Bias Explains Most of the Parent-of-Origin Effect on Breast Cancer Risk in BRCA1/2 Mutation Carriers

Information about the family history of cancer is an important predictor for estimating both the breast cancer risk and the probability of being a mutation carrier, for example, a carrier of a BRCA1 or BRCA2 mutation (1–6). It has been suggested that a paternal origin of the BRCA mutation increases the breast cancer risk, whereas it decreases the ovarian cancer risk (7–9). However, it could be that this parent-of-origin effect of the BRCA gene may be caused by biases due to the referral criteria, which are mainly based on the number of affected relatives and their ages at diagnosis (8, 10). For these reasons, some studies have taken referral bias into account by including only BRCA carriers who were unaffected at baseline and by correcting for breast cancer in first-degree relatives (FDR) in combination with environmental factors (8), or by including all carriers but with correcting for the year of birth and year of referral (7). The paternal BRCA1 mutation was both with and without bias correction associated with an breast cancer risk increase but for BRCA2 mutations nonsignificant risk-increasing and risk-decreasing trends have been published (7–9).

It could be that referral bias is only affecting the assessment of the parent-of-origin effect in the index cases and that it does not affect the assessment in the majority of the carriers, because once a mutation is detected in the family, genetic testing becomes available to the rest of the family following a cascade protocol irrespective of the family history (11). In the Netherlands, this means that family members who test positive for a (familial) mutation are provided with a letter to help inform their relatives about hereditary cancer. It has, however, been shown that sharing of information about carrier-ship is more likely to be blocked by male relatives in the family (12–16). Therefore, it might be that women with a paternal origin of the mutation are less often ‘triggered’ to undergo genetic testing until they develop cancer themselves, which causes referral bias (or genetic testing bias).

The aim of this study was to assess the parent-of-origin effect of the BRCA mutation on the breast cancer risk in proven BRCA1/2 mutation carriers and to assess whether this effect still remains when referral biases by family history or personal history of cancer are taken into account. Therefore, we considered it relevant to include only factors related to referral bias and to assess all dimensions of this bias which include the family history of both breast and ovarian cancer in first- and second-degree relatives (SDR) and the personal history of prevalent and/or incident cancer cases.

For this study, data from the Dutch national HEBON study (Hereditary Breast and Ovarian cancer study, the Netherlands) were used. The HEBON study was approved by the medical ethical committees of all participating hospitals, and subjects signed an informed consent form upon participation. From 1999 onwards, subjects who underwent genetic testing for the BRCA1/2 mutation in the course of genetic counselling at any of the Family Cancer Clinics in The Netherlands were invited to participate in the HEBON study. From 2011 onwards, all subjects were invited to fill in a (follow-up) questionnaire on clinical history and risk factors. Thus, the data in this study consist of retrospective data collection with prospective follow-up of carriers' cancer status via linkage with the National Dutch Cancer Registry and questionnaires. The clinics provided data on the date of birth, the cancer status, the mutation status, and the date of the DNA test. Data on the carriers' breast cancer status and ovarian cancer status were also obtained by linkage with the Dutch Cancer Registry with national coverage for the calendar years 1989–2011. Data on cancer status, risk-reducing surgeries, and family history were self-reported by means of the questionnaire.

The current study included females with a pathogenic BRCA1/2 mutation proven by genetic testing who completed the questionnaire. Exclusion criteria were: being the index carrier in the family, that is, the first member who tested positive for the mutation and who was affected at the time of DNA testing (n = 655); no available information on the parental origin of the BRCA mutation (data either missing or reported as unknown; n = 124); carrying a mutation from both parents or carrying both a BRCA1 and BRCA2 mutation (n = 48); or when a risk-reducing mastectomy (RRM) was performed, but the age at RRM was unknown in women without breast cancer (n = 7).

