Parity and endometrial cancer risk: a meta-analysis of epidemiological studies

The association between parity and endometrial cancer risk is inconsistent from observational studies. We aimed to quantitatively assess the relationship by summarizing all relevant epidemiological studies. PubMed (MEDLINE), Embase and Scopus were searched up to February 2015 for eligible case–control studies and prospective studies. Random-effects model was used to pool risk estimations. Ten prospective studies, 35 case-control studies and 1 pooled analysis of 10 cohort and 14 case-control studies including 69681 patients were identified. Pooled analysis revealed that there was a significant inverse association between parity and risk of endometrial cancer (relative risk (RR) for parous versus nulliparous: 0.69, 95% confidence interval (CI) 0.65–0.74; I2 = 76.9%). By evaluating the number of parity, we identified that parity number of 1, 2 or 3 versus nulliparous demonstrated significant negative association (RR = 0.73, 95% CI 0.64–0.84, I2 = 88.3%; RR = 0.62, 95% CI 0.53–0.74, I2 = 92.1%; and RR = 0.68, 95% CI 0.65–0.70, I2 = 20.0% respectively). The dose-response analysis suggested a nonlinear relationship between the number of parity and endometrial cancer risk. The RR decreased when the number of parity increased. This meta-analysis suggests that parity may be associated with a decreased risk of endometrial cancer. Further studies are warranted to replicate our findings.

Different number of parity. The associations between different number of parity (1, 2 or 3) and endometrial cancer risk were evaluated respectively. Parity number of 1 versus nulliparous was inversely associated with risk of endometrial cancer (RR = 0.73, 95% CI 0.64-0.84; I 2 = 88.3%), after summarizing estimates from 19 studies ( Table 5). The significant inverse association was detected in almost all strata of subgroup analyses (Table 5). According to the Galbraith plot (Supplementary Figure 2), 6 studies contributed to the heterogeneity 10,13,[26][27][28][29] . The heterogeneity disappeared after excluding these studies in the pooled analysis (I 2 = 0.0%). Similarly, after summarizing 13 studies, parity number of 2 versus nulliparous demonstrated a significant inverse association with risk of endometrial cancer (RR = 0.62, 95% CI 0.53-0.74; I 2 = 92.1%), which was also identified in different strata of subgroup analyses (Table 6). Five studies contributed to the heterogeneity according to the Galbraith plot (Supplementary Figure  3   Additionally, parity number of 3 versus nulliparous showed a significant inverse association with endometrial cancer risk (RR = 0.68, 95% CI 0.65-0.70; I 2 = 20.0%), after pooling 7 studies.
Dose-response analysis. Assuming a linear relationship, we detected that the combined RR per an additional live birth was 0.86 (95% CI 0.84-0.89), with considerable heterogeneity (P for heterogeneity < 0.0001). After testing a potential non-linear relationship, the test for nonlinearity suggested that a non-linear relationship might exist (p for nonlinearity: 0.0058). Under this model the RR also decreased when the number of parity increased. The nonlinear relationship between the number of parity and endometrial cancer risk in females was demonstrated in Fig. 3.

Discussion
We performed a comprehensive quantitative meta-analysis to evaluate the relationship between parity and endometrial cancer risk. After summarizing all available evidence, ever giving birth to children was associated with an inverse risk of developing endometrial cancer. The sensitivity analysis demonstrated that the result was not significantly affected by any individual study; also subgroup analyses revealed that the inverse association was detected in all strata. Additionally, analyses assessing each number of parity (1, 2 and 3) demonstrated that the inverse association persisted for all 3 scenarios. Furthermore, we identified a dose-response relationship between the number of parity and risk of endometrial cancer. Overall, our findings support that parity may be associated with risk of endometrial cancer. Our findings are plausible based on understandings from basic research. Estrogens are known to stimulate proliferation of cells in the endometrium and increase mitotic activity, which can induce cancer development 30,31 . On the other hand, progestins can decrease risk of developing endometrial cancer through reducing cell proliferation and stimulating differentiation 31 . During live birth, there is a hormonal balance shift toward less estrogen and more progesterone, which may further affect risk of developing endometrial cancer 32 . Our finding of the dose-response relationship between the number of parity and endometrial cancer risk may be attributable to repeatedly long-term progesterone actions for the antiestrogenic endometrial effects 33,34 . Another potential explanation is that at each birth delivery there is mechanical shedding of malignant/premalignant endometrial cells 28,35 .
Our study has several strengths. To the best of our knowledge, this is the first comprehensive meta-analysis evaluating the association between parity and endometrial cancer. Besides conducting subgroup analyses and sensitivity analysis to further evaluate the association, we assessed associations of different numbers of parity and conducted dose-response analysis to fully understand the relationship. Our analyses suggested that the finding of the inverse association between parity and endometrial cancer risk might be robust. Several potential limitations need be acknowledged for the appropriate interpretation of our findings. First, we do not have access to the individualized primary data from each of the included studies, which induces the possibility that the risk estimates used in our pooled analysis may not be fully adjusted for. For example, obesity and use of oral contraceptive are among the known factors affecting risk of  developing endometrial cancer 32,36 . However, in some of the included studies, they were not adjusted for the association between parity and endometrial cancer risk. Residual confounding may thus be an issue. Second, for the dose-response analysis, the highest categories of number of live birth have wide range of values in different studies. The exposure values may not be accurately assigned based on our assumptions in the methods section. However, this limitation is difficult to eliminate and the method we used is in concordance with the general approach in this area. Third, our study mainly summarizes evidence from observational studies, which are known to confer several relevant biases due to the observational nature. Further large scale multi-center prospective studies are warranted to replicate our findings. Forth, we notice considerable heterogeneities across studies in our pooled analyses. We conducted numerous subgroup analyses with the hope of detecting potential factors for such heterogeneities; however, it appears that in many subgroups the heterogeneity remains relatively high. According to the Galbraith plots, a proportion of the included studies contribute to the high heterogeneities. The heterogeneities disappear after excluding these studies in the pooled analyses. These need to be considered when interpreting our findings. Last, we would like to acknowledge that I 2 value should be interpreted with caution because it has certain uncertainty. The value has relatively low statistical power especially in scenarios of small numbers of available studies 37 . However, in the current meta-analysis there are a relatively large number of eligible studies. Thus the possibility of this limitation is low.
In conclusion, based on a summarization of all available evidence from epidemiological studies, parous versus nulliparous was inversely associated with risk of endometrial cancer. There was a nonlinear dose-response relationship between the number of live births and risk of endometrial cancer. Our findings suggested that parity might be a risk factor for endometrial cancer, suggesting roles of reproductive factors in the etiology of endometrial cancer.

