Although narrative reviews have concluded that there is strong support for familial clustering of prostate carcinoma, the association has never systematically been quantified in reviews. The purpose of this meta–analysis was to summarize and quantify the recurrence risk ratio with emphasis on the degree of relation, the specific relationship of the family member, the number of affected family members, and the age at diagnosis.
Publications were identified through computerized database searches for epidemiologic studies published up to December 2002. In addition, references cited in original and review papers were examined. Three blinded reviewers extracted both qualitative and quantitative information from each paper. Using random effects meta–regression analyses, the authors calculated summary recurrence risk ratios (Sλ). The reviewers also evaluated changes in Sλ according to differences in study methodology.
Thirty–three epidemiologic studies were included in the current review. Sλ was 2.53 (95% confidence interval, 2.24–2.85) for first–degree family members. Sλ appeared to be greater for men with an affected brother than for men with an affected father. Sλ for men who had second–degree relatives with prostate carcinoma was only slightly elevated. The nature of the familial clustering is such that Sλ increases with decreasing age of the patient and family members, with increasing genetic relatedness of the affected relative, and with increasing number of individuals affected within a family.
Prostate carcinoma is one of the most common cancers in the Western world. In the United States, this disease currently is the most commonly diagnosed noncutaneous malignancy among men. The incidence is increasing by 10–20% every 5 years, even when screen–detected carcinomas are disregarded.1 Although the incidence of latent prostate carcinoma appears to be constant throughout the world, the incidence of clinical forms varies substantially.2 African American men have long been known to have the highest rates of prostate carcinoma in the world, whereas native Japanese and Chinese men have the lowest known prostate carcinoma rates.3 This difference is likely due to both environmental and genetic influences.
Both common environmental risk factors and genetic influences can contribute to the familial clustering of prostate carcinoma. Next to age and race, family history appears to be one of the most important risk factors for prostate carcinoma. Attempts to elucidate the familial nature of prostate carcinoma began approximately 45 years ago after a study showed a higher incidence of prostate carcinoma in close relatives of patients with prostate carcinoma than in close relatives of control patients.4 Further epidemiologic studies confirmed this familial clustering of prostate carcinoma,5–37 although the reported magnitude of the estimated risk varied.
To our knowledge, no meta–analysis of all previous studies concerning familial clustering of prostate carcinoma has been conducted to date. Earlier narrative reviews of family history and prostate carcinoma have described the association for first–degree familial clustering of prostate carcinoma. However, these reviews did not calculate the corresponding recurrence risk ratio, i.e., the risk of prostate carcinoma for relatives of a patient with prostate carcinoma compared with the population risk.24, 38–45 According to these narrative reviews, men who have first–degree family members with prostate carcinoma have two to four times the risk of men who have no relatives with the disease. The magnitude of the other aspects of familial clustering (e.g., type of family member, number of affected family members, and age at diagnosis) also has not systematically been reviewed or quantified.
The objective of this study is to systematically review all epidemiologic studies up to December 2002; in order to obtain quantitative summary estimates of the familial clustering of prostate carcinoma with emphasis on the degree of relation, the specific relationship of the family member, the number of affected family members, and the age at diagnosis; and to evaluate changes in summary estimates due to differences in populations and study designs.
MATERIALS AND METHODS
Publications were identified by the principal investigator through computerized Medline, Embase, Cancerlit, and Current Contents searches for epidemiologic studies that had been published up to December 2002, with no language restriction. The keywords used were combinations of prostat*, cancer, carcinoma, famil*, father, and brother. In addition, references that were cited in original and review papers were examined until no further study was identified. To be included in this analysis, a paper had to describe a cohort study, a case–control study, or a cross–sectional study that compared prostate carcinoma incidence in relatives of patients with population incidence rates and provided sufficient information to estimate a recurrence risk ratio (denoted λ) as well as the associated standard error of incident primary prostate carcinoma comparing men with and without family members who had the disease. Studies of prostate carcinoma mortality were excluded because these studies, which frequently deal with aggressive rather than indolent disease, were expected to introduce too much heterogeneity into the meta–analysis. Also, death certificate diagnoses were considered problematic, as the age at onset of disease was relatively difficult to ascertain.
In compiling the database, a distinction was drawn between an article and a study. A study comprises all the analyses of a given group of research subjects. These analyses may be described in more than one article. When the same study population was referred to in more than one article, they were considered as part of a single study. When different results pertaining to the research subjects of a single study were published in more than one article, all such articles were included, but the data were combined to reflect the fact that only one group of patients was involved.
