Risk of developing a second primary cancer in male breast cancer survivors: a systematic review and meta-analysis

Background With increasing survival after cancer diagnoses, second primary cancers (SPCs) are becoming more prevalent. We investigated the incidence and site of non-breast SPC risks following male breast cancer (BC). Methods PubMed, Embase and Web of Science were systematically searched for studies reporting standardised incidence ratios (SIRs) for SPCs published by March 2022. Meta-analyses used the generic inverse-variance method, assuming a random-effects model. We evaluated SIRs for overall SPCs, site-specific risks, by age at BC onset, time since BC onset and geographic region. We assessed study quality using routine techniques. Results Eight population-based retrospective cohort studies were identified. SIRs ranged from 1.05 to 2.17. The summary SIR estimate was 1.27 (95% CI: 1.03–1.56, I2: 86%), and there were increased colorectal (SIR: 1.29, 95% CI: 1.03–1.61), pancreatic (SIR: 1.64, 95% CI: 1.05–2.55) and thyroid (SIR: 5.58, 95% CI: 1.04–30.05) SPC risks. When an outlying study was excluded, the summary SIR for men diagnosed with BC before age 50 was 1.50 (95% CI: 1.21–1.85), significantly higher than men diagnosed at older ages (SIR: 1.14, 95% CI: 0.98–1.33). Conclusions Male BC survivors are at elevated risks of developing second primary colorectal, pancreatic and thyroid cancers. The estimates may assist their clinical management and guide decisions on genetic testing.


BACKGROUND
Male breast cancer (BC) is rare, accounting for less than 1% of all BC cases [1,2]. As a result, few studies have investigated the risks of second primary cancers (SPCs) following male BC [3][4][5][6][7][8][9][10][11][12]. SPCs in male BC survivors are a growing health problem. The agestandardised incidence rate of male BC rose by 40% between 1975 and 2015 [13], whereas the age-standardised male BC specific mortality rate decreased by 22.5% and 12.4% between 2002 and 2016 in the European Union and the USA, respectively [14]. Most clinical management guidelines for male BC are extrapolated from information on BC in postmenopausal women [2], so this review could better inform clinical management decisions regarding SPC prevention measures following male BC.
No systematic review of SPC risks following male BC has been performed since 2008 [15]. No meta-analysis of SPC risks following male BC has been carried out to date. We therefore aimed to conduct a significantly updated systematic review (SR) and a novel meta-analysis of SPC risks in male BC survivors. Our objective was to review the latest evidence regarding the risks of developing SPCs following a first invasive primary male BC. A further objective was to assess site-specific second cancer risks among studies that also investigated the overall SPCs risks. Our final objective was to evaluate the variability in non-breast SPC risks by confounding variables, such as patient characteristics.

Exposure, outcome and measures of association
The exposure was defined as a previous first primary invasive male BC, with no prior cancer history. The outcome was defined as a non-breast SPC.
To minimise misclassification of recurrences or metastases of the first BC as second primaries, SPCs were determined using one of two possible sets of guidelines: those given by the Surveillance, Epidemiology and End Results (SEER) programme [16], primarily used in North America [17] and those given by the International Association of Cancer Registries (IACR)/International Agency for Research on Cancer (IARC) [18,19], used in all other regions [17]. An explicit statement that SPCs had been confirmed by a physician, with efforts made to differentiate SPCs from recurrences or metastases, was accepted if the guidelines used were unstated.
Second primary BC counts following a first BC are not comparable under the SEER and IACR/IARC guidelines, as observed in a 2014 study of SPC counts [17], as the different guidelines take different approaches to coding SPCs in paired organs [20]. However, the same study [17] found non-breast SPC counts to be almost identical under either set of guidelines. Therefore, only non-breast cancers were considered as SPCs in this review.
The chosen measure of association was the standardised incidence ratio (SIR), which compares the incidence of non-breast SPCs among men with a prior first primary BC to the corresponding expected incidence of nonbreast primaries in the general male population.

Data sources and search strategy
PubMed, Embase and Web of Science were each searched for relevant studies on March 11, 2022, using queries described in the Supplementary Material.

Inclusion and exclusion criteria
Studies were considered for inclusion if a SIR and associated standard error could be extracted that assessed the combined risk of non-breast SPCs in male BC survivors, they focussed primarily on adults and they were written in English. A final inclusion criterion was that a study must use IARC/IACR or SEER rules to identify SPCs, or if this was unstated, must state that diagnoses of any SPCs had been confirmed by a physician, with efforts made to differentiate from recurrences or metastases. Studies were excluded if they reported solely on SPC risks following a specific treatment (or lack thereof) of the first male BC, they reported solely on SPC risks following a non-invasive first male BC, or they had a cohort of fewer than 100 male BC survivors.
Studies with data overlapping entirely with another study were also excluded. Partially overlapping studies were included in the SR, although only the larger study was included in any meta-analyses. Data from the Swedish Family Cancer Database were considered to overlap with data from the Swedish national cancer registry due to close links between these resources [21].

