Risk of second breast cancer in female Hodgkin’s lymphoma survivors: a meta-analysis

Background Women treated for Hodgkin’s lymphoma (HL) have an elevated risk of developing second breast cancer (SBC) compared with the general population. We planned this meta-analysis to quantify the long-term risk of SBC and analyze the contributing risk factors among HL survivors. Methods According to predefined selection criteria, literature search identified 34 studies that were included in the analyses. Results After eliminating overlapping or duplicate data, 957 incidences of SBC were encountered in 24,505 females with HL over a median follow-up of 14.9 years. The medians: age at the diagnosis of HL, age at diagnosis of SBC, and latency since HL treatment to the development of SBC were 23.7, 35.0, and 17.7 years, respectively. The pooled relative risk (RR) of SBC was 8.23 (95% CI, 5.43-12.47, I2 = 96%), with a median absolute excess rate of 22.9 per 10,000 person-years. The RR was found inversely related to age at diagnosis of HL with the highest rate (68.7; [95%CI, 28.08-168.11], I2 = 79%), occurred in young patients (≤ 15 years old), where the RR in older women (≥ 40 years old) was not significant (0.55; [95% CI, 0.09-3.52]). Analysis of RR by 5-year increments since the treatment of HL showed that the risk was highest after 15–19 years of latency (13.87; [95% CI, 7.91-24.30], I2 = 89%). Analysis of the effect of treatment modalities showed that the RR rates were (4.70; [95% CI, 3.28-6.75], I2 = 74%), (5.65; [95%CI, 2.94-10.88], I2 = 91%), and (1.19; [95% CI, 0.50-2.82], I2 = 65%), for radiotherapy (RT) only, combined RT and chemotherapy (CT), and CT only, respectively. To investigate the demonstrated heterogeneity, meta-regression analysis was performed when feasible. In most such analyses, the natural logarithm of RR was inversely associated with age at HL diagnosis. Conclusions We conclude that, the current meta-analysis provided the most recent comprehensive estimate of the risk of SBC in a broad-range of HL survivors. Younger age at diagnosis proved to be a dominant risk factor. The obtained results would serve providing breast cancer screening recommendations for HL survivors.

However, increased risk of second cancer following effective treatment of HL has long been reported [6]. More recently, second neoplasms after HL are being encountered with increasing frequency due to the marked improvement in survival [2,7]. The particular elevated risk of second breast cancer (SBC) among this population is not surprising in view of the reported excess risks of breast cancer (BC) after incidental low doses of ionizing radiation [8,9], therapeutic RT [10,11], or as sequelae of the carcinogenic effects of CT [12,13].
Two pertinent meta-analyses have been published and they have addressed different questions. The first metaanalysis was published in 2006 and examined all second malignancy risk associated with HL treatment in 31 randomized trials and it included 65 incidence of SBC [14]. In the second meta-analysis [15], SBC risk and BC surveillance were investigated in young females (≤ 30 years at the primary tumor diagnosis) receiving moderate to high doses of RT targeted to mantle and modified mantle fields, mediastinum, lung, and thorax [15]. The latter meta-analysis comprised 11 studies that were not restricted to patients with HL but included all primary neoplasms in that age group.
Research on the late consequences of HL has often been limited by the size and composition of the study populations and by the duration and completeness of patient follow-up. To the best of our knowledge, there is no recently published meta-analysis intended to examine the risk of SBC in a broad range of ages at HL diagnosis, various follow-up periods, and subsequent to different therapeutic modalities. Also not precisely known, is the effect of other contributing risk factors. The lack of such data has prompted the current meta-analysis.

