A systematic review and meta-analysis of familial prostate cancer risk

Authors


Abstract

OBJECTIVE

To identify published studies quantifying familial prostate cancer risks in relatives of prostate cancer cases and, by meta-analysis, obtain more precise estimates of familial risk according to the family history.

METHODS

Thirteen case-control and cohort studies were identified which have reported risks of prostate cancer in relatives of prostate cancer cases. Pooled estimates of risk for various categories of family history were obtained by calculating the weighted average of the log relative risk (RR) estimates from studies.

RESULTS

The pooled RR (95% confidence interval) in first-degree relatives was 2.5 (2.2–2.8). There was evidence that this was highest in relatives of cases diagnosed before age 60 years and that RRs declined with age. The risk for the few men with two affected relatives was increased 3.5-fold (2.6–4.8). RRs to sons of cases appeared to be lower than in brothers; a complete explanation of this observation is uncertain.

CONCLUSION

Men with a family history of prostate cancer have a significantly greater risk of developing prostate cancer than those with no such history. Risks are greatest for relatives of cases diagnosed when young and those with more than one relative affected.

INTRODUCTION

Prostate cancer is one of most common non-cutaneous malignancies and a significant cause of cancer mortality in men in the UK. In 1998, ≈ 17 000 cases of prostate cancer were diagnosed, and ≈ 8500 deaths were caused by the disease [1,2]. Excluding screen-detected disease, a large proportion of prostate cancer is asymptomatic unless advanced, with 5-year survival rates reflecting that ≈ 40% of affected men have advanced metastatic disease at presentation [3]. The serious public health problem that this disease presents in Western countries is likely to be further compounded by the increasing age of the population.

Age-adjusted incidence and death rates from prostate cancer vary significantly among countries, even allowing for differences in the availability of screening programmes [2]. While variation in prostate cancer risk is likely to reflect in part the differences in exposure to environmental/lifestyle risk factors, no consistent aetiological risk factor has been identified [4]. However, epidemiological studies have consistently noted the familial clustering of the disease. The risk of prostate cancer in a first-degree relative (father, brother or son) increases a man's lifetime risk of the disease by 2–8 times [5,6], the strength of the relationship varying according to the age at diagnosis in the index case, type of relative and number of relatives affected.

There is considerable interest in estimating familial prostate risks, for genetic counselling and as a guide to determining entry into screening programmes and trials of chemoprevention. This requires risks to be quantified precisely. Compared with breast cancer, there have been fewer studies of familial prostate cancer risk. The purpose of the present study was to systematically review published studies that have quantified familial risks of prostate cancer and to summarize their findings through a meta-analysis.

METHODS

Published reports were identified using the electronic database Medline (National Library of Medicine, USA) for the years 1966 to August 2002 inclusive, using the terms ‘prostate’ in conjunction with the terms ‘family’, ‘familial’, ‘cohort’ and ‘case-control’. Additional articles were ascertained by hand searching through references cited in these publications. This procedure was also used to identify studies published before 1966. Few studies provide data on prostate cancer risks in other than first-degree relatives to calculate robust estimates of risk in these men. This systematic review is therefore restricted to studies that reported risks in first-degree relatives of cases.

The characteristics of the studies were extracted from published articles and summarized consistently to aid comparison. When more than one study used the same data, care was taken to include only one of the studies in the meta-analysis. Similarly, if a study had been superseded, the more recent study was used in the analysis.

STATISTICAL ANALYSES

A meta-analysis was undertaken to obtain a pooled estimate of familial prostate cancer risks from the published case-control and cohort studies. The summary statistics for case-control and cohort studies are the odds ratio and the ratio of observed to expected numbers of cases, respectively. For the purpose of the present analysis, both estimates of risk were considered to represent relative risks (RRs). Where both crude and adjusted estimates of risk were presented the adjusted estimates were used in the meta-analysis.

Pooled estimates of RR were obtained by calculating the weighted average of the logarithm of RRs [7]. Studies were weighted according to the inverse of the variance of logRR [7]. In cases where no CI was provided or could not be estimated, we adopted the same strategy proposed by Pharoah et al.[8], i.e. if the RR was reported as significant the variance was computed assuming the lower 95% CI was 1.0.