Data extraction
Titles and abstracts were screened independently by two authors, with a third author resolving any conflicts. For each study, the first author, publication date, country and centre of data derivation, design, time period, follow-up and definitions of the cohort and of SPCs were extracted, together with additional fields such as stratification details and sample sizes. One author was contacted for clarification. The extracted data were input into a Microsoft Excel table.

Statistical analysis
All statistical analyses were performed in R version 4.1.2 [22]. For each eligible study, the SIR of developing any non-breast SPC following an invasive first primary male BC was extracted as the principal summary measure. Meta-analyses were performed using the random-effects generic inverse-variance method, with DerSimonian-Laird estimators [23,24]. Standard errors were extracted by dividing the square root of observed non-breast SPC counts by the corresponding expected counts and were converted to the natural logarithm scale by dividing the result by the corresponding SIR [25]. When unreported, expected SPC counts were estimated by dividing observed SPC counts by SIRs. Unreported confidence intervals (CIs) were estimated using Byar's approximation, assuming observed SPC counts followed a Poisson distribution [25].
We performed unstratified meta-analyses and also stratified by age and time elapsed since the onset of the first BC. The stratification point for age was set at 50 years, although data on men aged up to 60 at BC onset were added into the younger group if no stratification at 50 was provided. We also performed two separate meta-analyses, respectively stratifying at 5 years and 10 years post diagnosis of the first BC. We considered reported SIRs stratified at 9 years equivalent to reported SIRs stratified at 10 years.
We also performed sixteen further meta-analyses, respectively evaluating second cancer risks at the following specific sites: bladder, blood (leukaemia, myeloma and non-Hodgkins lymphoma), brain and central nervous system (CNS), colorectum, head and neck, kidney, liver, lung, oesophagus, pancreas, prostate, skin (melanoma), stomach and thyroid. These are the male-specific subset of the 20 most common cancer sites in the UK from 2016 to 2018, after excluding BC and cancer of unknown primaries [26]. Since the purpose was to examine the distribution of the sites of any observed combined SPC risks, a study providing a SIR and associated standard error of developing cancer at a specific site was included in the corresponding site-specific metaanalysis only if it was also included in the meta-analysis that was unrestricted by SPC site.
We assessed between-study heterogeneity using Cochran's Q [27] and the I 2 statistic [28,29]. Publication bias was assessed using funnel plots and Egger's test [30]. A study was regarded as an outlier if there was no overlap between the study-specific and pooled (unstratified) meta-analysis confidence intervals [31]. All meta-analyses were performed first, including, then excluding, outlier studies. The results of all meta-analyses were visually represented as forest plots. Further sensitivity analyses took the form of subgroup analyses testing the effect of the geographical region (continent) of data derivation. Differences between summary SIRs based on multiple different datasets, such as different age groups, were assessed by treating each set of data as a subgroup and comparing the resulting Cochran's Q to a chi-squared distribution with the degrees of freedom being the number of subgroups minus one [31]. P values of less than 0.05 were deemed significant.
Study quality was assessed using the Newcastle-Ottawa scale (NOS) [32] (details in Supplementary Material).

Results of literature search
The database searches yielded 2011 studies following deduplication, 46 of which were deemed suitable for full-text screening as well as bibliography sweeping. To ensure the capture of all relevant studies, we also swept the bibliographies of 26 studies deemed unsuitable for full-text screening solely due to their focus on female BC survivors. Overall, the bibliography sweeps yielded 33 additional studies for full-text screening. In total, eight studies were included in the SR (Fig. 1 Fig. 1 Search process. The search process used to identify the studies in this review, as described in the Methods section.
Further study characteristics are described in Tables 1 and 2. NOS scores may be seen in the Supplementary Material.