The outcome was defined as the incidence of primary breast cancer. The prevalent cases were defined as all cases both before and after genetic testing—so including (semi-) incident cases. Incident cases were defined as breast cancer cases occurring ≥3 months after a woman's DNA testing date. Semi-incident cases were defined as breast cancer cases occurring after the DNA testing date of the family's index carrier, which means that the semi-incident cases included both cases that occurred both before and after the woman's DNA testing date.

The subjects' breast cancer and ovarian cancer status was either registry-confirmed or self-reported. If a woman reported that a malignancy was detected at the time of the RRM or at risk-reducing salpingo-oophorectomy (RRSO) in a calendar year outside the coverage of the Dutch Cancer Registry, her breast cancer or ovarian cancer status was adapted accordingly.

Referral bias was operationalized by family history or by personal history of cancer. Referral bias by family history was evaluated by taking each of the following factors of this bias into account in separate models: (i) the year of the family's ascertainment (i.e., year of index carriers' genetic testing) because referral criteria have become slightly less stringent over time; (ii) referral criteria, which are based on multiple components of family history; and (iii) separate components of family history for each tumor type, age group, and type of relatedness. The referral criteria (ii) were as assessed as following: ≥1 FDR or SDR with breast cancer by the age of 35 years; ≥1 FDR or SDR with ovarian cancer; ≥1 male FDR or SDR with breast cancer; ≥2 FDRs with breast cancer of whom ≥1 case by the age of 50 years; or ≥3 FDRs or SDRs with breast cancer of whom ≥1 affected FDR by the age of 50 years or ≥1 affected SDR by the age of 60 years. The family history (iii) of breast cancer, ovarian cancer, or another cancer was assessed as follows with and without stratification by gender: having ≥1 parents, ≥1 siblings only, ≥1 siblings or children, ≥1 FDRs or ≥1 SDRs affected with cancer by the age of 40, 50, 60 years or any age. Data on the family history of breast cancer, ovarian cancer, and other cancers were available for both FDRs and SDRs. For some FDRs and SDRs, the family history of cancer for one or more family members was not reported. In those cases, the family history was assumed to be negative.

Referral bias by personal history of cancer was evaluated by considering only incident and semi-incident cases.

Descriptive statistics were used to give an overview of the study population characteristics, and appropriate tests were used to test for differences between mutation carriers with a paternal or maternal origin of the BRCA1/2 mutation.

The effect of the parental origin of the BRCA mutation on the risk of developing breast cancer was estimated using Cox regression survival analyses. In all these analyses, the paternal group was compared with the maternal group (i.e., reference group). Robust SEs were calculated to account for the clustering of carriers within families. The assumption of proportional hazards was tested using Schoenfeld residuals and log-minus-log plots. Censoring was applied at the first moment any of the following events occurred: breast cancer, ovarian cancer, RRM, RRSO, or last date of information.

To assess whether the effect of parental origin of the BRCA mutation was independent of referral bias due to a family history of cancer, Cox regression analyses of prevalent cases were adjusted for each factor separately. Factors that altered the parent-of-origin effect by 5% or more were considered to be relevant and are reported in Results. We also took the second-degree family history into account to evaluate the potential referral bias. Because this information was not available for the complete cohort, we used a sub cohort of 742 (94%) BRCA1 carriers and 488 (93%) BRCA2 carriers.

To assess whether the parental-origin effect of the BRCA mutation was affected by referral bias due to a personal history of cancer, Cox regression analyses were performed on semi-incident and incident cases. For the semi-incident case analyses, follow-up time and cases were counted from the date of the family index carrier's DNA test. For the incident case analyses, follow-up time and cases (≥3 months after DNA test) were counted from the woman's date of DNA test onward. In these analyses, only mutation carriers who did not have breast or ovarian cancer, RRM, or RRSO before the start of the follow-up period were included as these were reasons for censoring.

Sensitivity analyses were performed for the unadjusted analyses by changing the outcome event to (i) prevalent registry-confirmed cases or (ii) all self-reported and registry-confirmed prevalent cases.

