Reduced risk of prostate cancer in childless men as compared to fathers: a systematic review and meta-analysis

The previously reported association between fatherhood status and prostate cancer risk was controversial. We carried out the present meta-analysis of all relevant studies to summarize evidence on this association. A comprehensive literature search of studies was performed in PubMed, Web of Science, and the Chinese National Knowledge Infrastructure (CNKI) databases, covering all the papers published from their inception to September 2015. The combined risk estimates with 95% confidence intervals (CIs) were calculated using a random effects model. Heterogeneity and publication bias were also evaluated. A total of 11 studies were finally included in this meta-analysis. We found a significantly reduced risk of prostate cancer associated with being childless (OR 0.91, 95% CI 0.87–0.96). There was statistically significant heterogeneity across the studies (P < 0.001, I2 = 88.2%). In summary, this meta-analysis supports that being fatherless is associated with a lower risk of prostate cancer. Because of the substantial heterogeneity and residual confounding, using other study designs to further explore this association and the underling mechanism is warranted.


Results
Literature search and study characteristics. The detailed processes of literature search are presented in Fig. 1. Eleven studies 9-11,14-21 (five case-control studies, five cohort/nested case-control studies, and one pooled analysis) were finally included in this meta-analysis of the association between fatherhood status and prostate cancer risk. All of these studies were published between 2001 and 2013. Types of exposure included being childless (n = 8) and infertility (n = 3). Except the pooled analysis study, the characteristics of ten individual studies were summarized as follows: study regions included North America (n = 4), Europe (n = 5), and New Zealand (n = 1); the number of cases ranged from 168 to 117,328, with a total of 182,012; study quality scores ranged from Evaluation of heterogeneity. In this study, the Q test and the I 2 index were adopted to examine the heterogeneity among included studies. As shown in Fig. 2, statistically significant heterogeneity was observed across the studies (P < 0.001 for heterogeneity, I 2 = 88.2%). Then the Galbraith plot was used to detect the potential sources of heterogeneity. As shown in Fig. 3A, the studies performed by Rosenblatt et al. 17 , Dennis et al. 18 , and Giwercman et al. 9 might contribute to the heterogeneity. The omission of these publications markedly reduced the heterogeneity (P = 0.337 for heterogeneity, I 2 = 12.0%) while the overall association remained significant (OR 0.91, 95% CI 0.88-0.94) (Fig. 3B).

Sensitivity analysis and cumulative meta-analysis.
In the sensitivity analysis, the influence of each study on the summary effect estimate was evaluated by repeating the meta-analysis after removing one study at a time. As shown in Fig. 4, the pooled ORs were not considerably affected by omitting any individual study, which indicated that our results were robust. Cumulative meta-analysis was performed by sorting the included studies based on publication date. Supplementary Figure S1 shows the results from the cumulative meta-analysis of the association between fatherhood status and prostate cancer risk in chronologic order.

Publication bias.
There was no evidence of significant publication bias with Begg's test (Fig. 5A, P = 0.533).

Discussion
This study summarizes and quantifies the current evidence on the association between fatherhood status and prostate cancer risk in a meta-analysis of observational studies, including five case-control studies, five cohort/nested case-control studies, and one pooled analysis study. To the best of our knowledge, this is the first meta-analysis evaluating the association between fatherhood status and prostate cancer risk. The results indicated that being fatherless was inversely associated with the incidence of prostate cancer (OR 0.91, 95% CI 0.87-0.96).  Androgen status may mediate the link between fatherhood status and prostate cancer risk. Infertile men have lower levels of testosterone and serum testosterone to estradiol ratios compared with fertile men 22 . Low circulating levels of testosterone have been suggested to be associated with a reduced risk of prostate cancer 23 , although the evidence from observational studies remains controversial 24 . In addition, inhibition of dihydrotestosterone (DHT) serum levels by use of 5α -reductase inhibitors markedly reduced the risk of prostate cancer in two large randomized clinical trials 25,26 .
It is worth noting that fatherhood status as a proxy for male fertility are hampered by the fact that the fatherhood status is influenced by various factors comprising fertility of the man, fertility of the partner, opportunity to start a family, and desire to have children. In subgroup analysis by marital status, the association was still statistically significant in the studies adjusted for marital status (OR 0.90, 95% CI 0.85-0.97). Thus, the presence of confounding from unmarried men could be partially ruled out. However, we were not able to assess the differential misclassification from men's reproductive intent and fertility of their partners.
Another important confounding factor was the difference in healthcare-seeking patterns, as married and better educated men may have a higher uptake of PSA testing 27 . In this study, we conducted stratified analysis by education and the pooled estimate remained consistently significant in studies adjusted for education. Nevertheless, as the majority of the included studies did not provide the information of PSA testing, we could not evaluate the role of PSA testing in the association between fatherhood status and incidence of prostate cancer. Therefore, there are reasons to believe that the beneficial effects of being fatherless were at least in part due to unmeasured and residual confounding.
Our study had some strengths. A total of 182,012 prostate cancer cases were included in this meta-analysis, which enhanced the statistical power and provided more reliable estimates. The estimates from fully adjusted models in each study were used in this study to minimize potential confounding. Various subgroup analyses and sensitivity analyses were carried out to assess the robustness of the results.
Our study also had several important methodological limitations. First, substantial heterogeneity was observed across individual studies (P < 0.001 for heterogeneity, I 2 = 88.2%), which might distort the pooled estimates. Through the Galbraith plot, we detected the studies that potentially contributed to the heterogeneity. After removing these studies, the combined estimate remained significant (OR 0.91, 95% CI 0.88-0.94) without obvious heterogeneity (P = 0.337 for heterogeneity, I 2 = 12.0%), which indicated that the heterogeneity didn't have a material impact on our conclusion. Second, publication bias was detected by Egger's test (P = 0.003), although the trim-and-fill analysis did not alter the findings substantially. Small negative studies were less likely to be published and we were not able to include some gray literature, such as conference abstracts and studies reported in languages other than English or Chinese. Third, half of the included studies adopted case-control design, which may introduce the possibility of select bias and recall bias. However, the observed association was robust and consistent in the subgroup of cohort/nested case-control studies (OR 0.90, 95% CI 0.85-0.95). Fourth, there were vast differences in size of included studies, ranging from 168 to 117,328. The pooled estimate was vulnerable to the results of large studies, such as Giwercman et al.'s and Wiren et al.'s studies 9,21 . In addition, these two largest studies 9,21 were both nationwide in Sweden and had an overlap of cases (between 1991 and 1998), which may lead to biased estimates.
Overall, this meta-analysis supports that being fatherless is associated with a low incidence of prostate cancer. Because of the substantial heterogeneity and residual confounding, using other study designs to further explore this association and the underling mechanism is warranted. . We adopted the following search algorithm: (fertility OR infertility OR infertile OR fatherhood OR father OR childless OR child OR children OR offspring OR marriage OR partner OR sex OR sexual OR sexually OR intercourse OR coitus) AND (prostate OR prostatic) AND (cancer OR neoplasm OR tumor OR malignancy) AND (cohort OR case-control OR case control). No language limitation was applied. We evaluated all retrieved publications carefully by examining their titles and abstracts, and the full texts of studies potentially matching the eligible criteria were further checked. We also reviewed reference lists of articles and reviews to identify any additional relevant studies. The present systematic review was planned, conducted, and reported in adherence to standards of quality for reporting meta-analyses 28 .