Results of meta-analyses
To aid the interpretation of the results of the meta-analyses, it should be noted that one study reported two sets of SIRs, including and excluding data from the first 2 months of follow-up [6]. The latter results were described by the study as the more reliable and hence were used in the meta-analyses. In addition, one study pooled their data from multiple centres across four continents [5]. This study was regarded as European for any metaanalyses stratified by geographic region, since the bulk of their data was drawn from European registries.
Unstratified results. The unstratified meta-analysis included six studies [3][4][5][6][7][8]. Only the German subset of the data used by Chen et al. [3] was included due to the rest of the data partially overlapping with a much larger study [5]. All studies reported an increase in SPC risks following a first primary male BC. Some variation in the reported SIRs was present, with the largest studies reporting estimates between 1.05 and 1.34 [3][4][5][6]8]. The only Asian study was an outlier, reporting a SIR of 2.17 [7]. There was no significant evidence for publication bias (Supplementary Material).
Whether including or excluding Hung et al., no significant evidence of heterogeneity was found within the continent-specific subgroups (Supplementary Material).
Effects of age at BC onset. The age-stratified meta-analyses consisted of 4 studies [5][6][7][8]. When including Hung et al., we found no significant evidence for a difference between the summary SIR of men aged under 50 at first BC onset and the summary SIR of men aged over 50 at first BC onset (Supplementary Material).
Results when excluding Hung et al. are shown in Fig. 4. There was significant evidence for a difference in summary SIR between the age groups (SIR: 1.50, 95% CI: 1.21-1.85) for those aged under 50 at first BC onset vs. 1.14 (95% CI: 0.98-1.33) for those aged over 50 at first BC onset, P for difference: 0.040).
Effects of follow-up time elapsed since BC onset. We found no significant evidence for the length of time elapsed since the onset of the first BC affecting SPC risks (Supplementary Material).
Site-specific associations. Hung et al. provided sufficient data for inclusion in the meta-analyses assessing the risks of SPCs at ten of the examined sites -the bladder, colorectum, head and neck, kidney, liver, lung, pancreas, prostate, stomach and thyroid. Summary SIRs from these meta-analyses ranged from 1.09 to 5.58. Among these sites, the risks of second primaries were significantly higher than the risks for first primaries for colorectal cancer (SIR: 1.29, 95% CI: 1.03-1.61), pancreatic cancer (SIR: 1.64, 95% CI: 1.05-2.55) and thyroid cancer (5.58, 95% CI: 1.04-30.05). Following the exclusion of Hung et al., there was no significant evidence of elevated cancer risks following male BC for any of these ten sites other than the colorectum (SIR: 1.21, 95% CI: 1.00-1.46)), although all associated point estimates were greater than 1. Hung et al. did not provide sufficient data for inclusion in the meta-analyses of SPC risks at the remaining examined sitesthe blood (leukaemia, myeloma and non-Hodgkins lymphoma), the brain and CNS, the oesophagus, and the skin (melanoma). The summary SIRs generated for these sites ranged from 1.00 for the blood (non-Hodgkins lymphoma) to 1.65 for the skin (melanoma), with no significant evidence for an increased risk of second primaries at any of these six sites.
Full results may be seen in Supplementary Material.

