Introduction

Cardiovascular disease (CVD) is one of the dominant causes of death and disability. It accounts for millions deaths worldwide each year and leads to be an important public issue1. Apart from age, hypertension, hyperlipidemia, diabetes, obesity and smoking, which are the most established risk factors for CVD, recent studies have hypothesized that sex hormone may play a role in the etiology of CVD1,2. Several studies have demonstrated that males are two to three times more likely to die prematurely from cardiovascular disease than females. This ratio persists in all countries independent of their overall death rate from this condition2. This theory may be affected by complex influences from sex hormone.

Pregnancy and parturition are important events in the life of a woman. Fluctuations of serum sex hormone levels during the process of pregnancy and delivery, perinatal hemodynamic changes, oxidative stress and other gestational factors exert complex influences on the cardiovascular system. Although previous studies have devoted great efforts to evaluate the relationship between childbearing history and mortality of CVD, the aforementioned association is still inconsistent. The association between parity and risk of CVD was first studied in the 1980s, which demonstrated the increased prevalence of CVD with parity number3. However, subsequent studies found minimal or no evidence for the aforementioned association3,4,5,6,7,8,9. Given the inconsistency of previous findings, we therefore carried out this systematic review and meta-analysis to summarize the evidence of the relationship between parity and CVD mortality.

Results

Study characteristics and quality assessment

Table 1 demonstrated the characteristics of the ten included prospective studies10,11,12,13,14,15,16,17,18,19, in which 16,601 cases and 978,209 non-cases was represented. Among the ten included studies, five were carried out in the United States11,15,17,18,19 and one for each in Israel10, Australia12, China14, Korea13 and Finland16. Cohort sizes ranged from 86718 to 585,45517 and the number of CVD cases varied from 4518 to 712514.

Table 1 Characteristics of studies of parity and CVD mortality.
Full size table

The information of study quality is described in Supplementary Table S1. Briefly, three studies13,14,15 were not assigned a star in the column of “representativeness of the exposed cohort” because these studies utilized special populations. Three studies13,16,17 were assigned two stars in the column of “control for important factor or additional factor” because they adjusted for more than two important confounders in the primary analyses. Two studies17,18 were not assigned a star in the column of “follow-up long enough for outcomes to occur” because the follow-up periods were less than ten years. Three studies12,13,19 were not assigned a star in the column of “adequacy of follow-up of cohorts” because the follow-up rates were less than 75%.

Ever parity versus nulliparous

Six studies11,14,15,17,18,19 focused on the relationship between ever parity and CVD mortality. The summary relative risk of CVD for the ever parity compared with nulliparous was 0.79 (95% CI, 0.60–1.06), with significant heterogeneity (I2 = 90.9%; P < 0.001; Fig. 1). We observed no publication bias through Egger’s test (P = 0.760) or Begg’s test (P = 0.260) (Supplementary Figure S1).

Figure 1
figure 1

Association between ever parity and cardiovascular disease mortality.

Full size image

Dose-response analysis of parity number

We included nine prospective studies10,11,13,14,15,16,17,18,19 in the dose-response analysis. The summary risk estimates for per live birth was 1.01 (95% CI, 0.97–1.05), with significant heterogeneity (I2 = 86.4%; P < 0.001; Fig. 2). In addition, we observed a significant nonlinear relationship between parity number and CVD mortality (P < 0.001). There was evidence of a J-shaped association in the non-linear dose-response meta-analysis of parity number and CVD mortality. The association of parity number with CVD mortality appears to follow an inversely linear dose-response pattern until the parity number reaches four live births. After this point, the association appears to rebound (Fig. 3).

Figure 2
figure 2

The relative risk(RR) for per live birth.

Full size image
Figure 3
figure 3

Dose-response analysis of parity number.