Qualitative Data Extraction
Three blinded reviewers extracted both qualitative and quantitative information from each paper. Original papers were blinded for authors, affiliations, journal name, publication year, and acknowledgments. We independently assessed the following qualitative items: general information (i.e., publication year and geographic area), population characteristics (i.e., population size and ethnic label), patient characteristics (i.e., number and mean age at diagnosis), and assessment method of family history. To investigate the potential influence of the timing of diagnosis, all studies were categorized as either “prospective studies,” in which family history of prostate carcinoma was assessed before diagnosis (cohort or cross–sectional studies); “retrospective studies;” in which family history of prostate carcinoma was assessed after diagnosis (cohort or case–control studies); or “registry linkage studies,” in which no patient reporting was utilized (cross–sectional studies). When continuous data were presented across different subgroups, we calculated a weighted mean, using the median of each subgroup weighted by the frequency of observations in that subgroup. When there was disagreement on an item, it was discussed until a consensus was reached.
Quantitative Data Extraction
We extracted quantitative data that allowed us to calculate λ values—estimated by the relative risk in cohort studies and the odds ratio in case–control and cross–sectional studies—and corresponding standard errors in order to estimate the association between prostate carcinoma risk and family history of prostate carcinoma for fathers (λf), brothers (λs), first–degree family members with prostate cancer (i.e., fathers, brothers, or sons) (λ1), 1 first–degree family member, more than 1 first–degree family member, first–degree family members age < 65 years, first–degree family members age > 65 years, first–degree relatives of patients who were diagnosed at age < 65 years, first–degree relatives of a patients who were diagnosed at age > 65 years, and second–degree family members (i.e., grandfather or uncle) (λ2). When the reported λ1 values within a study were stratified for age or other variables, we calculated an intrastudy pooled λ1 using fixed–effect pooling within age strata. When an age category included age 65 years, the corresponding λ1 was used in the calculation of both estimates (< and > age 65 years) at 50% of its original weight in each estimate. When the degree of family history was not reported, we assumed it to be first–degree.5, 17, 20, 25 Preferably adjusted λ values were extracted. When adjusted λ values could not be calculated, two–way contingency tables, based on exposure frequency distributions, were constructed to calculate the unadjusted λ values and corresponding standard errors. The standard error for the unadjusted λ values was calculated by the method of Woolf.46 Papers reporting ethnically stratified λ values were considered as separate studies so that the ethnic labels of study populations could be incorporated as covariates in meta–regression analysis in order to explore potential sources of heterogeneity.
To detect publication biases or other related biases, we explored heterogeneity in funnel plots—i.e., plots of effect estimates versus their estimated precision (the reciprocal of the variance). We visually examined funnel plot asymmetry and used Egger's unweighted regression asymmetry test47 to determine the degree of asymmetry. Standard meta–analytic procedures assume that results within a given analysis are independent. Our sorting of articles into their respective studies ensured that, for any given analysis of family history and prostate carcinoma, all of the summarized recurrence risk ratios would be based on independent samples. We used the Stata statistical software package48 (StataCorp, College Station, TX) to estimate the summary odds ratios and corresponding 95% confidence intervals (CIs) by random effects meta–regression analysis. The variance between studies was determined by a noniterative procedure that employed a method of moments estimator. To explore reasons for the observed heterogeneity, we performed sensitivity analyses on the study characteristics and tested their influence on the pooled effect estimates. The regression model relates the risk of prostate carcinoma to the study–level covariates, assuming a normal distribution for the residual errors with additive within–study and between–studies components.
