Comparison of the relation between smoking during pregnancy and offspring height or body size at the age of ten years in the three ethnic populations
Difference in used variables among distinct populations
It was seen that maternal smoking during pregnancy was much more frequent in the European population (30.9% for height and 30.2% for body size) than in the South Asian (2.4% for height and 2.2% for body size) and African (3.9% for height and 3.8% for body size) populations (Fig. 2); whereas the breastfeeding rate was considerably high across the three populations, especially in the South Asian and African populations (> 90%). The differences in income, gender, height, and body size were also pronounced among these populations; for instance, more low-income participants (nearly 50%) were observed in the African population compared to the other two populations. In addition, the proportion of being shorter was slightly greater in the European population (27.5%), while the rate of being fatter was relatively higher in the African population (27.4%).
Confounding Structure For Height
We here explored the relation between potential covariates and height, and observed that higher income was related to decreased rate of smoking during pregnancy in both the European and South Asian populations, but was associated with increased rate of smoking during pregnancy in the African population (Table 2); however, such association was only significant in the European population (P = 1.24×10− 9). As could be anticipated, higher birthweight was relevant to lower risk of being shorter in all ethnic populations (P < 0.05). Breastfeeding would likely reduce the risk of being shorter, but such reduction was only pronounced in the European population (P = 2.75×10− 13). Interestingly, income showed inconsistent associations with the risk of being shorter in diverse populations although all the relations were non-significant; briefly, a decreased risk of being shorter was seen in the European (P = 0.443) and African (P = 0.943) populations, but an increased risk of being shorter was observed in the South Asian population (P = 0.696). In addition, boys had lower risk of being shorter compared to girls in the three ethnic populations at the age of ten years; whereas this relation was only substantial in the European (P = 2.10×10− 9) and South Asian (P = 0.001) populations but not in the African population (P = 0.708).
Table 2
Associations of covariates with smoking during pregnancy and child outcomes in the three ethnic populations
health outcome
|
covariates
|
β (se, P) for European
|
β (se, P) for South Asian
|
β (se, P) for African
|
smoking during pregnancy a
|
smoking for height
|
income
|
-0.058 (0.010, 1.24×10− 9)
|
-0.168 (0.156, 0.279)
|
0.055 (0.146, 0.704)
|
smoking for body size
|
income
|
-0.057 (0.010, 2.29×10− 8)
|
-0.057 (0.198, 0.773)
|
-0.049 (0.162, 0.763)
|
child outcome b
|
height
|
birthweight
|
-0.444 (0.010, < 0.001)
|
-0.514 (0.079, 7.70×10− 11)
|
-0.243 (0.089, 0.006)
|
|
breastfeeding
|
-0.085 (0.012, 2.75×10− 13)
|
-0.134 (0.115, 0.244)
|
-0.045 (0.165, 0.787)
|
|
income
|
-0.008 (0.010, 0.443)
|
0.022 (0.057, 0.696)
|
-0.005 (0.064, 0.943)
|
|
gender
|
-0.060 (0.010, 2.10×10− 9)
|
-0.191 (0.057, 0.001)
|
-0.024 (0.065, 0.708)
|
body size
|
birthweight
|
0.166 (0.011, 6.61×10− 54)
|
0.128 (0.100, 0.203)
|
0.246 (0.092, 0.007)
|
|
breastfeeding
|
-0.191 (0.013, 4.20×10− 52)
|
-0.647 (0.130, 5.90×10− 7)
|
-0.362 (0.170, 0.033)
|
|
income
|
-0.029 (0.011, 0.007)
|
0.101 (0.072, 0.163)
|
-0.179 (0.071, 0.012)
|
|
gender
|
-0.271 (0.011, 1.45×10− 133)
|
-0.266 (0.071, 1.88×10− 4)
|
-0.814 (0.074, 2.38×10− 28)
|
a When smoking during pregnancy was the outcome (binary), we used logistic regression to assess the effect of income on smoking during pregnancy; b when child growth trait (height or body size) was the outcome (binary), we also applied logistic regression to assess the effect of birthweight, breastfeeding, income, and gender.