Search strategy
Between January 1966 and October 2011, we identified studies of interest by first conducting an electronic literature search of the databases MEDLINE, EMBASE, and the Cochrane Library. We also searched for relevant abstracts in the annual conference proceedings between January 1984 to October 2011 for the American Society of Clinical Oncology, European Society for Medical Oncology, and the San Antonio Breast Cancer Symposium. All ages of HL patients were eligible for inclusion.
We used exploded Medical Subject Heading terms or key words terms 'lymphoma' , 'Hodgkin' , 'Hodgkin's disease' and 'Hodgkin's lymphoma'. The terms were combined with 'neoplasm, second neoplasm, second primary' using the Boolean operator 'and'. Search results were also filtered against the terms 'breast, breast cancer, breast neoplasm). In the second step, these keywords were combined using the Boolean operator 'and' with 'standardized incidence ratio' , 'relative risk' , and 'observed to expected'. In addition, we manually reviewed the reference lists of relevant studies to identify additional pertinent published articles.

Selection criteria
We included studies that met each of following criteria: (i) published in English language between January 1985 and October 2011; (ii) included naive patients at any age and with any stage of HL; (iii) investigated the risk for second malignant neoplasms (SMNs) in HL survivors; (iv) reported relative risk (RR) and/or specified as standardized incidence ratios (SIR) or data allowing such outcomes to be derived; and (v) published as original articles (no case reports, case series, reviews, comments, letters, or editorials).
When two or more references reported duplicate data, we only included in the analysis the most recent data, studies with the longer follow-up, or the most relevant studies. We excluded studies that mainly addressed the clinical characteristics of SBC. We also excluded studies that mainly intended to evaluate the potential benefits and harms associated with breast cancer surveillance among women with HL. Case-control designs, i.e. HL patients who developed BC compared with patients who did not were excluded.

Data extraction
Two authors (KMA, and GAK) independently inspected each reference title identified by the search and applied the inclusion criteria. For possibly relevant articles and in cases of disagreement between reviewers, the full article was obtained and inspected independently by the five authors. The data intended for extraction were discussed, and decisions were documented. We used the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) reporting criteria to assess the quality of studies included in the meta-analysis [16]. Any significant lack of concordance in the scores assigned by authors was discussed to reach a consensus.
Standardized Excel sheet was used for each study that fulfilled the inclusion criteria. Extracted data included paper characteristics (first author's last name, publication year, country in which the study was carried out, and data source), study design, number of HL patients, mean/median age of patients, mean/median duration of follow-up, therapy details, number of observed and expected SBC cases, and RR or standardized-incidence rate (SIR) with corresponding 95% confidence interval (CI). The ratio of observed to expected numbers of cancers, SIR (referred to in the text as RR) was then used or calculated with likelihood-based 95% CI from Poisson models [17]. Where not reported, we computed the CI for the risk assuming a Poisson distribution for the observed number of cases. Standard error (SE) for the natural logarithm of RR (lnRR) was derived from CI, applying the following equation: SE = ln(upper 95% CI/ lower 95% CI)/(2 x z 1 -a/2 ). When appropriate, we also used the built-in calculator of the Review Manager Software (version 5.1.4 for Windows; The Cochrane Collaboration, Oxford, UK) to compute missing data. When studies showed that the observed number of cases was zero, we simply added 1 to both the observed and the expected number of cases to allow computation of an estimate of the lnRR and its associated SE [18].

Outcome measures
The primary outcome was the overall pooled RR of incidence of SBC among women survivors of HD. The secondary end points were RR vs. various variables: source of data, age at diagnosis of HD, length of follow-up, treatment modalities, and any additional relevant risk factors. RT in this meta-analysis is referred to supra-diaphragmatic irradiation with or without other radiation fields.

Statistical analyses
We assessed heterogeneity of the studies' results by inspecting graphical presentations and by calculating an x 2 test of heterogeneity and the I 2 statistic of inconsistency [19,20]. Statistically significant heterogeneity was defined as a x2 P value less than .1 or an I2 statistic greater than 50%. The estimates of RR, together with associated 95% CI, were obtained using the DerSimonian and Laird random-effects model [21]. Meta-regression analysis was performed to determine to what extent the heterogeneity is explained by various covariates using IBM SPSS statistical package v.19. The dependent variable was the lnRR weighted for the inverse of variance to perform weighted least square linear regression. We first conducted a univariate regression analysis for each variable followed by a multivariate regression including variables found significant in the univariate analysis.
Subgroup analyses were performed to assess potential contributions of various clinical variables to outcomes. A funnel plot estimating the precision of trials (plots of logarithm of the RR against the sample size) was examined for asymmetry to estimate publication bias [22]. Publication bias was also quantified by the regression asymmetry test by Egger [22].
All statistical tests were two-sided. RR was estimated according to the inverse of variance method with the use of Review Manager Software v5.1.4.