The search strategy revealed 45 articles that reported epidemiologic studies of family history and prostate carcinoma.5–37, 49–60 Twelve of these articles were excluded—two because they examined prostate carcinoma mortality rather than prostate carcinoma incidence,49, 50 six because no sufficient data could be extracted,51–56 and four because different study designs were used.57–60 Eight articles were combined in the analysis because the same study was published twice (Table 1).9, 10, 12, 14, 16, 20, 29, 36 The 33 articles that were included in the review5–37 described 5 prospective studies,5, 7, 23, 24, 26 20 retrospective case–control studies,6, 8, 10, 13, 15–22, 25, 27–34, 36 and 3 registry linkage studies,11, 12, 14, 35, 37 and comprised a total of 200,215 patients. All studies were published in English except for one article published in Italian22 and one published in Spanish,25 both of which were translated into English for data extraction by native–speaking colleagues. All studies that were identified were performed in Western countries; 22 studies5, 7–11, 13, 15–21, 23, 24, 27–34, 36 were performed in the United States of America or Canada, and 6 studies6, 12, 14, 22, 25, 26, 35, 37 were performed in Europe. Most studies investigated the association between family history and prostate carcinoma risk within Caucasian populations,5–12, 14–18, 21, 24, 26–30, 32–35, 37 although five studies were performed within mixed populations.13, 19, 20, 23, 31, 36 The two non–English articles did not provide specific information regarding ethnic labels, but these studies were assumed to have investigated Italian22 and Spanish25 populations. Nine studies5–7, 22, 23, 26, 27, 32, 34 used self–administered questionnaires to assess family history of prostate carcinoma, whereas 15 studies8–10, 13, 15–21, 24, 28–31, 33, 36 used interview–administered questionnaires. Three studies11, 12, 14, 25, 35, 37 consulted registries to retrieve information about family history. The mean age at diagnosis of patients across all studies is 66.7 years (range, 56.6–74.5 yrs; standard deviation, 4.3 yrs; Table 1).
Table 1. Study Characteristics of Published Epidemiologic Studies Concerning Family History of Prostate Cancer and Prostate Cancer Risk
No. of patients
Mean age (yrs)
NA: not available; 45+: greater than 45.
Mean age was not given, but recalculated from the original report.
Retrospective case-control study in which controls were recruited from the general population.
Retrospective case-control study in which controls were recruited from hospitals.
Prospective cohort study.
Structured interview using questionnaire.
Separate effect estimates available for African American study participants.
Separate effect estimates available for Caucasian study participants.
Retrospective case-control study in which controls were recruited from the general population using neighbors.
Cohort study in which family history was measured retrospectively.
Retrospective case-control study in which controls were recruited from spouses.
Family history was assessed using questionnaires for controls and medical records for patients.
Separate effect estimates available for Asian American study participants.
Separate effect estimates available for Hispanic study participants.
Only patients with family history of prostate cancer were reported.
We could not visually (Fig. 1) or statistically (P = 0.64, 0.71, 0.38, and 0.62 for father, brother, first–degree family members, and second–degree family members, respectively) identify heterogeneity in funnel plots.
Summary Recurrence Risk Ratios
Almost all studies provided information on the association between history of prostate carcinoma among first–degree family members and prostate carcinoma risk5–7, 9–29, 31–33, 35–37 (Fig. 2). Random effect pooling revealed a summary recurrence risk ratio (Sλ1) of 2.53 (95% CI, 2.24–2.85). The risk of prostate carcinoma appeared to be higher for men with an affected brother (Sλb, 3.37; 95% CI, 2.97–3.83)6–10, 13, 16, 18, 21, 23, 24, 26, 27, 29–31, 34, 37 than for men with an affected father (Sλf, 2.17; 95% CI, 1.90–2.49).6–10, 12–16, 18, 21, 23, 24, 26, 27, 29, 31, 33, 34, 37 In contrast, the recurrence risk ratio for men who had second–degree relatives with prostate carcinoma was only slightly elevated (Sλ2, 1.68; 95% CI, 1.07–2.64)16, 18, 24, 27, 29, 31 (Figure 2). The risk of prostate carcinoma for first–degree family members increased with increasing number of affected relatives, from 2.57 (95% CI, 2.32–2.84) for men who had 1 first–degree family member with prostate carcinoma9, 10, 16, 19, 21, 29, 31, 37 to 5.08 (95% CI, 3.31–7.79) for men who had 2 or more affected family members.9, 10, 16, 19, 21, 29, 31, 34, 37 The summary recurrence risk ratio decreased with increasing age of first–degree family members and the patient's age at diagnosis. Sλ1 was 3.34 (95% CI, 2.64–4.23) for family members age < 65 years6, 12–14, 16, 21, 29 and 2.35 (95% CI, 2.05–2.70) for family members age > 65 years.6, 12–14, 16, 21, 29 Furthermore, Sλ1 was 2.47 (95% CI, 1.71–3.55) for first–degree family members of patients who were diagnosed at age < 65 years12, 14, 21, 31 and 1.72 (95% CI, 1.41–2.10) for first–degree family members of patients who were diagnosed at age > 65 years.12, 14, 21, 31
We further examined the recurrence risk ratio for first–degree family members according to study design, publication year, geographic area, ethnic label, and family history assessment technique in order to explore the influence of these parameters on outcome estimates (Fig. 3). Subset–specific Sλ1 values did not differ substantially from each other, although Sλ1 for prospective studies appeared to be somewhat lower than for retrospective studies. Sλ1 was higher for studies conducted before 1990 compared with studies conducted after 1990 and for studies conducted in Europe compared with studies conducted in the US or Canada. Studies investigating Caucasian populations reported slightly lower Sλ1 values than studies investigating other populations. Sλ1 was slightly larger when family history was obtained through interviews than when family history was self–reported. However, none of these differences were statistically significant (Pdesign = = 0.22, Pera = 0.64, Parea = 0.35, Pethnicity = 0.11, Passessment = 0.84) (Figure 3). Sensitivity analysis of the study aggregated mean age at diagnosis did not reveal a substantial difference in the risk estimates.