Study selection.
The studies included in this meta-analysis met all of the following criteria: (i) the exposure of interest was being childless or infertility. Infertility is defined as the absence of a live birth for men who desire a child. Considering infertile men generally have no children, the studies with exposure of male infertility were also included in the pooled analysis; (ii) the outcome of interest was prostate cancer; (iii) study design was case-control or cohort; and (iv) the effect size estimates with their corresponding 95% confidence intervals (CIs) were provided (or data were sufficient to calculate these values). Whenever multiple articles from the same study population   were available, the study with the largest number of cases and most applicable information was included in this meta-analysis. In particular, we included a pooled analysis of 18 studies, rather than separate articles, since part of these articles are non-English literature or lack sufficient data to calculate the effect size estimate.
Quality assessment. The quality of each study was evaluated by two authors (Y.M. and X.X.) using the Newcastle-Ottawa Scale (NOS) (http://www.ohri.ca/programs/clinical_epidemiology/oxford.asp). This scale has three categories: Selection (four items), Comparability (one item), and Exposure for case-control studies or Outcome for cohort studies (three items). A study can be awarded a maximum of one star for each numbered item within the Selection and Exposure/Outcome categories. A maximum of two stars can be given for Comparability. The stars are then added up to obtain the total quality score.
Data extraction. The following information were extracted from each study: first author's name, year of publication, study design, the country in which the study was carried out, sample size, age of study population, exposure assessment, duration of follow-up for cohort/nested case-control studies, adjusted risk estimates with their 95% CIs, and matched or adjusted factors in the design or data analysis. Data extraction was performed independently by two authors (Y.M. and X.X.) using a predesigned data collection form. Discrepancies were resolved by group consensus and consulting a third reviewer.
Statistical methods. Considering that prostate cancer is a rare disease, the relative risk (RR) was assumed to be nearly equal to the odds ratio (OR), and the OR was used as the study outcome. The combined OR with its 95% CI was calculated to assess the strength of the association between fatherhood status and prostate cancer risk 29 . For studies which reported separate risk estimates for different number of children (e.g., one child, two children, three children, and so on), we pooled these risk estimates within each study, weighted by inverse of the variance 30 . Stratified analyses were carried out by study design, geographical region, number of cases, type of exposure, marital status, and education. The heterogeneity among included studies was tested by the Q statistic and the I 2 score 31 . Statistical significance of heterogeneity was set at 0.1. The value of I 2 was used to describe the degree of heterogeneity (I 2 < 25% no heterogeneity; I 2 = 25-50% moderate heterogeneity; I 2 > 50% large or extreme heterogeneity). Galbraith plot 32 was used to detect the possible sources of heterogeneity. A re-analysis was carried out after removing the studies possibly leading to the heterogeneity.  Sensitivity analysis was conducted by repeating the meta-analysis after the omission of every study in turn to assess the effect of each study on the combined estimate. Cumulative meta-analysis was also performed by sorting the studies based on publication year.
Potential small-study bias was evaluated by Begg's test (rank correlation method) 33 , Egger's test (linear regression method) 34 , and trim-and-fill method 35 . If P < 0.05, results were considered statistically significant. All of the statistical analyses were conducted with STATA 11.0 (StataCorp, College Station, TX), using two-sided P values.