Search results
We identified 1,647 potentially relevant articles ( Figure 1). After exclusion of duplicate references, none-relevant literature, and those that did not satisfy inclusion criteria, 41 candidate articles were considered for the meta-analysis . After careful review of the full text of these articles, 7 studies were excluded. In 5 studies the RR was reported based on case-control design, i.e. HL patients who developed SBC against patients who did not [33,43,47,63,64]. The RR in the sixth excluded study compared RR among HL patients according to presence or absence of a family history of BC [50]. The seventh excluded study was designed to examine RT dose and dose distribution in 41 HL patients (25 females) treated at a single Canadian institution [49].
The remaining 34 studies were included in the metaanalysis. The funnel plot of 25 non-overlapping studies showed mild asymmetry, however, the Egger linear regression test was not significant (P = .059).
Of the 34 included studies, there were several reports of overlapping and/or updated data with longer follow-up and more encountered events. For any analysis, only the updated results were used unless there were relevant data available in earlier report and were not included in more recent publication. That approach was used to deal with the overlapping data of van Leeuwen et al. [25] (2 Dutch institutes) and De Bruin et al. [55] (5 Dutch institutes); Mauch et al. [26] (5 USA institutes) and Ng et al. [41] (4 USA institutes); Aisenberg et al. [28] and Alm El-Din et al. [54] (single USA institute); Basu et al. [52] and Constine et al. [53] (5 USA institutes); Neglia et al. [39], Kenney et al. [45], and Castellino et al. [59] (the multiinstitutional USA Childhood Cancer Survivor Study); Metayer et al. [35] (16 population-based cancer registries in North America and Europe, Hodgson et al. [48] (13 population-based cancer registries in North America and Europe), and that of Dores et al. [57] (9 populationbased cancer registries in the USA); Hancock et al. [24], Wolden et al. (1998) [30], O'Brien [58], and Wolden et al. (2000) [37] (single USA institute); and Swerdlow et al. [36], and Swerdlow et al. [62] for the British National Lymphoma Investigation. Tables 1 and 2 show the abstracted and computed data of the included studies. Overlapping studies were shaded together. The data sets included patients diagnosed as early as 1935 to the late 2000s. The median duration of follow-up since the diagnosis of HL was 14.9 years (95% CI, 13.0-16.2 years; ranging from 6 to 23.8 years), while the median age at the diagnosis of HL was 23.7 years (95% CI, 18.9-25.5 years; ranging from 11.8 to 40 years), and the median age at the diagnosis of SBC was 35.0 years (95% CI, 30.2-40.0 years; ranging from 12.8 to 44.4 years). The median interval between primary diagnosis and development of SBC was 17.7 years (95% CI, 15.2-18.6 years; ranging from 12.2 to 21.6 years).
After eliminating overlapping and duplicate studies, there were 25,305 women with HL and 957 incidences of SBC. The median absolute excess rate (AER) of SBC incidence per 10,000 person-years of observation was 22.9 excess cases (95%CI, 15.6-55.7, ranging from 1.1 to 174). Few studies reported the cumulative incidence of SBC after 20 and 30 years of follow-up (average: 5.4% and 12%, respectively). Figure 2 depicts the Forest plot for the pooled RR. The fixed-effects model showed significant heterogeneity (I 2 statistic = 96%; p <0.0001). The random-effects model was computed instead and it showed that patients with HL have an almost 9-fold increase in the risk of SBC (RR = 8.23; [95% CI, 5.43-12.47], I 2 = 96%). We performed meta-regression analysis to determine to what extent the heterogeneity is explained by the effects of study size, age at HL diagnosis, and the latency since the completion of HL treatment. The univariate analysis showed significant inverse association between lnRR and age at diagnosis and a positive relation to latency since HL treatment. With multivariate analysis (Table 3), only younger age at diagnosis remained significant (<0.0001).
Several studies reported the RR at certain cut points of age at diagnosis and it also showed an inverse relation between risk and age ( Figure 5, fixed-effects model). In the latter analysis and contrary to analysis of RR against more age intervals, no significant heterogeneity was noted (I 2 = 41%). Of note, the RR of SBC for women who developed HL above the age of 40 years was not significant (RR = 0.55; [95% CI, 0.09-3.52]).

