Rapid advances in our understanding of the hereditary basis of disease over the past decade have spurred an interest in the identification of inherited genetic risk for many cancers, including prostate cancer. Although a hereditary component to the risk of prostate cancer has been recognized since 19601 and confirmed first through the now classic Utah Mormon genealogy studies2, 3 and more recently through a series of case-control and cohort studies, a predisposing gene(s) that is responsible for the majority of hereditary prostate cancer has yet to be discovered. However, linkage analysis studies have mapped several susceptibility loci at 1q24-25 (HPC1),4 1q42.2-43 (PCAP),5 Xq27-q28 (HPCX),6 1p36 (CAPB),7 20q13 (HPC20),8 8p22-239 and 17p11 (HPC2/ELAC2).10 Three genes have been cloned within these loci including the first to be cloned, HPC2/ELAC2 at 17p11, RNASEL (encoding ribonuclease L) at 1q24-25, with the Arg462G variant implicated in up to 13% of prostate cancer cases,11 and the 471delAAAG variant associated with prostate cancer in Ashkenazi Jewish men,12 and recently the macrophage scavenger receptor 1 (MSR1) at 8p22-23, which may promote prostate cancer through increased oxidative stress.13 Other susceptibility genes with an associated increased risk for prostate cancer include BRCA1, with carriers having a 3.3-fold increased risk of prostate cancer compared to the general population,14 BRCA2 associated with a 2–5% increased risk of disease15 and CHEK2, which is involved in the DNA-damage-signaling pathway and has been found in 4.8% of 578 patients with familial prostate cancer (defined as disease in a minimum of 3 men over at least 2 generations).16 Although it is currently unknown what fraction of population prostate cancer risk the above genes attribute to individually or collectively, it appears to be small, possibly less than 5–10%.
While the relentless search for genes moves forward, scientists and clinicians alike are interested in quantifying what we know about the risk related to familial prostate cancer. Although multiple publications have now documented varying levels of risk associated with familial prostate cancer, a meta-analysis has yet to be conducted to better quantify these risks. Better quantification of risk associated with degree of relationship between a proband and affected family member(s) would serve to inform counseling of men at increased risk of prostate cancer. This study was performed: (i) to identify published studies that quantify the relative risk (RR) of prostate cancer associated with a family history of the disease, through a structured review of literature; (ii) to summarize the evidence and compare RR associated with having a father, brother, first- or second-degree relative or other relative affected with prostate cancer.
MATERIAL AND METHODS
Published studies were identified using the Medline database (National Library of Medicine, Washington, DC) for the time period January 1, 1982 to November 8, 2000. Studies were also identified using published studies from the reference lists of key articles17, 18 for the same time period and from cyber updates from medical literature list servers. The search categories were then compared for duplicate articles. Inclusion and exclusion criteria are described in Tables I and II. Further, as quality control, suggested guidelines for meta-analysis reporting were drawn from several sources.19, 20, 21, 22 Studies were coded by one of the co-authors using predefined fields developed by the research team. Intercoder reliability was determined through independent coding of all 24 of the included studies and 10 of the excluded studies by a second co-author, with 95% agreement. All discrepancies involved the definition of cases and were resolved in favor of the original coding with discussion of the gray areas in the coding categories. Validity was established by combining only studies that met case control or cohort criteria with reported relative risk and 95% CI for family associations and prostate cancer risk.
Table I. Systematic Medline Review of Prostate Cancer and Family History: Inclusion Criteria and Results
Inclusion search criteria
No. of studies meeting inclusion criteria
1) “family history” + “prostate cancer”
a) Criteria in #1 + “case-control”
b) Criteria in #1 + “cohort”
c) Criteria in #1 + “relative risk”
2) Studies identified through reference list in Narod (1998)17
3) Other studies found in reference list of publications identified through search criteria
Duplicate articles identified in searches 1 through 1C
Articles excluded based on exclusion criteria as defined in Table II
Table II. Systematic Medline Review of Prostate Cancer and Family History: Exclusion Criteria
No. of studies meeting exclusion criteria
No confidence intervals reported
No relative risks reported
Other (editorials, comments or nonresearch related)
Other primary cancer site
Symptoms/treatment of cancer
Inadequate control for age
The initial search identified 332 articles that met the inclusion criteria listed in Table I. The final analysis included 23 studies after excluding those that met the criteria listed in Table II. Fourteen case-control and 9 cohort studies were identified and are summarized in Tables III and IV, respectively. Of the 14 case-control studies, 9 were population based and 6 were hospital based.
