of exposure.Maternal age was adjusted as categorical instead of a natural spline of the continuous variable.Numeric data can be found in Excel Tables S5 and S6.Note: DLNM Cox proportional hazard models were adjusted for infant sex, maternal age, race or ethnicity, marital status, parity, maternal smoking, remoteness, areal level socioeconomic status, year, and month of conception.
UTCI, Universal Thermal Climate Index in degree Celsius; HR, hazard ratio; CI, confidence interval; SGA, small for gestational age; LGA, large for gestational age; DLNM, Distributed lag non-linear model.
Figure S7.The exposure-response association between maternal weekly-specific average UTCI exposures over twelve weeks preconception through to gestational weeks at birth with reference to median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA at various thresholds of exposure.Season of conception was adjusted as categorical (summer, spring, winter, and autumn) variable instead of calendar month index.Numeric data can be found in Excel Tables S7 and S8.Note: DLNM Cox proportional hazard models were adjusted for infant sex, maternal age, race or ethnicity, marital status, parity, maternal smoking, remoteness, areal level socioeconomic status, year, and season of conception.UTCI, Universal Thermal Climate Index in degree Celsius; HR, hazard ratio; CI, confidence interval; SGA, small for gestational age; LGA, large for gestational age; DLNM; Distributed lag non-linear model.

SGA
LGA Figure S8.The exposure-response association between maternal weekly-specific average UTCI exposures over twelve weeks preconception through to gestational weeks at birth with reference to median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA at various thresholds of exposure.All eligible births with 22-42 gestational weeks were analysed instead of only term births.Numeric data can be found in Excel Tables S9 and S10.Note: DLNM Cox proportional hazard models were adjusted for infant sex, maternal age, race or ethnicity, marital status, parity, maternal smoking, remoteness, areal level socioeconomic status, and year and month of conception.UTCI, Universal Thermal Climate Index in degree Celsius; HR, hazard ratio; CI, confidence interval; SGA, small for gestational age; LGA, large for gestational age; DLNM; Distributed lag non-linear model.

SGA
LGA SGA LGA Figure S9.The exposure-response association between maternal weekly-specific average UTCI exposures from conception through to gestational weeks at birth with reference to median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA at various thresholds of exposure.Included only weekly exposures from conception to birth (that is, preconception exposures were excluded).Numeric data can be found in Excel Tables S11 and S12.Note: DLNM Cox proportional hazard models were adjusted for infant sex, maternal age, race or ethnicity, marital status, parity, maternal smoking, remoteness, areal level socioeconomic status, and year and month of conception.UTCI, Universal Thermal Climate Index in degree Celsius; HR, hazard ratio; CI, confidence interval; SGA, small for gestational age; LGA, large for gestational age; DLNM, Distributed lag non-linear model.

SGA
LGA Figure S10.The exposure-response association between maternal trimester-average cumulative UTCI exposures with reference to median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA.Three separate models were conducted for each trimester-average exposure instead of concurrent analysis in a single model.DLNM Cox proportional hazard models were adjusted for infant sex, maternal age, race or ethnicity, marital status, smoking status, parity, remoteness, socioeconomic status, and year and month of conception.Numeric data can be found in Excel Table S13.Note: HR, hazard ratio; CI, confidence interval; SGA, small for gestational age; LGA, large for gestational age; DLNM, Distributed lag non-linear model.

SGA
LGA Figure S11.The exposure-response association between maternal weekly-specific average UTCI exposures over twelve weeks preconception through to gestational weeks at birth with reference to median 14.2 ˚C and the odds ratios OR (95% CI) of SGA and LGA at various thresholds of exposure.DLNM logistic regression instead of DLNM Cox proportional hazard regression was used.Numeric data can be found in Excel Tables S14 and S15 in the excel spreadsheet.Note: Models were adjusted for infant sex, maternal age, race or ethnicity, marital status, parity, maternal smoking, remoteness, areal level socioeconomic status, year, and month of conception.UTCI, Universal Thermal Climate Index in degree Celsius; OR, odd ratio; CI, confidence interval; SGA, small for gestational age; LGA, large for gestational age; DLNM, Distributed lag non-linear model.

Figure S1 .
Figure S1.Flow chart for selecting the eligible births included in this study, Western Australia, 2000-2015.

Table S1 .
The exposure-response association between maternal weekly-specific UTCI exposures from twelve weeks before conception through to gestational weeks at birth and the hazard ratios HR (95% CI) of SGA at various percentiles of the exposure in Western Australia, 2000-2015.Model was adjusted for infant sex, maternal age, race or ethnicity, marital status, parity, maternal smoking, remoteness, areal level socioeconomic status, and year and month of conception.UTCI, Universal Thermal Climate Index in degree Celsius; HR, hazard ratio; LCI and UCI, 95% lower and upper confidence intervals; SGA, small for gestational age; P1-P99, first to 99 th centiles of exposure.

Table S2 .
The exposure-response association between maternal weekly-specific UTCI exposures from twelve weeks before conception through to gestational weeks at birth and the hazard ratios HR (95% CI) of LGA at various percentiles of the exposure in WesternAustralia, 2000Australia,  -2015.   .

Table S4 .
Model was adjusted for infant sex, maternal age, race or ethnicity, marital status, parity, maternal smoking, remoteness, areal level socioeconomic status, and year and month of conception.UTCI, Universal Thermal Climate Index in degree Celsius; HR, hazard ratio; LCI and UCI, 95% lower and upper confidence intervals; SGA, small for gestational age; P1-P99, first to 99 th centiles of exposure.The exposure-response association between maternal monthly-specific UTCI exposures from three months before conception through to gestational weeks at birth and the hazard ratios HR (95% CI) of LGA at various percentiles of the exposure in Western Australia, 2000-2015 Model was adjusted for infant sex, maternal age, race or ethnicity, marital status, parity, maternal smoking, remoteness, areal level socioeconomic status, and year and month of conception.UTCI, Universal Thermal Climate Index in degree Celsius; HR, hazard ratio; LCI and UCI, 95% lower and upper confidence intervals; LGA, large for gestational age; P1-P99, first to 99 th centiles of exposure.

Table S5 .
The exposure-response association between maternal cumulative UTCI exposures over twelve weeks preconception through to pregnancy and trimester-specific periods with reference to median 14.2 ˚C and the hazard ratios HR (95% CI) of SGA and LGA at various percentiles of the exposure in Western Australia, 2000- Socioeconomic status; UTCI, Universal Thermal Climate Index in degree Celsius; HR, hazard ratio; LCI and UCI, 95% lower and upper confidence intervals; LGA, large for gestational age. Figure S1.Flow chart for selecting the eligible births included in this study, Western Australia, 2000-2015.Note: SA1, statistical area level 1.