Revisiting the association between maternal and offspring preterm birth using a sibling design

Preterm birth (PTB) is one main factor associated with infant mortality and morbidity in industrialized countries [1, 2]. Preterm infants, especially those born earliest, are more likely to experience a myriad of health-related problems, from neurodevelopmental problems in childhood to cardiovascular and metabolic disorders later in life [3].

The etiology of preterm birth is multifactorial, complex and not well understood [4]. One important piece in the puzzle is the intergenerational recurrence of PTB. Several studies have consistently shown that mothers born preterm are more likely to have preterm babies [5‐8] although the mechanisms behind the association remain unclear. A meta-analysis summarizing unadjusted estimates recommended including maternal birth status in risk assessment when planning pregnancy care for individual women [9]. Meta-analyses of observational studies, however, should be taken with caution since spurious summary results may result from systematic biases shared by the included studies [10]. Although randomized trials are not an option to study this association, previous studies have rarely used observational designs for causal inference, such as sibling designs [11]. Being born preterm may not necessarily be an individual risk factor for delivering preterm infants. Rather, it may merely reflect the persistence of unmeasured environmental and family (genetic and behavioral) risk factors transmitted from one generation to another [12‐15]. Quasi-experimental sibling studies are well suited to isolate the causal effect of an individual risk factor not shared by siblings.

Our goal was to assess whether maternal preterm birth status is an individual risk factor per se for delivering preterm. To distinguish the risk of delivering preterm among women born preterm themselves from that of women born at term but with a family history of preterm birth, we used fixed-effects methods to compare preterm birth rates among the offspring (cousins) of sister mothers of discordant preterm birth status. Furthermore, we distinguished between spontaneous and provider-initiated PTB, which may be differentially involved in the recurrence of PTB across generations [16‐19].

There were 58,628 mothers born on or after April 1st, 1980 who gave birth to 116,790 children from April 1st, 1995 to March 31, 2016. We excluded mother-infant pair records that met one or more of the following exclusion criteria. Missing maternal gestational age (N = 17,687), extreme maternal gestational age (N = 242), mother born from multiple pregnancy (N = 356), missing child’s gestational age (N = 4613), child is a multiple (N = 2735) and extreme child’s gestational age (N = 4824). After exclusion of 19,055 mothers (32.5%) and 37,592 children (32.2%), the population cohort for analyses was composed of 39,573 mothers and 79,198 children. Of these mothers, 28,133 (71.1%) did not have a sister giving birth during the study period.

Among the 11,440 sisters, 1033 (9.0%) had discordant preterm birth status (455 born preterm and 578 at term) (Table 1), 69 (0.6%) were all preterm and the vast majority of sisters (n = 10,338; 90.4%) were born at term. Overall, offspring preterm birth rates were 6.2 per 100 live births in the population. Mothers without sisters giving birth in the cohort were more likely to deliver preterm if they themselves were born preterm. Among sister mothers with discordant PTB status, offspring preterm birth rates did not differ (10.2% among sisters born preterm and 10.0% among sisters born at term). The highest PTB rate occurred in a small number of families where all sister mothers were born preterm (13.1%) and the lowest rate where all sister mothers were born at term (6.1%).

Infant’s higher birth order was more common among mothers with sisters in the cohort, particularly among discordant sisters, as were other characteristics associated with adverse birth outcomes, such as teenage pregnancy, low socioeconomic status, residing in the rural North (mostly inhabited by Indigenous populations), history of substance abuse, smoking during pregnancy, inadequate prenatal care, receiving income assistance and low maternal education (Table 1). Infants whose mothers or aunts were born preterm had the most adverse profiles, which differed little between discordant sisters.

In the population cohort analysis, infants exposed to maternal preterm birth had higher risk of preterm birth before, and after adjustment [Adjusted Relative Risk (ARR): 1.38; 95% CI: 1.25, 1.55] (Table 2). The association was stronger when the mother’s gestational age was < 32 weeks (ARR: 1.77; 95% CI: 1.28, 2.42). In the sibling analysis, comparing sisters of discordant PTB status, maternal PTB was no longer associated with offspring PTB before or after adjustment overall (ARR: 1.02; 95% CI: 0.77, 1.34) or according to length of gestation.

Analyses splitting PTB into types in the whole population indicate that both maternal spontaneous and provider-initiated PTB were associated with spontaneous offspring PTB (Table 3), although maternal provider-initiated PTB was no longer associated with offspring PTB in analyses restricted to one child per mother. Maternal preterm birth types were not associated with offspring provider-initiated PTB before or after adjustment.

In the siblings’ analysis (Table 4) there was no statistically significant intergenerational recurrence patterns, with the possible exception of spontaneous PTB. Mothers of spontaneous PTB status were more likely to deliver spontaneous PTB infants [ARR: 1.39; 95% CI: 0.99, 1.93]. This borderline association barely changed when controlling for additional covariates but weakened after accounting for the recurrence of PTB within births of the same women [ARR: 1.26; 95% CI: 0.73, 2.20].

We found that maternal preterm birth status was positively associated with offspring preterm birth in the general population of Manitoba but not in a subcohort of sisters with discordant preterm birth status who subsequently gave birth. In the sister subcohort, preterm delivery was equally elevated among women born preterm themselves and among those who were born at term but had a sister who was born preterm. Observed intergenerational patterns were more apparent for spontaneous PTB.