To accommodate the possibility of heterogeneity among studies, a random-effects model was also used to derive RRs [9]. This model assumes that the studies in question are a random sample of a hypothetical population of studies, taking into account within- and between-study variability. The between-study variance is used to modify the weights used to derive the summary statistic. The pooled estimate of effect obtained from the random-effects model tends to be more conservative than when a fixed-effects model is used.

To assess the presence of publication bias, study RRs were plotted in order, according to the variance of the logRR estimate. Estimates from small studies that have less precision in estimating the underlying RR will scatter widely at the top of the plot, with a narrowing among larger studies. In the absence of publication bias, the plot resembles a symmetrical funnel. Conversely, if there is bias, the funnel plot will be asymmetrical [10]. For statistical manipulations the statistical program STATA was used (version 6.0, Stata Corporation, TX, USA) with the META module (Stata Technical Bulletin 38 and 42).

RESULTS

Twenty-two reports (either case-control or cohort) of prostate cancer in relatives of prostate cancer cases were identified [5,6,11–30]. Two of these provided no data on risks in first-degree relatives of cases [11,12]. The case-control study reported by Ghadarian et al.[6] was superseded by an analysis published later [22]. Similarly, the analyses of the Swedish National Cancer Registry [26,30] were also superseded [25].

Several studies were based on the analysis of the Utah Mormon Database [5,15,28,29]; to avoid duplication, the two analyses [5,14] were used, the latter specifically for estimating of risk in young relatives of cases diagnosed when young. Table 1 summarizes the details of the studies included in this overview.

Table 1. A summary of studies of prostate cancer risk in first-degree relatives of prostate cancer cases
ReferenceYear ofpublicationPlace of studyDate ofstudyAge, yearsNumber ofsubjects
  • *

    Metropolitan Detroit Cancer Surveillance System, New Jersey State Health Department.

Case-control studies
[ 13 ] 1977Minneapolis, USANot stated6440 cases
43 hospital controls
35 population controls
[ 14 ] 1982Utah, USANot stated 2824 cases
5648 population controls (Utah Cancer Registry)
[ 15 ] 1990Maryland, USA1982–8934–76691 cases
640 spouse controls
[ 16 ] 1991Texas, USANot stated 385 cases
385 hospital controls
[ 17 ] 1994San Francisco, Los Angeles, Hawaii Vancouver, Ontario1987–91All1500 cases
1581 population controls
[ 18 ] 1995Missouri, USANot stated38–871084 cases
935 spouse controls
[ 19 ] 1995Fulton & DeKalib Counties*1986–8940–79981 cases
1315 population controls
[ 20 ] 1995Baltimore, USA1982–8934–76690 cases
640 spouse-partner controls
[ 21 ] 1996Massachusetts, USA1992–94< 70563 cases
703 population controls
[ 22 ] 1997Montreal, Toronto,  Vancouver, Canada1989–93All640 cases
639 population controls
[ 23 ] 1998JamaicaNot stated72.0 (mean)263 cases
263 hospital patient controls
[ 24 ] 1999Lund, Sweden1995–9644–94356 cases
256 hospital patient controls
356 population controls
Cohort studies
[ 5 ] 1994Utah, USA −  1994All1376 cases
[ 25 ] 2002Sweden (National) −  2001All fathers
Brother < 67
569 cases

Thirteen studies [5,13,15–25], reporting the risk of prostate cancer associated with having at least one first-degree relative affected, are summarized in Fig. 1a. All studies found a positive relationship between having an affected first-degree relative and the risk of prostate cancer, with RRs of 2.0–3.9; using all studies the pooled RR (95% CI) was 2.5 (2.2–2.8).

Figure 1.

Figure 1.

Funnel plots of the RR of prostate cancer associated with: a , having one or more first-degree relatives affected with the disease, or with a history of the disease in b , a brother, and c , the father. Values are plotted in order of decreasing variance of the logRR. The green horizontal lines represent the 95% CI; each red box represents the RR point estimate and its area is proportional to the weight of the study. The black diamond (and broken line) represents the overall summary estimate, with the CI given by its width. The unbroken vertical line is at the null value (RR 1.0). *restricted to age < 67 years.