The association between family history and prostate carcinoma risk has been investigated extensively in 33 epidemiologic studies. These studies can be considered the best currently available data for the determination of the empiric risk for men with a family history of prostate carcinoma. The findings suggest a substantially increased prostate carcinoma risk for relatives of a patient. The nature of this familial clustering is such that the recurrence risk increases with decreasing age at diagnosis of the patient and decreasing age of other affected family members, and with increasing genetic relatedness of the affected relative and increasing number of individuals affected within a family. We observed a higher risk of prostate carcinoma in individuals with an affected brother compared with those who had an affected father.
The studies that were reviewed have been conducted in a number of different time frames and geographic regions by investigators who used a variety of methodologic and analytic techniques. Because of potential heterogeneity in populations, designs, and analyses of various studies, we assumed that the true effects being calculated would be affected by variation among studies in addition to the usual sampling variation within studies. To account for both sources of variation, we used random effects meta–regression analysis to combine the results from the primary studies.61 The random effects approach provides some allowance for heterogeneity in studies beyond sampling error.
The results from sensitivity analyses suggested that the recurrence risk ratios were consistent across studies that differed in publication year, geographic area, ethnic label of the study populations, and family history assessment techniques. Also, the recurrence risk for Caucasians was not significantly different from the risk for other populations. This finding might be explained by the knowledge that the prevalence of alleles that predispose an individual to familial or hereditary prostate carcinoma does not differ among populations or that environmental sharing within family units is similar across populations.
Retrospective studies may yield overstated effect estimates for family size; in these studies, the probability of recruiting a patient is positively associated with family size. Because the probability of having an affected family member also depends on the number of relatives a patient has, the proportion of patients with affected relatives can be higher than the proportion of controls with affected relatives even in the absence of familial clustering of prostate carcinoma.62–64 In prospective studies, however, this “length bias sampling” does not occur, as cases originate from a predefined cohort that is not dependent on family size. Indeed, in this study, the summary estimates for retrospective studies appeared to be somewhat higher than for prospective studies, although the difference was not statistically significant. It is unfortunate that, because the original studies did not report on family size sufficiently, we could not allow for this variable in the meta–analysis.
Most studies5–10, 12–15, 17–22, 25–28, 30, 32, 35–37 reported that the diagnoses of patients with prostate carcinoma were confirmed medically. For accuracy of data, diagnoses of prostate carcinoma should have been verified in family members as well. Most studies stated that diagnoses of prostate carcinoma in family members were self–reported. Only four studies could confirm diagnoses in family members using cancer registries.6, 11, 12, 14, 35, 37 It is reassuring that Bratt et al.6 demonstrated that self–reported family history of prostate carcinoma is a valid estimate of prostate carcinoma incidence compared with medical confirmation using carcinoma registries.
We did not attempt to uncover unpublished observations and excluded studies that did not meet the predetermined criteria. Publication bias might arise by excluding these studies. However, we could not visually or statistically identify funnel plot heterogeneity in our meta–analysis.