Full size image

Subgroup and sensitivity analysis

In the subgroup analysis of per one live birth and CVD mortality, we found non-significant results in the majority of the strata. However, there was a significant difference among the summarized results of the studies whether adjustment for cigarette smoking (Table 2), which might be partially responsible for the significant heterogeneity of the main result.

Table 2 Summary risk estimates of the association between parity number and cardiovascular disease mortality (per 1 live birth).
Full size table

In the sensitivity analysis, the six study-specific RR of the ever parity versus nulliparous ranged from a low of 0.76 (95% CI, 0.54–1.06; I2 = 92.5%; P < 0.001) after omission of the study by Jacobs and colleagues11 to a high of 0.91 (95% CI, 0.85–0.98; I2 = 0%; P = 0.957) after omission of the study by Jaffe and colleagues19. Additionally, The 9 study-specific RR of the parity number ranged from a low of 0.99 (95% CI, 0.96–1.02; I2 = 77.1%; P < 0.001) after omission of the study by Dior and colleagues10 to a high of 1.02 (95% CI, 0.98–1.06; I2 = 80.0%; P < 0.001) after omission of the study by Jaffe and colleagues19,20.

Discussion

Findings from this meta-analysis demonstrate a borderline significant reduced risk of CVD mortality in ever parity versus nulliparous women (RR = 0.79, 95% CI: 0.60–1.06). Additionally, non-linear dose-response analysis displays a potential J-shaped relationship between parity number and CVD mortality. We firstly comprehensively and quantitatively evaluate the relationship between parity and CVD mortality.

Several potential mechanisms might have been proposed but the exact biologic mechanisms are not fully understood. Previous studies showed that complicated metabolic changes, such as dyslipidemia, abnormal glucose tolerance and increased body mass index, developed with increasing parity number20,21. Besides, as pregnancy itself can be regarded as a state of increased insulin resistance, recurrent pregnancies may result in an additive effect on the later insulin resistance, which may give an explanation to the positive association between high (>4) parity and cardiovascular mortality16. The protective effect from moderate parity for CVD may link with the enhanced endothelial function in pregnancy, which results in greater bioavailable nitric oxide11. It is worth mentioning that increased endothelial function from pregnancy, unlike its concurrent metabolic change and other temporary disorders, may continue postpartum22.

As for the associations between endogenous estrogen exposure and CVD mortality, however, former studies have reported conflicting results13. Numerous pregnancies always result in prolonged exposure to high levels of estrogen and progesterone, which may reduce the risk of CVD. However, numerous pregnancies also relate to older maternal age, inflammation and oxidative stress, which are tightly associated with adverse predictors to CVD. It is also possible that higher fertility may reflect women who are better in general and therefore at relatively lower risk for CVD mortality11. However, increased risk of CVD was more likely to be attributable to lifestyle factors such as anxiety, stress, even fear of raising children6,13. Socioeconomic factors may also enhance the relationship between parity and CVD mortality, because both CVD and high parity have higher frequency in lower social classes10.

Compared to nulliparous, our meta-analysis shows that ever parity has a significantly protective effect against the death caused by CVD within a limited number of parity. Interestingly, as the 10 included cohorts displayed different viewpoints of parity and CVD mortality, the published information on this issue is scarce and inconsistent. Some studies reported a non-linear association between parity and CVD mortality10,19. For example, Dior et al.10 reported that higher mortality rates were observed for mothers of 1 child (HR = 1.18; 95% CI, 1.04–1.4), mothers of 5–9 children (HR = 1.21; 95% CI, 1.09–1.33) and mothers of ≥10 children (HR = 1.49; 95% CI,1.12–1.99) than mothers of 2–4, which was similar to the findings of present study. Additionally, Gallagher et al.14 showed slightly increased risk of CVD mortality associated with more than five births while Jaffe confirmed that the risk estimates of CVD mortality were higher among women with no children (HR 2.43, CI 1.49, 3.96) and women with more than 8 children (HR 1.64, CI 1.02, 2.65) than those with two children19,20. In contrast, a cohort study reported by Jacobs et al. suggested that women with more than 4 pregnancies were at lower CVD mortality risk11, with further reduction of mortality as parity increases12. Conversely, several studies denied the direct association between parity and CVD mortality13,15. These conflicting results from previous investigations make the present study more meaningful. Possible reasons for this discrepancy include a variation of cohort size, different races, subgroup analysis and so on. The results of the present study suggest that women with 4 or 5 children have the lowest risk of CVD induced mortality.