Figure 1.

Figure 1.

Funnel plots of the RR of prostate cancer associated with: a , having one or more first-degree relatives affected with the disease, or with a history of the disease in b , a brother, and c , the father. Values are plotted in order of decreasing variance of the logRR. The green horizontal lines represent the 95% CI; each red box represents the RR point estimate and its area is proportional to the weight of the study. The black diamond (and broken line) represents the overall summary estimate, with the CI given by its width. The unbroken vertical line is at the null value (RR 1.0). *restricted to age < 67 years.

Figure 1.

Figure 1.

Funnel plots of the RR of prostate cancer associated with: a , having one or more first-degree relatives affected with the disease, or with a history of the disease in b , a brother, and c , the father. Values are plotted in order of decreasing variance of the logRR. The green horizontal lines represent the 95% CI; each red box represents the RR point estimate and its area is proportional to the weight of the study. The black diamond (and broken line) represents the overall summary estimate, with the CI given by its width. The unbroken vertical line is at the null value (RR 1.0). *restricted to age < 67 years.

Eight studies reported separate risks in father-, son- and brother relationships [15–19,21,22,25]; all found greater risks associated with both types of family history. In all but one of the studies [22] the RRs in brothers were higher than in fathers. Estimates of risk in brothers reported by Hemminki and Czene [25] were truncated at age 66 years; in view of this the risks abstracted for the father-son relationship were truncated to the same age. Estimates of father-son RRs were 1.9–3.8 and the pooled estimate of the RR was 2.5 (2.1–3.1). Estimates of RRs in brothers were 2.6–5.3 and the pooled estimate 3.4 (2.9–4.1). Omitting the study reported by Hemminki and Czene [25] had little effect on the pooled estimate of RR, at 2.4 (2.1–2.9) and 3.3 (2.7–4.0) in father-sons and -brothers, respectively.

Four studies reported risks according to the ages of first-degree relatives of index cases [17,19,21,24,]; all reported the risk of prostate cancer to be higher in younger relatives. For men aged < 65 years with an affected first-degree relative, the RR of prostate cancer was 4.3 (2.9–6.3), while the RR for those aged> 65 years was 2.4 (2.0–2.9).

Two studies [5,25] reported the RR of relatives according to age of diagnosis or death in the index case. Goldgar et al.[5] estimated the RR in relatives of cases diagnosed before age 60 years to be 5.1 (2.4–10.0). Hemminki and Czene [25] provided a very similar estimate of risk, albeit truncated to age 67 years, of 3.8 (2.9–5.1); pooling the estimates the RR was 3.9 (2.9–5.1).

Pooling data from the studies, the RR in relatives of cases diagnosed before age 60 years was 4.3 (2.9–6.3). A similar trend of an increased risk was apparent for the age of relative. Two studies reported risks according to age of the index case; in relatives of cases diagnosed before age 60 the RR was 2.9  (1.9–4.5).

Two studies [14,25] provided estimates of risk by age of the index case and age of the relative. Both found that men with the highest RR were young with a relative diagnosed young. For relatives aged < 65 years who were relatives of cases diagnosed before age 65 the pooled RR was 3.4 (2.7–4.2).

Five studies reported the risk of prostate cancer associated with having more than one affected first-degree relative [15,17,21,22,25]; in all but one [22] the risk to relatives was greater than having only one affected relative. Estimates of risk were 2.8–9.4 (Fig. 1b,c), with a pooled estimate of the familial RR of 4.6 (2.7–8.0). However, the estimate provided by one study [25], of 9.4 (5.8–14.0), was restricted to fathers and brothers diagnosed before age 67 years, which inevitably inflates the estimate. Omitting this study the pooled estimate of familial risk was 3.5 (2.6–4.8).

DISCUSSION

Published studies were identified that have reported the risk of prostate cancer in relatives of prostate cancer cases. In ascertaining studies, a systematic review process was adopted, avoiding selection on the basis of study quality. Using these studies it was possible to derive pooled estimates of risk for several different types of family history. The results provide strong evidence that familial prostate cancer risks are closely related to the age at diagnosis and increase markedly with many affected relatives.