Participants in most studies were predominantly Caucasian men. Only 3 studies included African-American men,23, 24, 25 whereas 1 study26 investigated a population of Jamaican men, and 1 study evaluated the association between family history and risk of prostate cancer in Asian men.25
In terms of quality of the studies, even within study categories (case control [population or hospital based], cohort) there was a wide range of ascertainment methods, ranging from high-level, well-described, well-maintained regional registries to lower-level methods, such as single department and less well-defined populations of convenience. In addition, one study, Ghadirian et al.,27 tried to match by age but was unable to adequately control for age differences between cases and controls and was consequently excluded from all of the analyses.
Several studies ascertained cases through the same population. Bratt et al. 199728 and Bratt et al. 199929 both used patients from the Swedish Regional Tumor Registry, although Bratt et al. 199728 is a cohort study, whereas Bratt et al. 199929 is a case-control study. Gronberg18, 30 used sons of men who were diagnosed from 1959 to 1963 and were ascertained through the same Swedish Cancer Registry used by Bratt et al. However, they sampled a much larger population. Issacs et al.31 used the same hospital-based cases ascertained from the same men undergoing radical prostatectomy over the same time period as Steinberg et al.,32 although Issacs included family members of probands and a small national sample of families with multiple affected members.
Because of these duplications in study populations, we conducted 3 meta-analyses. The first analysis was all inclusive of our search criteria and yielded what we term a “summary estimate”. The other 2 analyses were conducted to investigate the sensitivity of our results to population duplication. To exclude potential bias from data repetition, both of these analyses excluded studies with population overlap. In what we consider the strongest of the 3 estimates, or the “primary” analysis, we included only the most recent reports from studies of overlapping populations. When there was overlap in dates, the primary analysis used cohort studies over case-control studies. Cohort studies were preferred because they are less subject to recall bias. For example, the cohort study of Bratt et al. 199728 ascertained family history through use of Parish and Cancer Registries, whereas the case-control study of Bratt et al. 199929 relied only on subject interviews. Further, Kalish et al.33 reported that subjects in their cohort study were less likely to report a positive family history of prostate cancer after their own diagnosis than prior to diagnosis, suggesting that positive associations between prostate cancer risk and family history are not due to recall bias.34 However, the evidence against recall bias in a case-control study on a similar population35 was based more on a similarity to results from other studies less subject to recall bias than on evidence from their own data. In the third meta-analysis, studies excluded in the primary analysis were included, whereas the corresponding studies included in the primary analysis were excluded. Results from all 3 analyses were not significantly different, therefore we report only the primary and summary estimates.
All studies that met inclusion criteria are listed in Tables III and IV and in Figures 1 and 2. Studies that were excluded from the “primary” analysis are indicated by the symbol “X”. In addition, some of the results from the primary analysis only are displayed as forest plots in Figures 2 and 3. For each study in these figures, the case-control studies are presented first in alphabetical order and, after a blank space, the cohort studies are presented. The RR from each study is presented graphically as a solid square, and the 95% confidence interval is distance of the straight line drawn through the square. The area of the solid square is proportional to the size of the study and inversely proportional to the variance of the estimate. Thus, small confidence intervals correspond to large squares and vice versa. A solid diamond is drawn at the bottom of the plot. The diamond at the bottom of the plot is centered at the weighted average of the RRs and the width of the diamond indicates the 95% confidence interval for the weighted average. The R statistical package36 was used to compute the weighted RR estimates and to produce the forest plots.
We evaluated the significance of differences between the RR of prostate cancer associated with having a father vs. a brother with prostate cancer using a weighted paired t-test with weights proportional to the inverse of the variance of the difference. A 2-sided p-value was computed from a t-distribution and used a boot-strap to protect against possible nonnormality.37
As a quality-control assessment, we constructed a funnel plot (Fig. 3) to evaluate the potential effect of publication bias on results of our meta-analysis.19, 20, 21, 22 This plot suggests potential publication bias in the study by McCahy et al.,38 which may lead to a slight overestimate of RR in the meta-analysis.39 Further, since meta-analyses often exhibit heterogeneity beyond sampling variation,19, 20, 21, 22 we tested this assumption. A χ2 test of heterogeneity of risks among our studies did not detect any for first-degree relatives (p-value = 0.135) but did find evidence of heterogeneity for fathers (p-value = 0.002) and brothers (p-value = 0.011). A comparison of the log relative risk estimates for population-based case-control studies, hospital-based case-control studies and cohort studies showed no differences among the 3 groups (weighted analysis of variance, p-value = 0.35, 0.50 and 0.16 for first-degree relatives, fathers and brothers, respectively). Hence, the heterogeneity is not a result of combining these 3 groups.