RR and follow-up latency
We performed analysis of RR by 5-year increments since the treatment of HL ( Figure 6, random-effects model, I 2 = 73%). By and large, the analysis demonstrated an       [55], while all the other studies demonstrated a decreased RR after 25-29 years compared with that after 20-24 years of followup. Due to unreported data, meta-regression analysis of the heterogeneity could only include age at diagnosis of HL as the sole explanatory variable and it showed an inverse association with lnRR (Table 3). After ≥ 20 years of latency since diagnosis, 9 studies [37,40,41,44,45,48,51,54,55] reported RR of 6.95 (95% CI, 4.8-10.1). That RR was not significantly different from the rate encountered after ≥ 30 years of followup (RR = 7.03; [95% CI, 5.2-9.5]) as reported from 5 studies [42,44,51,54,55].

RR vs. Age at HL diagnosis and follow-up latency
To examine the interaction of both age at HL diagnosis and length of follow-up versus risk, few studies have reported adequate data. De Bruin et al. [55] reported that after 5-14 years of follow-up, those who were ≤ 20 years at HL diagnosis had significantly higher risk (RR = 20.0; [95% CI, 7.3-43.4]) as compared with those who were older (21-30 years old) (RR = 5.3; [95% CI, 1.9-16.6]). Similarly, after ≥ 25 years of follow-up, Figure 2 Summary statistics and corresponding forest plot for the overall relative risk (RR) of second breast cancer as reported from 23 studies. RRs were calculated using a random-effects model. Table 3 The results of meta-regression analyses the corresponding RRs for younger and older patients were 14.2 (95% CI, 7.9-25.4), and 9.0 (95% CI, 4.9-16.5), respectively. Figure 7 shows the random-effects model for the RR according to HL treatment modalities. Significant heterogeneity was demonstrated (I 2 = 87%). RT used as the sole therapeutic modality was associated with an almost 5-fold increase in risk (RR = 4.70; [95% CI, 3.28-6.75]), I 2 = 74%) and even higher rate (RR = 14.08; [95% CI, 9.93-19.98]) when RT was used for patients ≤ 30 years of age [32,36,41]. Two studies [24,54], reported on the mantle field RT dose where there was a small difference in RR between dose < 40 Gy and ≥ 40 Gy (5.99, and 6.13, respectively). In the first study [24], 1 patient per 567 person-years risk versus 23 patients per 7876 personyears developed SBC in the lower versus higher RT dose, respectively. In the second study, 17 of 135 versus 18 of 109 patients developed SBC in the lower versus higher RT dose, respectively [54]. When reported, almost all SBC arose within or at the margin of RT field. Figure 7 also shows that adding any CT to RT numerically increased the risk as compared with the risk asso-  To explore the heterogeneity of the RR vs. therapeutic modalities, we performed a univariate analysis that showed an inverse association between lnRR and age at HL diagnosis, positive interaction with latency, and no significant effect for study size. The multivariate meta-regression analysis, however, only showed that younger age at diagnosis retained a significant independent risk (Table 3).