Data Sources and Search Strategies. A literature search of PubMed (MEDLINE), Embase and
Scopus databases was conducted from the inception to February 2015. We used the following search keywords: (((((((((parity) OR pregnancy) OR livebirth) OR reproductive) OR reproduction) OR reproductive factors) OR reproductive factor)) AND ((endometrium) OR endometrial)) AND ((((((((((malignancies) OR malignancy) OR neoplasm) OR neoplasms) OR cancer) OR cancers) OR adenoma) OR adenomas) OR carcinoma) OR carcinomas). We also screened references of included articles and relevant review papers to identify other potential studies. Study Selection. Studies were eligible if they (i) were prospective studies or case-control studies or pooled analysis of epidemiological studies; (ii) evaluated the association between parity and risk of endometrial cancer; (iii) presented RR, odds ratio (OR), or hazard ratio (HR) values with 95% CI or necessary data for determination. Cross-sectional studies were excluded. Epidemiological studies comparing endometrial cancer cases with controls with gynecology conditions were excluded as well. If we identified multiple articles involving same participants, the study with the largest number of patients and most relevant information was included.  Data Extraction and Quality Assessment. Two investigators independently carried out the abstract screening, full text screening, and data extraction. Disagreements were resolved by discussion, with input from other investigators. Data extracted from each study included: the first author's name, publication year, study country, study design, characteristics of study population (sample size, age, length of follow-up, measures and numbers of parity, and association effect sizes). If more than 1 estimate were reported, we used the estimate that was adjusted for the most appropriate covariates, like the previous studies [38][39][40][41][42] . In situations where only unadjusted estimates were provided, we used the crude estimate in the analysis. The qualities of included studies were assessed with the Newcastle-Ottawa Quality Assessment Scale 43 . Specifically, aspects of population and sample methods, exposure and outcome descriptions, and statistical matching/adjustments of the data were assessed. With this scale each study was assigned a score  Table 4. Summary risk estimates of the association between parity and endometrial cancer risk (parous versus nulliparous).
(maximum score is 9 points). Studies with an overall score of higher than or equal to 7 points were categorized as high-quality studies; others were categorized as low-quality studies.  Statistical Methods. The RR and 95% CI from included studies were used as the measure of association. Due to the rarity of endometrial cancer, ORs and HRs were deemed equivalent to RRs and RRs were used to represent measures. I 2 was used to assess the heterogeneity across studies, where a I 2 > 50% suggests considerable heterogeneity 44 . We pooled the log transformed RR using the fixed-effects model 45 when there was no considerable heterogeneity. We used the random-effects model 46 Table 5. Summary risk estimates of the association between parity and endometrial cancer risk (parity number of 1 versus nulliparous).
With regards to the dose-response analysis, we explored potential linear relationship between the number of parity and risk of endometrial cancer 47 . If studies reported the parity number by ranges, we used the midpoint of each category in the analysis. For studies in which the highest category did not have an upper end, the width of the highest category was assumed to be the same as the adjacent category, like previous studies 48,49 . Furthermore, we assessed potential non-linear relationship for the association. For this analysis, fractional polynomial models with restricted cubic splines and 3 knots at fixed percentiles (10%, 50%, and 90%) of the distribution were used 50,51 . We then performed a likelihood ratio test to determine whether nonlinear or linear relationship was suggested.
Publication bias was evaluated via Begg's test 52 . A P-value of 0.05 was used as the threshold to determine significant publication bias. All statistical analyses were performed with Stata (version 13; StataCorp, College Station, TX).  Table 6. Summary risk estimates of the association between parity and endometrial cancer risk (parity number of 2 versus nulliparous).