When we carried out the analysis of parity and CVD mortality by geographic location, significant positive association were observed in Israelite10. Considering limited studies from other European countries(Table 1), the interpretation of the results should be taken restrainedly. Although more American studies focused on this topic11,15,17,18,19, the results were still confusing. Nevertheless, we found few studies from Asia and only two of them focused on this topic from China14 and Korea13, which both showed borderline relationship between parity and CVD mortality without statistical significance. However, comparing with the reports from northern hemisphere, Simons and colleagues12, from Australia, reported that increased parity was associated with a tendency for decreased risk of CVD mortality, but without notably statistical significance either.

Our meta-analysis has several strengths. The included studies showed conflicting results, which may result from their limited statistical range and power. Our study, however, was conducted based on approximately 994,810 study participants and 16,601 CVD cases from 10 cohort studies. This massive database generated a stronger statistical power for us to detect and verify this putative association. Moreover, prospective design also helped to make the present study more robust by eliminating selection bias and recall bias. At the same time, our approach with meta-analysis toward the studies was one of the powerful tools to assess the role of parity in the risk of CVD mortality. To guarantee the analysis quality, this meta-analysis had a big sample size and the follow-up duration was considerably long. Although outcome evidence from long-term randomized trials is ideal, these studies are too difficult to implement on a practical basis, especially regarding reproductive factors.

Meanwhile, several limitations should also be acknowledged. First, the discovered information toward the relationship between parity and mortality from subtypes of CVD and each diagnosis criteria was limited, which remained this topic open to further research in the field of observational cohort studies. Second, parity is associated with several other factors, such as gestational hypertension, gestational diabetes mellitus, gynecological tumor and so on, which are established risk factors for CVD. Third, higher deliveries was generally associated with unhealthy factors such as a higher possibility of abortion, oral contraceptive and so on. The observed association between moderate parity and a lower risk of CVD is unlikely to consider these confounders. Fourth, delivery mode was not taken into account in our study., natural labor and caesarean section might lead to different prognoses. Last, the quality of individual original studies varied. Quality scoring might ignore some important information or introduce somewhat arbitrary subjective factors into the analysis because of combining disparate study features into a single score. Although we used the Newcastle-Ottawa Scale (NOS)23 to assess the quality of included studies, we did not score these studies or describe them as high or low quality quantitatively. Studies included in this review were all prospective studies and met most validity criteria, but three of them13,14,15 utilized special populations, four studies14,15,18,19 adjusted for limited confounders in the primary analyses, two studies17,18 didn’t follow up for at least ten years and the follow-up rates of three studies12,13,19 were less than 75%. These flaws might bring some bias, which should be paid more attention by investigators in the future. We have observed a significant heterogeneity across studies in our analysis pool, showing notable clinical relevance. We believe this may result from our large sample size, which can confer greater statistical power to heterogeneity tests.

In conclusion, the present meta-analysis suggests a potential non-linear J-shaped relationship between the number of parity and CVD mortality. Since the number of included studies was limited, further studies are warranted to confirm our findings as well as to stratify the results by the types of CVD and other risk factors to eliminate the residual confounding.

Methods

Search Strategy

We applied the Meta-analysis Of Observational Studies in Epidemiology (MOOSE) as the guidance to conduct and report this study24. A holistic review of the published articles (through the end of March 2015) with limitation to humans was performed by using MEDLINE (PubMed), EMBASE and ISI Web of Science databases. We used the following search strategy and keywords: (reproduction or reproductive factors or livebirth or pregnancy or parity) and (cardiovascular diseases or coronary heart disease) and (cohort study or prospective study). Furthermore, the listed article references were also examined and analyzed for additional studies.