In addition, RRSO was included as a time-dependent covariate and no censoring was applied at this event. Other sensitivity analyses were performed only for the prevalent case analyses adjusted for family history of cancer. First, the analyses were additionally adjusted for the number of reported family members. Second, carriers with missing information for the family history factors were excluded. Third, analyses were performed with the family history as documented at the time of personal genetic testing because this might be the more optimal scenario to assess referral bias.

All analyses were performed with stratification by the BRCA1/2 gene. The analyses were performed using R, and statistical significance was defined as P < 0.05 (17).

In total, 1,314 BRCA mutation carriers were included in this study, of whom 788 (60%) harbored a mutation in the BRCA1 gene and 526 (40%) in the BRCA2 gene. Of these, 236 (33%) of the BRCA1 carriers and 120 (23%) of the BRCA2 carriers developed breast cancer. For both BRCA1 and BRCA2 carriers, the percentage of ovarian cancer was relatively low, 2% to 4%, most likely due to the high rate of RRSO (∼65%). Of all carriers with information on the date of genetic testing [n = 968 (74%)], the testing was performed presymptomatically in 493 (89%) of the BRCA1 carriers and in 370 (89%) of the BRCA2 carriers (Table 1).

BRCA1 carriers had more FDRs affected with breast cancer diagnosed before the age of 50 years (38% vs. 31%, P = 0.010) or ovarian cancer at any age (20% vs. 11%, P < 0.001) compared with BRCA2 carriers but had fewer male FDRs with breast cancer (0.9% vs. 3.6%, P < 0.001) which is in line with known differences in cancer penetrance between BRCA1 and BRCA2 families (18–21).

The BRCA mutation was of paternal origin in 552 (42%) women: 330 (42%, P < 0.001) BRCA1 and 222 (42%, P < 0.001) BRCA2 carriers (Table 1). For both BRCA1 and BRCA2 carriers, women with a paternal origin of the BRCA mutation differed significantly from the women with a maternal origin in the following ways: they were more likely to have breast cancer themselves (BRCA1_pat 42% vs. BRCA1_mat 26%; BRCA2_pat 28% vs. BRCA2_mat 19%); less likely to have FDRs affected with breast or ovarian cancer (BRCA1_pat 37% vs. BRCA1_mat 82%; BRCA2_pat 37% vs. BRCA2_mat 74%); less likely to have a mother affected with any cancer (BRCA1_pat 17% vs. BRCA1_mat 83%; BRCA2_pat 17% vs. BRCA2_mat 73%); and more likely to have a father with any cancer (BRCA1_pat 34% vs. BRCA1_mat 20%; BRCA2_pat 44% vs. BRCA2_mat 30%). While breast cancer was the most common cancer among all mothers of mutation carriers, only a small fraction of all fathers was affected with breast cancer (Table 2).

BRCA1/2 mutation carriers with a paternal origin of the mutation were more often affected with cancer at the time of the index carrier's DNA test (BRCA1_pat 20% vs. BRCA1_mat 7%; BRCA2_pat 10% vs. BRCA2_mat 7%), and by the time of their personal DNA test, BRCA1/2 mutation carriers with a paternal origin of the mutation were even more often affected (BRCA1_pat 25% vs. BRCA1_mat 13%; BRCA2_pat 13% vs. BRCA2_mat 13%). BRCA1/2 mutation carriers with a paternal mutation who were unaffected at the time of DNA testing were more likely to have affected siblings, children, or SDRs at this time compared with unaffected mutation carriers with a maternal origin of the mutation (Supplementary Table S1).

A woman's breast cancer risk was significantly increased when the BRCA1 mutation was of paternal origin [HR, 1.54; 95% confidence interval (CI), 1.19–2.00]. For BRCA2 carriers, this increase was of the same order of magnitude but not statistically significant (HR, 1.40; 95% CI, 0.95–2.06; Table 3).