Familial aggregations of prostate carcinoma that were found in these epidemiologic studies cannot serve to identify genetic liability in the absence of specific genetic tests. Such familial clustering may be due not only to genetic risk factors for disease, but also to common exposure of relatives to environmental carcinogens.12, 42 However, there are features of this clustering that may suggest a genetic component. The genetic contribution to diseases of complex origin, such as cancer, often is most salient to families of patients with early onset. Therefore, one feature of an inherited form of prostate carcinoma is the increased clustering of prostate carcinoma in families of patients with early onset; clustering of this type is shown in the current meta–analysis. In addition, the observation of a higher recurrence risk for family members who had a brother with prostate carcinoma compared with those who had an affected father is consistent with a recessive, or X–linked, genetic component of prostate carcinoma susceptibility. The observation that the recurrence risk ratio for first degree family members is higher than for second degree family members is expected in complex inheritance. Nevertheless, the higher recurrence risk among brothers can also be rationalized by the fact that many individuals with affected brothers experience an earlier onset of prostate carcinoma than those with affected fathers, which might indicate a larger (maternal) genetic influence.
Twin studies are more suitable for distinguishing shared environmental and genetic factors. The National Academy of Sciences Twin Cohort, which comprises nearly 16,000 male veteran twins, revealed a monozygotic concordance for prostate carcinoma of 27.1%, compared with 7.1% for dizygotic twins. This finding provides strong evidence of the influence of genetic susceptibility on prostate carcinoma in which multiple loci might be involved. The authors estimated the narrow–sense heritability of liability to be 57%.65 These findings have been confirmed by two Swedish twin registries, both linked to the Swedish Cancer Registry, that have identified incidences of carcinoma diagnosed from 1959–1992 in twins born from 1886 onward.66
Segregation analyses also support the idea of genetic susceptibility to prostate carcinoma.58 Susceptibility genes with autosomal dominant transmission may account for about 9% of all prostate carcinoma incidences, with penetrances of 88–89% by age 85 years.40, 59 These findings led to the development of the Hopkins Criteria for families that have a hereditary high risk of prostate carcinoma; a family has a high risk if either 1) prostate carcinoma has occurred in 3 or more first–degree relatives; 2) prostate carcinoma has occurred in 3 successive generations of either the maternal or paternal lineages; or 3) 2 relatives have been affected at age < 55 years. Linkage analysis of such families has suggested the existence of susceptibility genes on chromosomes 1p36, 1q24–25, 1q42.2–q43, 20q13, and Xq27–q28, although age–related penetrances associated with mutations in any of these loci have not been reported.67 Evidence for RNaseL as an HPC1 gene at the HPC1 locus on chromosome 1q24–q25 recently has been presented.68–71 It has been inferred that such high–risk genes might function as tumor suppressors.39, 68
Heritable prostate carcinoma with fewer affected family members and therefore lower recurrence risk ratios may result from polymorphisms in other genes. Among the candidate genes are those associated with androgen production or response, including the androgen receptor (AR), steroid 5–α–reductase Type II (SRD5A2), adrenal hyperplasia V(CPY17), aromatase (CYP19), and 3–β–hydroxysteroid dehydrogenase (HSD3B2) Type II genes.72–78 The prevalence of high–risk alleles of some of these genes is also in agreement with the data regarding the risk of prostate carcinoma by ethnic group.76, 79 However, in one study, polymorphisms in the AR gene did not appear to play a major role in familial prostate carcinoma.80 Thus it is not yet clear which genetic tests should be offered to men who have a brother or father affected by prostate carcinoma at age < 55 years.
Nevertheless, physicians and genetic counselors can use this information about the risks of prostate carcinoma associated with positive family history to counsel men who currently are unaffected by the disease. As estimated by the current meta–analysis, the empiric risk for first–degree relatives of a man affected by prostate carcinoma is more than twice the risk for patients who have no family history of the disease. The recurrence risk ratio can be multiplied by the lifetime risk of prostate carcinoma for the general population to estimate the absolute risk of prostate carcinoma for an individual with an affected family member. Epidemiologic studies to date have revealed no other risk factor that is as consistently and strongly associated with the development of prostate carcinoma as a positive family history of the disease.
Screening for prostate carcinoma is endorsed by some, but not all, professional groups.81 Those who argue against prostate carcinoma screening believe that screening leads to the detection of too many incidental, insignificant cancers. However, proponents of screening argue that since the widespread advent of prostate–specific antigen (PSA) testing, there has been a migration to lower stages of disease at diagnosis, which ultimately would lead to an overall reduced cause–specific mortality rate.82, 83 Men with a positive family history of disease constitute an easily identifiable high–risk group that could benefit from PSA screening at an earlier age and at shorter intervals compared with the general male population.40, 42 Ultimately, support for the argument in favor of prostate carcinoma screening will come from a decrease in cause–specific death rates. Ongoing trials are in the process of evaluating the question of whether screening is desirable.