Age at first birth and risk of later-life cardiovascular disease: a systematic review of the literature, its limitation, and recommendations for future research

As will be highlighted more in depth in latter sections of the results, the appointment of reference category, operalization of exposure categories, and adjustment for covariates is highly divergent between studies, rendering it difficult to draw straightforward conclusions. Therefore, we broadly synthesize the results per cardiovascular outcome, but encourage readers to refer to Tables 3 and 4 in which information on risk categorization, reference group, sample size and number of events by exposure category, point estimates, confidence intervals, and covariate choices are schematically displayed. In these tables we also highlight potential sources of biases related to these parameters.

In the five cohort studies reporting on CVD mortality, young AFB was positively associated with CVD mortality. In one of the five studies [32], the confidence interval crossed the one (HR: 1.15, 95%CI: 0.99–1.3) (Table 4).

In four [9, 10, 25, 32] out of six studies reporting on CHD mortality, the youngest AFB group was at highest risk for CHD mortality. In three of these studies [10, 25, 32], the confidence interval crossed the 1 (Table 4). In the fifth study [28], women with an AFB of 25–29 had a lower HR (0.75, 95%CI: 0.6–0.95) compared to women with an AFB of 20–24 (reference category). Women in the youngest AFB group, <20, were not compared to women in the lowest risk group (AFB of 25–29) and the CI largely overlapped 1 when comparing those <20 to the reference group of AFB 20–24 (HR: 0.93, 95%CI: 0.7–1.3). In the sixth study [36], women with an AFB of ≥26 and women with an AFB of <23 were at higher risk compared to women with an AFB of 23–25, but this finding was not statistically significant (HR: 1.09, 95%CI: 0.8–1.5 and HR: 1.17, 95%CI: 0.9–1.5 respectively).

One case-control study [37] reporting on sudden cardiac death observed a positive association of young AFB with sudden cardiac death when adjusted for age. This association was attenuated when the authors adjusted for smoking, but no longer for age (Table 3).

In all four cohort studies [9, 25, 32, 36] reporting on cerebrovascular death (stroke), the women with the youngest AFB were at greatest risk. All associations attenuated in the fully adjusted models with point estimates ranging from 1.14 to 1.4 and confidence intervals just crossing the 1 (Table 4).

One study [28] reported on mortality from ischemic stroke and hemorrhagic stroke. Women with an AFB <20 compared to women with an AFB of 20–24 had an HR of 1.23 (95%CI: 0.96–1.6) and 1.09 (95%CI: 0.9–1.3) for ischemic and hemorrhagic stroke, respectively.

One cohort study [38] reporting on subarachnoid hemorrhagic stroke observed an increased relative risk for women with a first birth at age 26–30, as well as for women aged older than 30 at first birth, compared to women <26 at first birth. In an additional model, in which AFB was analyzed as a continuous variable, the study observed an increased relative risk for subarachnoid hemorrhagic stroke for each year increase in age at first birth (Table 4).

All three studies [24, 31, 34] reporting on acute non-fatal myocardial infarction, observed women with an early AFB at highest risk (Table 3). It is important to note that the reference category is nulliparous women in the study by La Vecchia et al. [31] In the study by Bertuccio et al. [24] the association attenuated in their fully adjusted model.

One study [30] observed a higher log odds ratio (0.36, P < 0.01) for the presence of heart disease in women with an AFB of <20 compared to 20 or older (Table 4).

One study [30] observed no evidence of an association of an early AFB and the presence of stroke (Table 4).

The vast majority of studies investigate early AFB rather than late AFB. The age cut-offs differ from one study to another. The youngest AFB category was defined as follows: one study used <19; 12 studies used <20; two studies used <21; and one study each applied <23, <25, <27 as cut-offs. One study [34] used both <18 and <20 as an early AFB cut-off. Another study [33] used a cut-off of <26 and additionally, looked at AFB as a continuous variable. The oldest AFB cut-offs were ≥31 in two studies, ≥32 and ≥33 in one study each and, ≥35 in two studies, all other studies had a lower cut-off (Table 3). In general, authors did not justify their choice of cut-off.