RR and additional contributing factors
Comparing HL patients diagnosed from 1960s to the early 1970s, late 1970s to early 1980s, and more recent years the reported RRs were 3.7, 5.9, and 10.7, respectively [23,28,48,51,55].  Table 4 shows data related to potential contributing factors. Pooled analysis was not attempted due to the small number of studies with sufficient data. Table 4 shows that HL patients who presented with mediastinal mass had higher risk compared with those without mediastinal mass. Table 4 also shows the inconsistency of the reported RR among those who had splenectomy versus those who did no, however, the findings were based on three studies only. Table 4 also shows the potential protective role of pelvic RT as reported by De Bruin et al. [55], where patients receiving that modality showed a risk that was not as high compared with those who did not. HL survivors who received RT had a higher risk of developing estrogen receptor (ER)-negative/progesterone receptor (PR)-negative SBC as compared with ER-positive/PR-positive tumors [57].

Discussion
HL has been a successful model for the development of effective treatment approach in clinical oncology. Long-term survivors of that disease have also allowed better recognition and understanding of the late effects of therapy. In a large cohort of 25,305 women with HL, and with 957 incidences of SBC reported from North American and European institutions, the current meta-analysis quantified the risk of SBC. The overall pooled analysis showed that there was an approximate 9-fold increase in the risk of SBC incidence (pooled RR = 8.23), and AER of 23 patients per 10,000 person-years.
Patients included in this meta-analysis developed HD at a median age of 23.7 years. There has been incongruity about the influence of younger age at HL diagnosis and the higher risk of SBC, where some studies have failed to prove that effect [39,42]. The current meta-analysis clearly showed that younger age at HL diagnosis was associated with increased risk of SBC, and the risk remained after adjusting for other covariates. Moreover, we demonstrated that the RR of SBC for women who developed HL above the age of 40 years was not significant. It is presumed that the higher risk associated with Figure 5 Summary statistics and corresponding forest plot for the relative risk (RR) of second breast cancer based on specific cut points of age at Hodgkin's lymphoma diagnosis. RRs were calculated using a fixed-effects model. Figure 6 Summary statistics and corresponding forest plot for the relative risk (RR) of second breast cancer based on follow-up latency since Hodgkin's lymphoma diagnosis. RRs were calculated using a random-effects model. young age at HD diagnosis is attributed to the effect of RT delivered at a time when breast tissue is proliferating.
RT used as the sole therapeutic modality was associated with a 5-fold (RR = 4.70) increase in risk and a 14-fold (RR = 14.08) increase among young (≤ 30 years of age). Almost all SBC arose within or at the margin of the RT field. The RR of combined RT and any CT was slightly higher than that associated with RT only (5.65 vs. 4.70). An even higher, was the RR associated with the combination of RT and alkylating CT (RR = 6.59), thus, the potential protective effect of gonado-toxic alkylating CT was not demonstrated. Several studies showed an inverse association between the use of alkylating CT in HL and SBC risk [24,65], however, other investigators reported increased risk [30,66]. In this meta-analysis and based on data reported from three studies, the risk associated with the combination of RT and non-alkylating CT was not significant [37,54,58], also found insignificant, was the risk related to the use of CT only.
Analysis of potential additional risk factors was limited due to lack of sufficient data and/or inability to compute missing information, therefore, cautions should be exerted in interpreting results. Two studies [24,54], reported on the mantle field RT dose where there was only a small difference in RR between dose < 40 Gy and ≥ 40 Gy (5.99, and 6.13, respectively). While some studies showed that subjects with SBC were found to be significantly more likely to have received higher doses of mantle RT [52], this observation was complicated by the fact that patients treated with higher radiation doses have had longer follow-up. Guibout et al. [46], did not find a significant association between RT dose and SBC, suggesting that the increased risk after HL may indicate a specific susceptibility for developing SBC, or a particular susceptibility to radiation and/or chemotherapy, or both. Conversely, De Bruin et al. showed that the risk of SBC is related to the RT volume [55], where mantle field irradiation was associated with a 2.7-fold increased risk of SBC compared with mediastinal irradiation alone. Besides, the meta-analysis reported by Franklin et al. showed a RR of 3 comparing extended field versus involved field RT [14].