History of prostate cancer in any relative
There were 7 studies that reported RR in any relative: 4 were case-control studies32, 35, 38, 40 and 3 were cohort studies.33, 41, 42 These findings all show a positive association between having any family history of prostate cancer and increased risk for disease in yet unaffected family members. Relative risks range from 1.7 to 8.22, with the pooled summary estimate of RR = 2.04 (95% CI 1.64–2.55). The results for the primary estimate are RR = 1.93, 95% CI 1.65–2.26.
History of prostate cancer in a first-degree relative
There were 16 studies that reported RR in a first-degree (father or brother) relative, 11 of these case-control studies23, 24, 25, 26, 29, 31, 32, 35, 38, 43, 44 and 5 of these cohort studies.28, 41, 45, 46, 47 RRs estimated from hospital-based case-control studies, population-based case-control studies and cohorts were compared using a weighted analysis of variance and did not differ significantly (p = 0.35). Figure 1 summarizes these findings as a forest plot, and all but one28 show a positive association between having a first-degree relative with prostate cancer and increased risk for disease. RRs range from 1.43–17.83 with a pooled RR = 2.24, 95% CI 2.08–2.41, as indicated by the summary meta-analysis. The results for the primary analysis are RR = 2.22, 95% CI 2.06–2.40.
History of prostate cancer in a second-degree relative
Six studies reported risk associated with a second-degree (grandfather or uncle) relative having prostate cancer. However, only 2 of these studies reported no significant difference without reporting RR,24, 26 and could not be used in the meta-analyses. Of the 4 studies reporting RR in a second-degree relative, 3 were case-control studies,32, 38, 48 and 1 was a cohort study.47 Two studies showed a significant positive association,32, 48 and 2 showed no significant association38, 47 between having a second-degree relative with prostate cancer and increased risk for disease. Relative risks range from 1.24 to 3.12, with a pooled summary estimate of RR = 1.91, 95% CI 1.58–2.30). The results for the primary analysis are RR = 1.88, 95% CI 1.54–2.30.
History of prostate cancer in a father
There were 14 studies that reported RR of having a father with prostate cancer, 8 of which were case-control studies,23, 24, 29, 32, 35, 43, 48, 49 and 6 were cohort studies.18, 30, 41, 42, 45, 47 Figure 2 summarizes these findings (plotted in gray); all but 241, 42 (both cohort studies) demonstrate a significantly positive association between having a father with prostate cancer and increased risk for disease. Relative risks range from 1.23 to 3.77, with a pooled summary estimate of RR = 2.14, CI 1.83–2.51). The results for the primary analysis are RR = 2.12, CI 1.82–2.51.
History of prostate cancer in a brother
There were 12 studies that reported RR of having a brother with prostate cancer, 8 of which were case-control studies,23, 24, 29, 32, 35, 43, 48, 49 and 4 were cohort studies.41, 42, 45, 47 Figure 2 also summarizes these findings (plotted in black), which show that all but one41 (a cohort study) demonstrate a significantly positive association between having a brother with prostate cancer and increased risk for disease. Relative risks range from 1.41 to 6.29, with a pooled summary estimate of RR = 2.87, CI 2.21–3.73). The results for the primary analysis are RR = 2.84 CI 2.16–3.72.
History of prostate cancer in a father vs. a brother
In Figure 3, the relative risks for having a father with prostate cancer are plotted in gray and the relative risks for having a brother with prostate cancer are plotted in black. Twelve studies reported the RR of having a father and brother with prostate cancer; 8 were case-control,23, 24, 29, 32, 35, 43, 48, 49 and 4 were cohort studies.41, 42, 45, 47 In 10 of the 12 studies that report RR for both fathers and brothers, the RR for a brother was higher than that for a father (Fig. 2). In combination, a t-test of the paired comparisons yields a 2-sided p-value of 0.04, indicating a statistically significant difference between the relative risks for fathers and brothers. A bootstrap test of paired comparisons among these 10 studies provides even greater statistical support for these findings (p-value = 0.005).