Confounding structure for body size
We further examined the relation between potential covariates and offspring body size (Table 2), and discovered that higher income was related to reduced risk of smoking during pregnancy in the European population (P = 2.29×10− 8); but this negative association was not replicated in the South Asian (P = 0.773) or African (P = 0.763) population. As can be expected, higher birthweight was strongly associated with the risk of being fatter in the European (P = 6.61×10− 54) and African (P = 0.007) populations except for the South Asian population (P = 0.203). Breastfeeding and gender were found to significantly be related to the risk of being fatter in all the three populations. In addition, higher income was related to decreased risk of being fatter in the European (P = 0.007) and African (P = 0.012) populations, but to increased risk of being fatter in the South Asian population although such relation was non-significant (P = 0.163).
Relationship between smoking during pregnancy and offspring height or body size at the age of ten years in the three ethnic populations
According to the above results, we could conclude that there existed evidently structural differences among the three ethnic populations. Under this context, we further studied whether a substantial association between smoking during pregnancy and offspring height or body size at the age of ten years could be consistently identified across these populations.
Estimated relation in the European population
In the European population, we found that smoking during pregnancy had substantial influence on offspring's height at ten years old (P = 5.16×10− 42), indicating that the risk of being shorter would increase approximately 15.3% (95%CIs: 13.0 ~ 17.7%) for individuals whose mother smoked during pregnancy (Table 3). Except for income (P = 0.561), other covariates (e.g., birthweight, breastfeeding, and gender) had significant impact on offspring's height at ten years old (P < 0.05). Higher birthweight would reduce the risk of being shorter at age ten (OR = 0.65, 95%CIs: 0.63 ~ 0.66); boys or breastfed offspring often had lower risk of being shorter, with the risk decreased approximately 6.2% (95%CIs: 4.3 ~ 8.0%) and 7.2% (95%CIs: 5.1 ~ 9.3%), respectively.
Next, we identified that smoking during pregnancy could affect offspring body size at ten years old (P = 1.52×10− 132), with the risk of being fatter increasing approximately 32.4% (95%CIs: 29.5 ~ 35.4%). Except for birthweight which had a positive association with fat body size at ten years old, all remaining covariates such as breastfeeding, income, and gender had an inverse association with fat body size. Specifically, greater birthweight resulted in higher risk of being fatter at age ten, with an estimated OR of 1.20 (95%CIs: 1.17 ~ 1.22); for boys, breastfed individuals, or those with higher household income (≥£31,000), the risk of being fatter would reduce approximately 24.4% (95%CIs: 22.7 ~ 26.0%), 15.8% (95%CIs: 13.7 ~ 17.8%), or 2.5% (95%CIs: 0.4 ~ 4.6%), respectively.
Estimated relation in the South Asian and African populations
These associations also largely held in the South Asian and African populations, although sometimes the estimated effect sizes were not always significant due to low power of insufficient samples (Table 3 and Figure S1). Specifically, it was shown that smoking during pregnancy was significantly related to increased risk of being shorter in the South Asian population (OR = 1.49, 95%CIs: 1.07 ~ 2.07) but only marginally significant in the African population (OR = 1.30, P = 0.095). In addition, smoking during pregnancy was significantly related to increased risk of being fatter in the African population (OR = 1.48, 95%CIs: 1.05 ~ 2.07) but non-significant in the South Asian population (OR = 1.08, P = 0.746).