When taking bias by family history into account in separate models, the risk increase associated with the paternal origin of the BRCA mutation varied from 1.41 to 1.83 in BRCA1 carriers and 1.27 to 1.62 in BRCA2 carriers (Fig. 1).

For BRCA1 carriers, the effect of the paternal origin remained significantly increased irrespective of the adjustment. For BRCA2 carriers, the effect of the paternal origin was only significantly increased when adjusted for maternal breast cancer up to the age of 60 years or for having a sibling with breast cancer. For both BRCA1 and BRCA2 carriers, adjustment for having FDRs with ovarian cancer mitigated the effect of paternal origin, whereas having FDRs with breast cancer increased the effect. Overall, adjustment for the second-degree family history had a similar but weaker impact on the paternal origin effect of the BRCA mutation.

The semi-incident case analyses included 46 BRCA1-related and 36 BRCA2-related breast cancers. The HRs of the parent-of-origin effects were insignificant: 1.02 (95% CI, 0.56–1.88) for BRCA1 carriers and 0.94 (95% CI, 0.48–1.85) for BRCA2 carriers (Table 3). The incident case analyses included 16 BRCA1-related and 15 BRCA2-related breast cancers. The HRs of the parent-of-origin effect were 0.77 (95% CI, 0.25–1.71) for BRCA1 carriers and 1.14 (95% CI, 0.24–3.15) for BRCA2 carriers, both being insignificant.

In the sensitivity analyses restricted to only registry-confirmed prevalent cases, as compared with the prevalent case analyses, the unadjusted effect of the paternal origin was 1.52 (−3%) in BRCA1 carriers and 1.44 (+2%) BRCA2 carriers (Table 3). For the analyses including both registry-confirmed and self-reported prevalent cases, these numbers were 1.56 (−1%) in BRCA1 carriers and 1.28 (−8%) in BRCA2 carriers.

Subsequently, the analyses were adjusted for family history but in addition also included the reported number of affected and unaffected FDRs and SDRs. In addition to our already reported factors, adjustment for having any FDRs with breast cancer or having any (male) FDRs with any other cancer (not breast or ovarian) increased the effect of paternal origin of BRCA1 by 5% or more and adjustment for having any female FDRs with any other cancer by the age of 60 years increased the effect of paternal origin of BRCA2 by 5% or more.

When women with missing data were excluded, the impact of family factors on the paternal origin effect was slightly stronger than in the main analyses. For BRCA2 carriers, in addition to the already reported factors, adjustment for having any male FDRs with any cancer other than breast cancer reduced the paternal origin effect by 5%.

In our analyses, we used family history at time of the questionnaire. We wondered whether the family history at time of genetic testing would be more optimal for assessing referral bias because this is the family history that contributed to a person's ascertainment. Therefore, these analyses were performed on a sub-cohort of 550 (70%) BRCA1 carriers and 416 (79%) BRCA2 carriers for whom this data were available; however, no differences with the other prevalent case analyses were observed (data not shown).

The results of the prevalent case analyses showed that paternal inheritance of a BRCA mutation increases the breast cancer risk for both BRCA1 carriers (HR, 1.54; 95% CI, 1.19–2.00) and BRCA2 carriers (HR, 1.40, 95% CI, 0.95–2.06). This risk increase was present, irrespective of the referral criteria because the HRs adjusted separately for several factors of family history ranged from 1.41 to 1.83 in BRCA1 carriers and 1.27 to 1.62 in BRCA2 carriers, and they were associated with a similar level of significance as the unadjusted effect. However, no parent-of-origin effect was observed when referral bias by personal history of cancer was taken into account. These HRs were 0.66 (95% CI, 0.25–1.71) for BRCA1 and 1.14 (95% CI, 0.42–3.15) for BRCA2.

