A Genetically Informed Study of the Associations Between Maternal Age at Childbearing and Adverse Perinatal Outcomes

Abstract

We examined associations of maternal age at childbearing (MAC) with gestational age and fetal growth (i.e., birth weight adjusting for gestational age), using two genetically informed designs (cousin and sibling comparisons) and data from two cohorts, a population-based Swedish sample and a nationally representative United States sample. We also conducted sensitivity analyses to test limitations of the designs. The findings were consistent across samples and suggested that, associations observed in the population between younger MAC and shorter gestational age were confounded by shared familial factors; however, associations of advanced MAC with shorter gestational age remained robust after accounting for shared familial factors. In contrast to the gestational age findings, neither early nor advanced MAC was associated with lower fetal growth after accounting for shared familial factors. Given certain assumptions, these findings provide support for a causal association between advanced MAC and shorter gestational age. The results also suggest that there are not causal associations between early MAC and shorter gestational age, between early MAC and lower fetal growth, and between advanced MAC and lower fetal growth.

Appendices

Appendix 1

In order to test carry-over effects, we fit a model that controlled for maternal age at first childbirth and birth order in a subsample of the data that excluded first-born offspring. We compared the results from this model to Model 1, the population model, which only controlled for birth order (see Figs. 3 and 4). In general, the analyses controlling for maternal age at first childbirth (see Table 3 and the second column of Figs. 3 and 4) were consistent with the results in the main text. The models showed a similar pattern of results for the association between MAC and GA, suggesting that differences in GA among offspring were not due to when a woman had her first child. Furthermore, the models showed similar results for the association between older MAC and BWGA. However, including maternal age at first childbirth as a covariate slightly attenuated the association between younger MAC and BWGA, suggesting that maternal age at first childbirth may partially account for the observed population-wide association between younger MAC and BWGA.

Fig. 3

Gestational age. The first column shows Model 1, the population model that only controlled for offspring birth order. The second column shows models that controlled for maternal age at first childbirth and offspring birth order in a subsample of the data that excluded first-born offspring. Ninety-five percent confidence intervals are shown for the binned Swedish models. The shaded regions on the U.S. continuous models represent 95 % confidence intervals

Fig. 4

Birth weight for gestational age. The first column shows Model 1, the population model that only controlled for offspring birth order. The second column shows models that controlled for maternal age at first childbirth and offspring birth order in a subsample of the data that excluded first-born offspring. Ninety-five percent confidence intervals are shown for the binned Swedish models. The shaded regions on the U.S. continuous models represent 95 % confidence intervals

Table 3 Test of carry-over effects

Appendix 2

We did not include offspring year of birth as a covariate in the main analyses because differences in offspring year of birth would be almost perfectly correlated with differences in MAC in the sibling comparison models. Thus, in order to test whether associations may be due to birth cohort effects (i.e., changes in society that have occurred over time) we fit population models that controlled for offspring year of birth. The population-wide models with (see Table 4 and the second column of Figs. 5 and 6) and without (see the first column of Figs. 5 and 6) year of birth included as a covariate yielded the same pattern of results, suggesting that birth cohort effects did not bias our interpretation of the results in the main text.

Fig. 5

Gestational age. The first column shows Model 1, the population model that only controlled for offspring birth order. The second column shows population models, which controlled for offspring birth order and offspring year of birth. Ninety-five percent confidence intervals are shown for the binned Swedish models. The shaded regions on the U.S. continuous models represent 95 % confidence intervals

Fig. 6

Birth weight for gestational age. The first column shows Model 1, the population model that only controlled for offspring birth order. The second column shows population models, which controlled for offspring birth order and offspring year of birth. Ninety-five percent confidence intervals are shown for the binned Swedish models. The shaded regions on the U.S. continuous models represent 95 % confidence intervals

Table 4 Test of birth cohort effects

Appendix 3

We re-ran the adjusted population models (Model 2) and the sibling comparison models (Model 4) in the Swedish sample, leaving out paternal covariates to examine the role of measured paternal characteristics (see Figs. 7 and 8 for population models and Figs. 9 and 10 for sibling comparison models). The model provided us with some insight into the possible confounding roles of measured paternal factors, which we did not have access to in the US sample. The models without paternal covariates (see Table 5 and column two of Figs. 7, 8, 9, 10) showed the same patterns of findings as the models with paternal covariates (see column one of Figs. 7, 8, 9, 10), suggesting that measured paternal characteristics did not account for the results in the Swedish analyses.

Fig. 7

Gestational age. The first columns shows population models, which control for offspring, maternal, and paternal covariates. The second column shows population models, which controlled for offspring and maternal covariates; paternal covariates were not included. Ninety-five percent confidence intervals are shown

Fig. 8

Birth weight for gestational age. The first columns shows population models, which control for offspring, maternal, and paternal covariates. The second column shows population models, which controlled for offspring and maternal covariates; paternal covariates were not included. Ninety-five percent confidence intervals are shown

Fig. 9

Gestational age. The first column shows sibling comparison models, which control for offspring birth order and paternal covariates. The second column shows sibling comparison models, which controlled for offspring birth order; paternal covariates were not included. Ninety-five percent confidence intervals are shown

Fig. 10

Birth weight for gestational age. The first column shows sibling comparison models, which control for offspring birth order and paternal covariates. The second column shows sibling comparison models, which controlled for offspring birth order; paternal covariates were not included. Ninety-five percent confidence intervals are shown

Table 5 Test of the role of paternal covariates

Appendix 4

Because GA and birth weight are correlated with birth order we fit a cousin comparison models restricted to the subsample of first-born individuals in the Swedish dataset (see Figs. 11 and 12) to assess whether birth order may have influenced the main results. The first-born cousin comparisons (see Table 6 and column two of Figs. 11 and 12) showed the same patterns of findings as the main analyses full-cousin comparisons (Model 3; see column one of Figs. 11 and 12), suggesting that our main results were not due to birth order effects.

Fig. 11

Gestational age. The first column shows the full-cousin comparison model and the second column shows the cousin comparison, which included only first-born cousins. Ninety-five percent confidence intervals are shown

Fig. 12

Birth weight for gestational age. The first column shows the full-cousin comparison model and the second column shows the cousin comparison, which included only first-born cousins. Ninety-five percent confidence intervals are shown

Table 6 Test of birth order effects

Appendix 5

To examine the clinical significance of the finding further, in the Swedish sample we re-ran models used in the main analyses to predict the log-odds of PTB (see Table 7 and Fig. 13) and found that the results were consistent with the GA findings from the main analyses. The results suggested the observed population-wide association between early MAC and increased risk for PTB was confounded by familial factors. However, the association between advanced MAC and increased risk for PTB was observed in all models, thus, providing support for an association between advanced MAC and increased risk for PTB, independent of measured covariates and unmeasured shared familial factors.

Fig. 13

Preterm birth. Ninety-five percent confidence intervals are shown

Table 7 Preterm birth results