Table 3
Associations between smoking during pregnancy and offspring height or body size at ten years old in the three ethnicities
covariates
|
β (se, P) for height
|
β (se, P) for body size
|
European
|
|
|
maternal smoking
|
0.143 (0.011, 5.16×10− 42)
|
0.281 (0.011, 1.52×10− 132)
|
birthweight
|
-0.437 (0.010, < 0.001)
|
0.180 (0.011, 4.74×10− 63)
|
breastfeeding
|
-0.075 (0.012, 1.02×10− 10)
|
-0.172 (0.013, 4.09×10− 42)
|
income
|
-0.006 (0.010, 0.561)
|
-0.026 (0.011, 0.019)
|
gender
|
-0.064 (0.010, 1.76×10− 10)
|
-0.279 (0.011, 2.20×10− 141)
|
South Asian
|
|
|
maternal smoking
|
0.396 (0.170, 0.020)
|
0.076 (0.234, 0.746)
|
birthweight
|
-0.517 (0.079, 6.36×10− 11)
|
0.127 (0.101, 0.205)
|
breastfeeding
|
-0.130 (0.115, 0.260)
|
-0.645 (0.130, 6.53×10− 7)
|
income
|
0.024 (0.057, 0.673)
|
0.101 (0.072, 0.162)
|
gender
|
-0.191 (0.057, 0.001)
|
-0.266 (0.071, 1.87×10− 4)
|
African
|
|
|
maternal smoking
|
0.262 (0.157, 0.095)
|
0.388 (0.172, 0.024)
|
birthweight
|
-0.243 (0.089, 0.006)
|
0.247 (0.091, 0.007)
|
breastfeeding
|
-0.028 (0.166, 0.865)
|
-0.326 (0.171, 0.057)
|
income
|
-0.005 (0.064, 0.937)
|
-0.178 (0.071, 0.012)
|
gender
|
-0.021 (0.065, 0.747)
|
-0.811 (0.074, 4.45×10− 28)
|
As anticipated, the non-significant associations discovered above were likely due to low power of small samples in the South Asian and African populations. Particularly, in the African population, the power associated with the current number of individuals (N = 5,000) was estimated to be 38% for height, and 19,128 individuals would be required for an expected power of 90% (Figure S1A). In the South Asian population, the power corresponding to the number of currently available individuals (N = 4,775) was estimated to only 6.2% for body size, and 500,532 individuals would be needed for an expected power of 90% (Figure S1B).
Estimated relation for stratification analysis in the European population
Further stratification analysis in the European population indicated the absence of heterogeneous effect of smoking during pregnancy on offspring height or body size at age ten (P > 0.05) for all other covariates except for gender (Figure S2). We identified that smoking during pregnancy could lead to higher risk of being shorter (OR = 1.18 vs. 1.13, P = 0.021) but lower risk of being fatter (OR = 1.28 vs. 1.35, P = 0.021) for boys at ten years old compared to girls. We could not carry out the stratification analysis in the South Asian and African populations due to their small sample sizes.
Relationship between smoking during pregnancy and offspring height or body size at the age of ten years in the European sibling comparison study
In the European sibling cohort, smoking during pregnancy also significantly affected offspring height (P = 2.10×10− 3) and body size (P = 7.31×10− 10) at ten years old with a consistent effect direction as in the European full cohort (Table 4). Briefly, in these sibling pairs, the risk of being shorter than average would increase approximately 12.6% (95%CIs: 5.0 ~ 20.3%) and the risk of being fatter than average would elevate approximately 36.1% (95%CIs: 26.3 ~ 45.9%) for individuals whose mother smoked during pregnancy. Additionally, birth order was significantly associated with both height (P = 4.30×10− 5) and body size (P = 2.24×10− 12). Interestingly, the above estimates, also including estimated effects for other covariates, were in line with those generated in the full European cohort, although some of them were no longer significant largely due to limited samples in the sibling cohort.
Table 4
Estimated effects of maternal smoking during pregnancy on offspring height or body size at ten years old in the European sibling pairs of the UK Biobank cohort
factor
|
β (se, P) for height
|
β (se, P) for body size
|
maternal smoking
|
0.119 (0.039, 2.10×10− 3)
|
0.308 (0.050, 7.31×10− 10)
|
birthweight
|
-0.482 (0.034, < 0.001)
|
0.263 (0.041, 9.20×10− 11)
|
breastfeeding
|
-0.054 (0.041, 0.189)
|
-0.059 (0.053, 0.260)
|
income
|
0.043(0.032, 0.180)
|
0.039 (0.040, 0.328)
|
gender
|
-0.047 (0.032, 0.142)
|
-0.372 (0.041, < 0.001)
|
birth order
|
0.116 (0.028, 4.30×10− 5)
|
0.242 (0.034, 2.24×10− 12)
|
Results after taking genetic factors into account
Furthermore, after taking genetic factors into account by including a GRS of growth trait, we discovered that smoking during pregnancy was still significantly associated with offspring's height and body size at ten years old especially in the European cohort, with the estimated effects nearly unchanged compared to those generated without considering GRS (Tables S5-S7). Note that, the GRS itself was positively correlated with height or body size in the European population (Table S8); the negative relation between the GRS and height in the South Asian and African populations was possibly due to racial heterogeneity, which resulted in the poor performance of genetic transferability across diverse populations [37, 38].