As with any systematic review there are several sources of potential bias that affect the results of the meta-analyses. The first is whether all relevant articles were identified; we consider that we identified the key published studies in which estimates of prostate cancer risks in first-degree relatives was a major component and thus avoided selection bias. The second source of bias is publication bias; this is a major concern in all forms of meta-analysis, because studies reporting significant or positive findings are more likely to be published [10]. It is not unusual for small early studies to report a positive relationship or a large effect that subsequent much larger studies cannot replicate. The third source of bias is heterogeneity, but the importance of heterogeneity between studies on summary estimates of familial risk is difficult to assess. Given the differences in study groups, some degree of heterogeneity is to be expected. Selection of cases on the basis of family history is another possible source of heterogeneity, but there is no evidence that any of the studies were enriched for familial cases per se.

Another potential source of bias is recall bias. Many studies reported to date have placed heavy reliance on unverified diagnoses in relatives. Most studies that have estimated familial prostate cancer risks have been either case-control or retrospective cohort in design, and it is conceivable that cases are more likely to provide a positive history of prostate cancer. With the exception of the early small studies this is unlikely to be a major issue, as the estimates of risk from the larger studies are similar to those obtained in the studies based on cancer registry data.

The advent of widespread PSA screening for prostate cancer in recent years may significantly bias future estimates of familial risk. Disease diagnosed by PSA does not necessarily equate to invasive disease, on which existing estimates are primarily based. Furthermore, relatives of cases may be more likely to seek screening and be diagnosed with in situ or pre-invasive disease than men with no family history of the disease, so biasing estimates.

This systematic review provides strong evidence that familial prostate cancer risks are closely related to age at diagnosis and are markedly greater with many affected relatives. This undoubtedly reflects, in part, the higher probability that there is a major genetic component to early-onset disease. A recent twin study indicated that the heritability of prostate cancer is high, at ≈ 42%[31], but hereditary disease appears to be heterogeneous. Genome-wide linkage searches have implicated several regions in inherited prostate cancer, e.g. HPC1 (1q24–q25), PCaP (1q42-q43), HPCX (Xq27–q28), CAPB (1p36), HPC20 (20q13), HPC2/ECAC2 (17p11) and 10q25 [32–40], but confirmatory studies at these loci have produced discordant results. Although an autosomal dominant model has been shown to provide the best model of the familial aggregation of prostate cancer in many studies [41–46], some analyses suggest that multiple alleles, each conferring modest risks such as RNASEL variants [47], may be responsible for most inherited disease [48]. The involvement of low-penetrance alleles whose effect may be modified by environmental effects, together with locus heterogeneity, provides an explanation for the difficulties with identifying a gene through linkage.

Most studies suggest that risks in brothers are greater than in father-son relationships. Familial aggregation of prostate cancer can be caused by genetic factors that are not inherited. Hence, in the context of an increasing population incidence of prostate cancer, and possibly increasing exposure to environmental risk factors, differences in familial risk would be compatible with a gene–environment interaction. Alternatively, from a purely genetic perspective, such a pattern of familial risk (i.e. higher risks in brothers than in father-son relationships) would be indicative of the action of recessively acting susceptibility alleles. It is impossible to preclude that a difference in risk in fathers compared with brothers could be the result of differential recall in case-control studies, whereby cases are more likely to report an affected sibling than a parent. In addition, failure to compare age-specific risks for fathers and brothers is an issue. Where this has been addressed a difference has still been reported [25] but it is less profound. It is possible that the residual difference partly reflects changing temporal trends.

There is now increased concern about the early detection of prostate cancer, heightened by the ageing population. However, the conclusion that prostate-cancer screening applied to the general population is cost-effective is contentious [46,47]. Most studies have found no evidence for a significant difference between familial and sporadic forms of prostate cancer in terms of either clinicopathological features, response to treatment, or outcome [48–51]. Given the increased risk of the disease in men with a father or brother with early-onset disease (defined by an age of onset of < 60 years, or with two affected first-degree relatives) screening these men should increase the yield and might prove to be a cost-effective strategy.

Abbreviations
RR

relative risk.

Ancillary