The funnel plot suggests publication bias probably did not have a major impact on results of our meta-analysis. Although the RR reported by McCahy et al.38 was considerably larger than RRs reported by other studies, because of the large variance in the estimate it does not greatly influence the pooled estimates of RR.
This meta-analysis supports an increased risk of prostate cancer for men with a positive family history. RRs for men with any affected family member, an affected first-degree relative and an affected second-degree relative were 1.93, 2.22 and 1.88, respectively. Furthermore, among first-degree relatives, risk was significantly higher for men with an affected brother compared to those with an affected father; pooled RRs were 2.87 and 2.12, respectively.
This meta-analysis supports smaller samples that have reported a higher RR of prostate cancer when a brother, rather than a father, is affected. Having a brother with prostate cancer is associated with a statistically significant increased risk of prostate cancer as compared to having an affected father or any other combination of affected relatives assessed in our study. Reasons an affected brother portending the highest RR among familial associations is unknown. The result suggests the possibility that earlier age of onset is associated with a stronger genetic risk for prostate cancer. There also is some evidence in the literature for an X-linked mode of inheritance of genetic susceptibility for prostate cancer,50, 51 although not all studies support this association,52 and none of the extensive work conducted to identify an X-linked gene has yielded positive results. Although men with an affected brother were at highest risk of prostate cancer in our analysis, men with affected fathers were also at substantially elevated risk, which argues against a role of X-linked transmission for most cases. Our results are consistent with what many scientists have begun to conclude, that a large proportion of familial prostate cancer may be due to shared environment in combination with familial sharing of low-penetrance alleles at many loci, each contributing to a small increase in cancer risk, rather than a few rare highly penetrant genes. Although a small number of individual genes may certainly contribute to a Mendelian inheritance model of prostate cancer, a complex mixed model that simultaneously takes into account multiple susceptibility genes, including genes of moderate or low penetrance, might more appropriately define the genetic transmission of prostate cancer.53
Interestingly, the pooled RR for fathers (primary estimate 2.12) did not differ substantially from that of second-degree relatives (primary estimate 1.88), presumably mostly uncles, suggesting that temporal and environmental influences may also be important in familial clustering of prostate cancer. Other possible explanations include improved diagnostics or simply increased awareness. Men currently have access to screening and diagnostic options with the prostate-specific antigen (PSA) that were not available to many of their fathers and uncles who would have been more likely to have died with asymptomatic, undetected disease. In addition, changes in environmental, occupational and dietary exposures over time could contribute to generational differences. Environmental differences may be particularly important in explaining differences in risk of prostate cancer associated with a positive family history depending on whether the affected relative was a brother or father if childhood and adolescence is a critical time for exposure as has been suggested,54 since brothers share a common environment during this time but fathers and sons do not.
Strengths of our study include use of clear, conservative search criteria and coding guidelines that resulted in a high interrater reliability when evaluating studies for inclusion. Further, the robustness of the analysis was demonstrated by the lack of sensitivity of results to exclusion of studies that used overlapping study populations. Average RRs differed by less than 2% for all categories, and corresponding confidence intervals were similar. Furthermore, there was little evidence that a single study was an influential outlier that had a disproportionate effect on pooled estimates of RR. Limitations of our analysis include inability to discriminate men who only had an affected brother, men who only had an affected father and men who had an affected brother and an affected father, which may have decreased observed differences in risk depending on whether the affected first-degree relative was the father or a brother. In addition, we did not have data on the number of brothers and other relatives with cancer relative to the number at risk of cancer, we were unable to estimate age-specific risks, and we could not account for known prostate cancer risk factors such as race and diet in our analysis.
Meta-analysis of the current literature on risk of prostate cancer among men with a positive family history indicates a RR of 1.8-, 2.1- and 2.9-fold increased risk, respectively, depending on whether the affected relative was a second-degree relative, the father or a brother. A structured literature review identified 12 studies that reported a RR for both fathers and brothers, 10 of which show a higher RR for having an affected brother than for having an affected father. However, none of these studies quantified the statistical significance between these relative risks. Our study is the first to report a statistically higher risk associated with having a brother with prostate cancer than having an affected father. Both genetic and environmental factors likely contribute to this increased risk and require further research.
The authors thank Dr. M. Daly for her thoughtful review and Ms. L. Bagden for her excellent secretarial assistance.