The studies vary in the appointment of the reference group. A small number of studies selected the youngest or oldest age category as the reference category, while most studies selected one of the intermediate age groups as the referent group. A few studies include nulliparous women in their analyses and appoint that group as the reference group. Most studies limited their analyses to parous women.

There is large variation in study sample size, years of follow-up and age at follow-up (Tables 1 and 2). Studies with a small sample size, short follow-up, young age at follow-up, or a combination of the three are likely to record fewer CVD events. Four cohort studies [9, 10, 27, 39] report on one or more associations between AFB and a CVD event with fewer than 150 total events.

A large variety of potential confounding variables were included in the studies. Only half of the studies provided explanations for their covariate choices [9, 10, 22, 23, 25, 28‐30, 35, 39]. Explanations included “established risk factors” [39], “potential confounding effects” [23] and “a priori defined plausible confounder” [28]. Three articles [22, 29, 30] explicitly presented a theoretical framework, while for the other studies a theoretical underpinning was implied. A number of studies present different statistical models in which they increasingly adjust for covariates [9, 22, 24, 25, 29, 35]. While intermediate variables are hypothesized, none of the articles applied methods such as structural equation modelling or mediation modelling to inform theories on possible pathways. In the fully adjusted models, all but one study adjusted for age. Eight studies adjusted for one or more other demographic covariates besides age. Eleven studies adjusted for one or more socio-economic covariates. Eight studies adjusted for one or more biomedical covariates. Nine studies adjusted for one or more health behaviors. Nine studies adjusted for one or more reproductive health covariates. Two studies [26, 31] with limited statistical adjustment mention that such adjustment for covariates did not alter their results, and thus presented only the unadjusted models.

Twelve studies [9, 22, 23, 25, 27‐31, 34, 35, 38] observed an association between at least one of the exposure groups and the reference group. Five studies did not propose a pathway to explain the results, two studies proposed a physiological pathway and five studies considered a combination of physiological and social pathways.

Almost none of the studies investigated the extreme ends of the distributions for reproductive ages. Only two studies [26, 34] examined adolescent childbirth (<19) and none of the studies looked at very young childbirth (<17). Yet, research on the association between AFB and all-cause mortality suggests the greatest mortality occurs among the youngest women [9]. Most studies used <20 as the youngest age category, while six used a higher cut-off (<21, <23, <25, <26 or <27).

Differences in the operationalization of the exposure groups likely contributed to the heterogeneity of the study results. Out of the six studies that used a higher cut-off (<21, <23, <25, <26 or <27) for their youngest age category, none reported an association of young AFB and CVD after covariate adjustment. If indeed the adolescent (<19) AFB-group is at highest risk for the studied outcome, combining that group with an older AFB-group, which is at lower risk, would lead to misclassification bias due to imperfect specificity and would bias the results towards the null-hypothesis. It is important to recognize that a number of studies reported on AFB as a secondary outcome and were not specifically designed to investigate the association between AFB and CVD. Furthermore, some studies indicate being unable to analyze early AFB due to a low number of individuals in that exposure category (college student cohort [27] and Nurses’ Health Study [26]).

The choice of reference category in several studies influences the interpretation of the reported findings. For some of the articles, it may have ‘masked’ an association with AFB. For example, Gallagher et al. [28] would have observed a lower HR for women with an AFB of 25–29 and possibly women with an AFB of ≥30, if they had defined AFB <20 as their reference group, rather than AFB 20–24. Also, a number of studies used nulliparous women as their reference group. For some studies, nulliparous women were combined with women who had not yet given birth [10, 31] or with an older AFB group [23]. For these, we cannot distinguish whether any observed associations are due to AFB, parity, or both. Thus, observed associations may be due entirely or partially to another factor (e.g. parity, infertility) and not the AFB category under examination. Moreover, when younger women who have not yet given birth are grouped together with nulliparous (older) women (who may have an underlying health condition related to their fertility status and CVD) [40], it may bias the findings.