DISCUSSION
Most published studies reporting SPC risks following male BC draw their data from European [3,4,6,9] or North American [8,10,12] population-based cancer registries. The majority reported elevated risks [3][4][5][6][7][8][9]12]. Male BC survivors have often been found to be at greater risk of primary cancers of the prostate [5][6][7]10], skin [5,7,8,10] and digestive system [5][6][7][8], although with varying magnitudes. This systematic review and meta-analysis of such studies confirm the combined risks of non-breast SPCs to be significantly elevated following a first primary male BC. When excluding the outlier study by Hung et al., male BC survivors aged under 50 at the initial BC diagnosis were found to be at significantly higher risk than those over 50. This difference in risks may even have been slightly underestimated, due to our decision to include data on men aged under 60 at first BC diagnosis in the younger stratum when no direct stratification at 50 was provided [33]. We also found significant differences between risks reported by studies from Asia, North America and Europe, although larger studies from a wider range of countries are needed to clarify the extent of any risk differences between geographic regions. Finally, we found that male BC survivors are at increased risk of second colorectal, pancreatic and thyroid cancer.
The results of the age-stratified, continent-stratified and sitespecific meta-analyses differed depending on whether Hung et al. were included. We therefore discuss the robustness of these results here. Firstly, we found significant evidence for SPC risks varying by geographical region only when including Hung et al. Since Hung et al. was the sole Asian study, this indicates that the difference was driven by this study rather than differences between the European and North American studies. Hung et al. is a well-designed study, using a complete and accurate database [34]. It therefore seems this finding may reflect a higher SPC risk in Asian (specifically Taiwanese) male BC survivors than a flaw in study design.
Although  Year/years. c Confidence interval generated using Byar's approximation.
I. Allen et al.
Significant evidence for increased risks of second primary pancreatic and thyroid cancers was only found when Hung et al. was included in the relevant meta-analyses. The largest study in this review also found the risks of pancreatic SPCs to be significantly elevated [5]. There is also evidence of shared risk factors for male BC and pancreatic cancers. For example, pathogenic BRCA1 [35,36], BRCA2 [35][36][37] and PALB2 [37,38] variants are associated with both male BC and pancreatic cancer. Therefore, the finding that male BC survivors are at increased pancreatic cancer risk seems plausible. In contrast, the finding of increased second primary thyroid cancer risks was mainly driven by data from Hung et al. and was based on a total of just 4 observed cases. Although previous BC has been linked to elevated thyroid cancer risks in women [39], larger studies are needed to clarify this association in men. Finally, it should be noted that Hung et al. reported combined risks of colorectal and anal cancers and of lung and mediastinum cancers. Hence, the point estimates estimated for second colorectal and lung cancers when data from Hung et al. were included may be distorted slightly, although the fact that second colorectal cancer risks remained significant even following the exclusion of these data indicates that second colorectal primary risk is likely to be elevated in male BC survivors.
The strengths of this SR include the number of studies with large sample sizes, considering the rarity of male BC [5,6,8,12]. There was also no significant evidence of publication bias (see Supplementary Material). This SR was built on studies of high   Fig. 4 Forest plot showing standardised incidence ratios and a pooled estimate of second primary cancer risks, stratified by age group at breast cancer onset. Association between a first primary male breast cancer and the onset of a non-breast second primary cancer, in comparison to the general male population, stratified by age group at breast cancer onset, excluding the outlying study by Hung et al.    methodological quality, with all studies being assigned NOS scores of 6 or higher. Finally, there was limited heterogeneity among European studies, which was the largest continent-specific subset of studies available.
It is known that BC treatments such as chemotherapy, radiotherapy, or hormonal therapy increase SPC risks in women [40][41][42]. Treatment effects could also partly explain our findings in men. Other non-genetic risk factors which may influence risks for the first primary male BC, such as hormonal imbalances or a family history of male BC [1], may also contribute to the observed elevated SIRs. However, this information was not available in the studies. Notably, in addition to pancreatic cancer, some cancers found to be at greater risk following male BC are also associated with pathogenic variants in genes linked to BC susceptibility in men. For example, both male BC susceptibility [37,43] and colorectal cancer susceptibility [44,45] are associated with pathogenic variants in the CHEK2 gene. We also found some evidence of elevated second stomach and prostate cancer risks when including Hung  This evidence suggests that SPC risks for BC survivors with a genetic predisposition to BC may be increased in comparison to BC survivors without such a predisposition. Research in this area has been undertaken for contralateral BC in women [52], but is otherwise very scarce. There is some evidence that a higher proportion of male than female BC cases are due to pathogenic variants in BC susceptibility genes [53,54], with the largest study of germline susceptibility in male BC cases finding 13.7% of male BC survivors to carry such variants [37]. Pathogenic germline variants in BC susceptibility genes could account for a sizeable proportion of second primaries following male BC, with a recent large study confirming non-breast primaries to be 58% more common among male carriers of deleterious BRCA1/2 variants than among male relatives of carriers who were either untested for, or confirmed not to carry, such a variant [35]. Further research in this area may thus be particularly relevant for male BC survivors. Genetic susceptibility could also account for part of the observed association between early-onset male BC and raised SPC risks, since pathogenic variants in such genes are associated with an earlier age at BC diagnosis [55][56][57]. An additional explanation for this relationship is that more aggressive treatment regimens tend to be offered to younger BC patients [58,59], but these treatments can confer a higher risk of developing SPCs [40][41][42].
The study has some limitations. The estimated SIRs may have been affected by surveillance bias, whereby cancers are detected in BC survivors that would have gone unnoticed in individuals without any cancer history due to increased surveillance [6,60]. However, this was likely reduced by the inclusion of data from four studies [4,6,8,12] excluding SPCs occurring within a time period of at least 2 months immediately following the initial BC diagnosis, where surveillance bias is likely to be most intensive [6]. The paucity of studies reporting effects of treatments of the first male BC [7,8] and the lack of studies reporting the influence of hormonal imbalances and family histories of male BC also meant that we could not adjust for several potential confounders. The rarity of second cancers at certain sites may also mean some analyses were underpowered, as evidenced by the wide confidence intervals. Therefore, it cannot be concluded that other associations do not exist. It also cannot be ruled out that some relevant published studies were missed, although the doublescreening process and the sweeps of reference lists should minimise the likelihood of this.
To our knowledge, this is the first meta-analysis of SPC risks in male BC survivors to have been performed and the first systematic review since 2008. This study provided site-specific SIRs and assessed the variability in the estimates by age at first BC diagnosis, follow-up time and geographical region (continent). Future large cohort studies might consider the effects of BC treatment, family history, or hormonal imbalances, as they receive relatively little focus in the current literature. There is also a clear need for further research on the influence of pathogenic variants in BC susceptibility genes on SPC risks following male BC.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.

DATA AVAILABILITY
The data from the study by the AIRTUM Working Group are publicly available at I tumori in Italia. Rapporto 2013: I tumori multipli | Epidemiologia&Prevenzione (epiprev.it). All data from the remaining studies were taken directly from their corresponding published, publicly available manuscripts or Supplementary Materials.

CODE AVAILABILITY
All codes used to generate the results in this manuscript can be provided upon request.