Our findings for the population cohort align remarkably well with those of previous studies. In the population cohort we obtained an unadjusted RR: 1.42 (1.27, 1.58), similar to that of a meta-analysis involving seven studies with low heterogeneity [RR: 1.41 (1.26, 1.59)] [9]. Like others, we also found a dose-response association with mothers born at earlier gestation in the population cohort [15]. However, the apparent lack of association in our sibling analyses indicates that the higher risk observed among women born preterm was shared by their sisters not born preterm, suggesting that the actual risk factor for preterm birth may not be maternal preterm birth status per se but a broader familial occurrence of preterm birth. Two studies using different designs also found a higher risk of delivering preterm shared by sisters of concordant or discordant PTB status but not by brothers [30, 31], which suggests that paternal and fetal genes may not play an important role and that the transmission of PTB mainly occurs through the female line [15, 30, 31].

We also observed that sisters of concordant PTB status exhibited extreme outcomes. PTB rates were lowest among sisters all born at term (6.1%) and highest among sisters all born preterm (13.1%). These findings are consistent with polygenic theory according to which higher recurrence rates occur among close relatives if more than one family member is affected [32]. Although our study points at familial female aggregation of PTB, it cannot specify the contributions of genetic and environmental factors involved, since the sibling design keeps these characteristics shared by siblings constant [29]. Studies on the heritability of preterm birth have shown that fetal and maternal genetic components play a role, although secondary to that of environmental factors [13, 14]. Further studies may elucidate the differential contribution of genetic and familial environmental influences.

In our analyses stratified by preterm birth types, there was no intergenerational association for infant provider-initiated PTB in the population cohort or in the sibling cohort. Infant spontaneous PTB exhibited clear associations with maternal spontaneous PTB in the population cohort across all models. In the sibling analysis, these associations were positive but not statistically significant presumably due to reduced statistical power to detect associations. Given this limitation, the possibility of a true maternal transmission of spontaneous PTB cannot be ruled out. Further studies may clarify this issue.

Our study has limitations. The use of administrative data collected for other purposes restricted the information available for analyses. At the same time, routine data collection on a relatively stable and large population over decades made this study possible. Our sibling design is particularly suited to draw causal inferences regarding an individual exposure that differs between siblings, provided that there is no exposure measurement error or confounding due to factors not shared by siblings [33]. Gestational age information was lacking for 19% of pregnancies, most of them affecting maternal birth status. A validation study indicated that PTB (< 37 weeks) in Canadian hospital records has sensitivity and specificity of 91.2 and 98.8%, respectively [23]. Measurement of gestational age is known to vary over time, being more likely to be influenced by the last menstrual period (LMP) when sister mothers were born than when they delivered. LMP may overestimate PTB rates [34], which would artificially increase maternal exposure to PTB and dilute associations, particularly in the siblings’ analyses [33]. However, in our data very and moderate PTB rates were not higher in the 1980’s and 1990’s, when mothers were born, and secular trends are consistent with those reported in North America [35]. Despite controlling by design for genetic and environmental factors shared by siblings, unmeasured individual risk factors not shared by sister mothers may contribute to some degree of residual confounding. Our sensitivity analysis adjusting for additional maternal risk factors (e.g., maternal smoking, education, prenatal care, no partner, receipt of income assistance) suggests that such bias would be small since estimates remained virtually unchanged. Previous studies have reported a similar low impact of covariate control [5‐8, 31]. We did not control for race or ethnicity, which may be warranted in a multiethnic population characterized by immigration and a large share of Indigenous people, who experience social and health disadvantage [36]. However, unlike in the population analyses, race/ethnicity is controlled for by the sibling design, which in turn partially controls for other characteristics correlated to race/ethnicity and reduces their variability between siblings. We did not control for maternal morbidity, which may be in the causal pathway. We could not include stillbirths, many of whom are preterm, because only live births receive a birth record linkable to the mother. Underestimation of preterm birth rates would be small due to the rare occurrence of stillbirths. Finally, we did not have usable data on fathers or partners who may have helped determine if women in the sibling cohort were full or half sisters. Paternal factors may also influence pregnancy courses, although not through their genes [15].

Our findings provide new insights into the study of intergenerational effects of PTB. First, our study provides additional evidence that PTB clusters within families and that maternal preterm birth status may not be a predictor of preterm delivery as sensitive as a family history of PTB. Interestingly, the risk of offspring PTB within families in which at least one mother-sister was born preterm was higher (≥10%) than among women born preterm without sisters (7.7%). Part of this excess risk may be due to poorer sociodemographic profiles aggravated by larger family size, so that the environmental factors contributing to PTB among women born preterm seem to affect equally their sisters not born preterm themselves. The association between socioeconomic disadvantage and adverse birth outcomes is well established not only cross-sectionally [12] but also across generations [37]. The clustering of PTB risk within families may also be influenced by gene-gene and gene-environment interactions. Anthropometric features, such as maternal height and body size, are also inherited to some extent and associated with PTB [38, 39]. The heritability of PTB has been estimated in the range of 15 to 40% [17], although an unknown part of the recurrence across generations may involve non-genetic factors [15].

Second, both the population and the sibling analyses indicated that provider-initiated PTB was not influenced by maternal or familial PTB. The possibility that the event of being born preterm, irrespective of the cause, may physiologically predispose women to deliver preterm later in life (i.e., adverse fetal programming) is not supported by our findings. A focus on spontaneous PTB may better serve to identify mechanisms involved in familial patterns of PTB.

Our findings do not support the notion of seeing maternal PTB status per se as an individual risk factor for preterm delivery. Rather, consideration of a female family history of PTB may better identify women at higher risk of preterm delivery and perhaps other pregnancy complications. The clustering of PTB within families suggests that family-based approaches to understanding the occurrence of preterm birth and its sequelae may be more fruitful than solely focusing on individual risk factors.