Factors such as sample size, the age of the cohort, and follow-up time contributed to a low number of cases with disease in some studies, rendering it difficult to interpret reported findings. One study [27], for example, that observed a higher rate of CHD amongst women with an AFB of 33–43, compared to women with an AFB of 25–29, only recorded a total of 33 CHD cases in all parous women in the study. Another study [9], which followed women for 5 years, was conducted amongst relatively young women (aged 30–45 during follow-up). Despite the short follow-up and the young age of the women, the total number of CVD events among parous women in this study was 151, which was sufficient to reveal higher mortality for CVD amongst women with an AFB of <20 compared to an AFB of 20–29. However, the same study also examined the association between AFB and coronary heart disease and cerebrovascular events separately. Likely, the number of cerebrovascular events alone was too low to detect a difference between AFB categories. The authors do not report the number of cerebrovascular events.

The distribution of the populations studied in terms of reproductive exposure (AFB) may be restrictive. For example, the study by Colditz et al. [26] was conducted amongst registered nurses and as the authors highlight in the discussion, this sample may not be representative of US women in general, as only 1 % had an AFB of <20. This was likely the case for several other studies included in the review, such as Beard et al. [23] who only included married women.

Lastly, selective survival may have influenced findings in the studies that began following women later in life. Numerous studies have documented associations between early AFB and all-cause mortality [7‐10]. Thus, associations may converge as a larger proportion of the highest risk women are selected out of the study sample.

A very wide range of covariates were included in many of the statistical models. Some models adjusted only for age, while others adjusted for several socioeconomic, biomedical, health behaviors and reproductive covariates. Few studies justified their choices for inclusion of covariates and few of the studies that did observe an association provided a thorough interpretation of their results. In the absence of more comprehensive conceptual frameworks, it is difficult to ascertain whether the statistical adjustment was sufficient, insufficient, or excessive. As Barclay et al. [22], Henretta [30], and Grundy and Kravdal [29] highlight in their studies, there might be different pathways by which reproductive exposure relates to (cardiovascular) disease. On the one hand, background variables such as childhood SES, as well as personality traits, may contribute to early childbearing (selection effect) [13]. At the same time, an early AFB could have consequences for SES by reducing future educational and occupational opportunities. It is difficult to discern to what extent early parenthood circumstances influence the relation between background SES and later life SES. As highlighted before, there is a similar uncertainty regarding physiological pathways. High childhood BMI is related to an early menarche [41], which in turn gives rise to more ‘opportunity’ for an early AFB (a selection effect). On the other hand, an early pregnancy may cause a higher BMI, since early pregnancy is related to greater weight gain during pregnancy compared to pregnancy in adulthood [18, 19]. And again in this relationship, childhood BMI is related to BMI in later life [41, 42]. It is therefore difficult to ascertain whether there is a need to statistically adjust for many of these covariates, as they may be intermediate variables.

Extensive statistical adjustment may have masked “real” associations between AFB and CVD. In a recent well-conducted study of Swedish women at age 40, younger age at first birth was associated with hypertension [43], which is one of the most important risk factors for CVD. When studies adjust for biomedical indicators such as BMI, hypertension, and cholesterol, which are markers of CVD, they implicitly assume an alternative pathway linking AFB to CVD that is not mediated by the classic risk factors. None of the authors who adjusted for such biomedical indicators provided a justification for an alternative pathway. They may have therefore over-adjusted their statistical models. Under adjustment may also be a concern. One study [38] that observed a higher risk for subarachnoid hemorrhage among women with an AFB of 26–30 and ≥31, compared to women with an AFB of ≤25, did not report adjusting for age. Since the women were tracked from the time of their first birth (between ‘78 and ‘87) until the end of follow-up in 2003, age at first birth was directly linked to participant age during follow-up. Women with a young age at first birth were younger compared to women with an older age at first birth in this sample and incidence rates of subarachnoid hemorrhage increase with age [44].

Overall, and acknowledging the limitations above, if we solely examine the results of the 14 studies that defined their young age at first birth group as <20 or younger, we observe the youngest AFB group at highest risk in 21 out of the 23 studied associations. In 13, the association was statically significant. As for the ten remaining associations from ten different studies, in three [24, 25, 37], the association was significant in the analyses that were age-adjusted only. In one study [28], an association would have likely been observed if the young age at first birth group had been compared to the age at first birth category with the lowest risk and, in five studies a low number of events [9, 10, 25] or a low number of women in the young age at first birth category [23, 26] likely limited study power.