Study Selection

The literature was selected by 4 criteria: (1) the study had a prospective design; (2) the exposure was either parity or the number of livebirth; (3) the outcome was mortality from CVD, coronary heart diseases (CHD), or ischemic heart disease (IHD); and (4) risk estimates with 95% confidence intervals (CIs). If multiple literatures were based on the same population, we chose the literature with larger sample size. We identified 10 potentially relevant prospective studies10,11,12,13,14,15,16,17,18,19 from 7973 articles (Fig. 4). Four studies10,12,13,16 only reported the results of highest compared with lowest number of livebirth, which were only brought into the analysis of parity number.

Figure 4
figure 4

Selection of studies for inclusion in meta-analysis.

Full size image

Data Extraction

Two independent investigators (Haichen Lv and Hongyi Wu) evaluated the eligibility and abstracted the data of each study. Discrepancies were settled by discussion. We summarized the following data from the selected studies: name of first author, year of publication, study’s country and design, sample size, follow-up year, exposure and outcome method, adjusted risk estimates and their 95%CIs of each included study for ever parity compared with nulliparous and potential confounders adjusted for in the primary analysis. We also calculated the risk estimate from the raw data demonstrated in the literature19 when it was not presented in the study.

Quality Assessment

The Newcastle-Ottawa Scale (NOS)23 included 3 quality parameters for cohort studies: the selection of study groups, comparability of groups and ascertainment of either the exposure or outcome of interest was used by two independent researchers (Haichen Lv and Hongyi Wu) to assess study quality.

Statistical Analysis

Considering heterogeneity between the studies, we used fixed-effects25 or the random-effects model26 to calculate summary risk estimate and 95% CIs for the ever parity compared with nulliparous and for the dose-response analysis. The method proposed by Hamling et al.27 was used to recalculate RRs for studies10,11,14,15,16,17 that did not use the category with the lowest number of parity as the reference or those that reported the risk estimates of each exposed category instead of combined estimates.

The methods described by Greenland et al.28 and Orsini et al.29 were used for the dose-response analysis, which require the distribution of cases and person-years or non-cases and the RRs with the variance estimates for at least three quantitative exposure categories30. We assigned the median or mean level of parity number in each category to the corresponding RR for each study demonstrated in the literature. We estimated the mean duration in each category by calculating the average of the lower and upper bound for studies that reported parity number by ranges of duration. We made the assumption that the open-ended interval length was the same as the adjacent interval when the highest category was open-ended. We set the lower bound to zero. We examined a potential nonlinear dose-response relationship between parity number and CVD risk using fractional polynomial models31, when the lowest category did not have a lower bound. We determined the best-fitting second-order fractional polynomial regression model, defined as the one with the lowest deviance. We used a likelihood ratio test to assess the difference between the nonlinear and linear models to test for nonlinearity32. The dose-response results are presented at one live birth increments. Finally, we carried out a sensitivity analysis by excluding each study in turn to evaluate the impact of individual data set on the overall estimate.

In the studies, heterogeneity was assessed by using the I2 statistics25. Subgroup analyses were carried out based on the CVD case number (≤500 versus >500), geographic location (North America, Europe and Asia) and whether adjustment for potential confounders. Small study bias, such as publication bias, was evaluated via Egger’s test33, Begg’s test34 and funnel plots. All statistical analyses were conducted via Stata software (version 11.2; StataCorp). P values were two sided with a significance level of 0.05.

Additional Information

How to cite this article: Lv, H. et al. Parity and Cardiovascular Disease Mortality: a Dose-Response Meta-Analysis of Cohort Studies. Sci. Rep. 5, 13411; doi: 10.1038/srep13411 (2015).