The reason for failing to show a convincing evidence of RT dose-response effect associated with SBC risk is at best divisive. There is evidence for a strongly linear radiation dose response, but only in the lower dose range (up to 5 or 10 Gy) [67,68]. It has been suggested that cell killing tends to decrease the carcinogenic effect of RT along an exponential curve at doses above 10 Gy [69]. However, it is known that BC is a known complication of low-dose breast radiation [67], thus BC may remain a risk among adolescent women who receive any dose of thoracic irradiation for HL.
Although new RT techniques and treatment strategies have the potential to reduce the future burden of late effects, nevertheless, we have shown that an even higher risk was reported in more recent years suggesting that there remains a significant cohort at an increased risk of SBC.
Pelvic RT was found to be associated with a protective effect as reported by De Bruin et al. [55]. The same effect was also noted by Basu et al., where 3.4% of patients who developed SBC received pelvic RT as compared with 26.3% among those who did not [52]. It is presumed that the protective effect of pelvic RT is attributed to the induction of premature menopause and the role played by hormone stimulation in RT-induced breast carcinogenesis [43,70]. The influence of splenectomy on SBC risk has been controversial. While some studies reported a modest higher risk [71], other studies failed to show that effect [30,52].
Only one study examined the receptor status of SBC [57]. The RR of ER-negative/PR-negative SBC was 66% higher than ER-positive/PR-positive SBC among 5-year HL survivors, and nearly two-fold higher among 15-year survivors. Conversely, other studies of small numbers of SBC patients have not found a significant variation in hormone-receptor status when compared with primary BC controls [72,73]. While the incidence of hormone receptor-positive BC in the general population exceeds that of ER-negative/PR-negative BC, it is postulated, however, that young women treated for HL may experience premature ovarian failure related to HL therapy, Table 4 Relative risk of second breast cancer vs. selected risk variables and therefore, their hormonal BC risk factors may differ from those in the general population.
The present meta-analysis has several limitations. First, it is not possible to completely exclude the possibility that the HL itself carries with it an increased risk of second malignancy including SBC. Second, it is very difficult to quantify the possible effect of confounding factors such as lifestyle factors, personal risk, family history, etc. For example, Landgren et al. found increased RR (1.81) of breast cancer among HL patient with positive (vs. negative) family history of cancer [50]. Third, the analyses showed significant heterogeneity in risk estimate, nevertheless, investigating heterogeneity using meta-regression technique showed the dominant role of age at HL diagnosis. Other limitations include the lack of comprehensive treatment data including information on RT dose and additional treatments, and the lack of sufficient data to model the protective effect of endogenous hormone ablation against the risk associated with exposure to exogenous hormones. Moreover, it is not clear if a similar magnitude of risk is to be expected in a different patient population where the incidence of sporadic BC is low. Finally, the meta-analysis lacks the analysis of SBC outcome. However, SBC incidence rather its mortality was the main objective of the meta-analysis. Moreover, not all studies reported on mortality, besides, analysis of SBC mortality would be confounded by the mortality from HL itself or its therapy-related effect, ascertainment of the cause of death, age of diagnosis of HL or SBC, and length of follow-up.

Conclusions
We conclude that the current meta-analysis provided the most recent comprehensive estimate of the risk of SBC in a broad-range of HL survivors with inclusive analysis of relevant clinical and treatment variables. Based on the derived data where the median age at the diagnosis of SBC was 35.0 years and at a median latency of 17.7 years, screening recommendations for HL survivors need to be reemphasized. The results from the current meta-analysis support the favorable outcome of the risk-guided BC screening for such patients according to three prospective studies [74][75][76]. It is probably more appropriate that female patients with HL who are at a higher risk for developing SBC to be screened annually and at an earlier age rather than biennially starting at the age of 50 years as currently recommended for general population [77]. Our data also support the recent trend of risk-adapted management of HL to reduce the risk of short-and long-term adverse events associated with needless overtreatment [78].