An increase in breast cancer risk in case of a paternal origin of the mutation has been reported for BRCA1, whereas for BRCA2 mutations, nonsignificant risk-increasing as well as risk-decreasing trends have been published (7–9). For both genes, we observed a risk increase in case of paternal transmission of the mutation, even when the effect was adjusted for possible referral bias by family history. This is in line with the results of a previous study in which an increased risk was observed in case of a paternal mutation when the analyses were adjusted for the year of birth, year of referral, and for oral contraceptive use (7). They observed a similar risk increase in the unadjusted and adjusted analyses [BRCA1: HR = 1.50 (P = 0.02) and HR = 1.53 (P = 0.02), respectively; BRCA2: HR = 1.23 (P = 0.37) and HR = 1.21 (P = 0.41), respectively]. Another study adjusted all analyses for referral bias by personal cancer history and included only women who were unaffected at baseline. In addition, the analyses were adjusted for referral bias by family history—the number of FDRs with breast cancer, and breast cancer in the mother—breast feeding, age at menarche, and country of residence (8). For BRCA1, they reported also an increased unadjusted and adjusted risk in case of paternal transmission of the mutation [HR = 1.46 (P = 0.06) and HR = 1.36 (P = 0.19), respectively]. For BRCA2 carriers, they observed a risk decreasing trend in both the unadjusted and adjusted analyses in case of paternal transmission [HR = 0.81 (P = 0.65) and HR = 0.88 (P = 0.82), respectively]. These findings are similar to our results from the incident cases analyses with RRSO as a time-dependent variable (BRCA1: HR = 1.10, BRCA2: HR = 0.90). However, when the RRSO was incorporated as a censoring event, the risk ratios changed from risk increasing to decreasing and vice versa (BRCA1: HR = 0.66, BRCA2: HR = 1.14). Although none of these parent-of-origin effects were statistically significant.

Our study is the first to adjust the parent-of-origin analyses for referral bias by incorporating a family history of cancer in both male and female FDRs and SDRs and by including a history of cancer other than breast cancer. We show that having a family history of ovarian cancer in any FDRs and/or SDRs mitigates the paternal origin effect, whereas a family history of breast cancer increases this effect. However, we also addressed the impact of referral bias by a personal history of cancer. More mutation carriers with a paternally derived mutation were affected with breast cancer at the time of ascertainment of the family and at the time of their personal DNA test, as compared with mutation carriers with a maternally derived mutation, especially for BRCA1, which might suggest referral bias and genetic testing bias. When this was taken into account in the semi-incident and incident case analyses, no parent-of-origin effect could be observed anymore, although numbers were small and thus statistically insignificant.

Sensitivity analyses showed similar results for the unadjusted prevalent and (semi-)incident case analyses using only registry-confirmed cases and/or all self-reported breast cancer cases. Moreover, in the adjusted analyses, similar results were found when missing family history was considered as missing instead of a negative family history.

For this study, we used a well-structured national cohort of BRCA1/2 mutation carriers, and breast cancers were confirmed by linkage with the national cancer registry whenever possible. Only the impact of the parent-of-origin on breast cancer was assessed because the number of ovarian cancer cases was too small for risk assessment. A minor limitation is that the family history and risk-reducing surgeries were self-reported, but studies have shown that individuals report their family history of breast cancer quite accurately for FDRs and fairly accurately for SDRs (22–24). The self-reported family history of ovarian cancer is less accurate, especially in SDRs. However, results adjusted for SDR family history were in line with what was expected on the basis of the results of the FDR history. Information on the parental origin of the BRCA mutation and on second-degree family history was not available for the complete cohort. However, comparison of the selected study population with the complete cohort showed that our study population was a representative sample including a somewhat younger cohort with more risk-reducing surgeries and fewer cancers.

Both retrospective and prospective cohort studies can be used to assess risk factors, but biases should be addressed carefully in either study design, as there will always be some form of selection of the study population. In this study, the referral bias by family history was assessed by excluding the index cases and accounting for possible confounding by family history. However, our risk factor of interest, parent-of-origin, seems to be related to referral bias due to a personal history of cancer, and this could not be disentangled with retrospective analyses alone. The referral bias due to a personal history of cancer could only be addressed prospectively without any further adjustment for family history due to small numbers.

