We appreciate the comments by Rickard et al.  of our recent paper which evaluated the effects of bearing twins on maternal reproduction and survival using a large sample of women in the Utah Population Database (UPDB) . While they concur with our major finding showing improved post-reproductive survival among women bearing twins compared with mothers of singletons only, they report results of a simulation using synthetic data that questions the interpretation of one of our key results. Specifically, Rickard et al. challenge whether the higher lifetime fertility of mothers who have produced twins in our study reflects their propensity for twinning or is the inevitable result of accumulating more opportunities to twin than women who gave birth fewer times. We address this alternative hypothesis by re-examining the UPDB using two analytical approaches: parity progression ratios (PPRs) and odds ratios based on logistic regression. Our results demonstrate that the higher fertility of twinning mothers in the UPDB compared with non-twinning mothers is not consistent with the Rickard et al. hypothesis. We also question two assumptions in Rickard et al.'s simulation and suggest that these assumptions could explain how their findings diverge from our results using UPDB data directly.
First, we compared PPRs for 4603 mothers of twins to 54 183 non-twinning mothers who were separated into two birth cohorts: pre-1870 and post-1870, to examine the effects of the demographic transition. PPRs are a standard demographic technique that measures the proportion of women at a given parity who proceed to have an additional child. In this instance (figure 1), we compared the proportion of twinning mothers who bear an additional child relative with non-twinning mothers at that same parity. For the pre-1870 cohort in which natural fertility conditions were largely prevalent, twinning mothers had equal or higher PPRs than their non-twinning counterparts across all parities. The post-1870 birth cohort followed a similar but attenuated pattern, reflecting the onset of the demographic transition.
Second, we calculated the ratio of PPRs between twinning and non-twinning mothers. When the ratio is at 1, both twinning and non-twinning mothers progress to the next parity equally. The pattern illustrated in figure 2 indicates that twinning mothers are more likely to advance to the next parity than the non-twinning mothers, especially at lower and higher parity steps, most apparent in the earlier birth cohort. Figure 2 also shows that mothers of twins begin reproducing more quickly (their initial PPRs are higher) and continue doing so, whereas non-twinning mothers advance to higher parities more slowly during their early years of reproduction. Non-twinning mothers appear to ‘catch-up’ with the twinning mothers at middle parities (5–6, 6–7, 7–8), but twinning mothers again surpass them at higher parities. In the pre-1870 birth cohort, the number of women at these high parities is significant; the median parity for this subset is 9, therefore representing more than half the mothers in this group. These PPR estimates support our maternal heterogeneity hypothesis proposing that mothers of twins have reproductive advantages over their singleton-only bearing counterparts owing to a more robust phenotype.
Third, we report in table 1 odds ratios (and their 95% confidence intervals) based on cohort-specific (mothers born before 1870 versus after) logistic regressions that control for birth year, age at marriage, whether the previous child had died, whether the husband was still living and affiliation with the Church of Jesus Christ of Latter-day Saints (LDS status). These odds ratios refer to the odds that a mother of twins will have an additional parity relative to mothers who did not have twins for a given parity, controlling for these additional variables. For instance, the odds that a woman with two children will have a third (controlling for age at marriage, mother's birth year, previous child's death, whether the spouse was alive and LDS status) is 68 per cent higher for the pre-1870 cohort and 59 per cent higher in the post-1870 cohort if the mother had already produced twins. These results confirm the same pattern seen in the PPR comparisons. Mothers who previously twinned had higher odds ratios at all parities (except intervals 5–6, 6–7, 7–8 in the post-1870 cohort) and, again, the effects were largest at lower and higher parities.
These results based on UPDB data show that twinning propensity and higher fertility cannot be due to the statistical artefact that Rickard et al. propose because our analyses are based on mothers who have twinned compared with those who have not with the identical number of existing children. Furthermore, the odds ratio results show that women who have had twins are more likely to progress to another parity than a woman who did not have twins at that parity net of other confounders, an effect that is strongest during natural fertility conditions. Together, these results are inconsistent with the suggestion that the basis for our original finding was because women who have more children are more likely to bear twins.
Finally, we have concerns about assumptions in the Rickard et al. simulation model which may account for our divergent findings. First, it is well established that twins suffer higher infant mortality rates than singletons . In our article, we accounted for the possibility that mothers of twins may have higher lifetime fertility rates as a result of child replacement by controlling for child mortality before the age of 18 years old. While the fertility advantage of twinning decreased after accounting for this factor, the fertility of twinning mothers frequently remained statistically greater than non-twinning mothers. Second, the Rickard et al. model assumes a constant probability of twinning for women of all ages even though they acknowledge that increases in twinning at later maternal ages has been well established . In our article , we account for this phenomenon (that the probability of twinning increases with maternal age and, therefore, also by parity) by introducing a statistical interaction term (twin × parity) into our mortality analyses. We found a significant survival difference between twinning and non-twinning mothers even after controlling for the parity effect. We also account for twinning variation at each parity progression herein. Rickard et al. do not control for either high infant mortality on maternal interbirth intervals or on lifetime fertility nor the differing probabilities of twinning across maternal ages and parities in their simulations.
In our original article we demonstrated, after controlling for many potential confounders, that mothers who bear twins also have higher than expected lifetime fertility owing to shorter interbirth intervals, longer reproductive span, later ages at last birth as well experiencing longer lives which sets them apart from their non-twinning counterparts in this population. We were careful not to suggest that twinning is the catalyst for this pattern of heterogeneity. Rather, we argue that there is a variation in maternal capacity to bear the costs of twinning and those women who twin reflect a robust phenotype which is also reflected in their enhanced survival and other measures of reproduction. The analyses we present here further strengthen our original findings. PPRs and odds ratios show that for a given parity mothers of twins were more likely to progress to an additional birth than non-twinning mothers, especially at low and high parities. These results confirm our conclusion that mothers of twins have higher lifetime fertility as a result of a robust phenotype.
This study was supported by The National Institute of Ageing grant no. AG022095 (The Utah Study of Fertility, Longevity and Ageing). We thank the Pedigree and Population Resource (funded by the Huntsman Cancer Foundation) for its role in the ongoing collection, maintenance and support of the Utah Population Database (UPDB).
The accompanying comment can be viewed at http://dx.doi.org/10.1098/rspb.2012.0191.
- Received February 24, 2012.
- Accepted March 22, 2012.
- This journal is © 2012 The Royal Society