Another selection bias that could have affected the observed parent-of-origin effect is a possible difference in genetic fitness between paternal and maternal mutation carriers (25). This might imply that a female carrier with a more severe genetic make-up will have an earlier disease onset, a worse prognosis, and be therefore less reproductive. However, we found no difference in the number of siblings between paternal and maternal mutation carriers.

Although the results from our epidemiologic study make the parent-of-origin effect seem less likely, research on the possible biologic mechanisms underlying this effect, like difference in maternal and paternal imprinting of the BRCA genes and their modifier genes, may help resolve the issue (7, 8).

In conclusion, the existence of a parent-of-origin effect depends on how researchers correct for referral bias. Correction of referral bias as defined by family history did not substantially impact this effect, whereas bias correction for the personal cancer history made the parent-of-origin effect disappear. This bias, when uncorrected for, may have produced the positive association between paternal origin of the BRCA1/2 mutation and an increased risk of breast cancer reported in earlier studies. As the prospective cohort was relatively small, a larger prospective cohort study should address the combined impact of referral bias by family and personal history. Currently, there is no evidence for incorporating the parent-of-origin effect in risk prediction models for breast cancer in BRCA1/2 mutation carriers.

No potential conflicts of interest were disclosed.

Conception and design: J.R. Vos, J.C. Oosterwijk, M.A. Rookus, M.J. Mourits, G.H. de Bock

Development of methodology: J.R. Vos, J.C. Oosterwijk, M.A. Rookus, M.J. Mourits, G.H. de Bock

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.C. Oosterwijk, C.M. Aalfs, M.A. Rookus, M.A. Adank, A.H. van der Hout, C.J. van Asperen, E.B. Gómez-Garcia, A.R. Mensenkamp, M.G.E.M. Ausems, A. Jager

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.R. Vos, J.C. Oosterwijk, M.A. Rookus, M.J. Mourits, G.H. de Bock

Writing, review, and/or revision of the manuscript: J.R. Vos, J.C. Oosterwijk, C.M. Aalfs, M.A. Rookus, M.A. Adank, C.J. van Asperen, E.B. Gómez-Garcia, A.R. Mensenkamp, A. Jager, M.G.E.M. Ausems, M.J. Mourits, G.H. de Bock

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.R. Vos, M.A. Rookus, A.H. van der Hout

Study supervision: J.R. Vos, J.C. Oosterwijk, M.J. Mourits, G.H. de Bock

The Hereditary Breast and Ovarian Cancer Research Group Netherlands (HEBON) consists of the following Collaborating Centers: Coordinating center - Netherlands Cancer Institute, Amsterdam, the Netherlands: M.A. Rookus, F.B.L. Hogervorst, F.E. van Leeuwen, S. Verhoef, M.K. Schmidt, N.S. Russell, J.L. de Lange, R. Wijnands; Erasmus Medical Center, Rotterdam, the Netherlands: J.M. Collée, A.M.W. van den Ouweland, M.J. Hooning, C. Seynaeve, C.H.M. van Deurzen, I.M. Obdeijn; Leiden University Medical Center, the Netherlands: C.J. van Asperen, J.T. Wijnen, R.A.E.M. Tollenaar, P. Devilee, T.C.T.E.F. van Cronenburg; Radboud University Nijmegen Medical Center, the Netherlands: C.M. Kets, A.R. Mensenkamp; University Medical Center Utrecht, the Netherlands: M.G.E.M. Ausems, R.B. van der Luijt, C.C. van der Pol; Amsterdam Medical Center, the Netherlands: C.M. Aalfs, T.A.M. van Os; VU University Medical Center, Amsterdam, the Netherlands: J.J.P. Gille, Q. Waisfisz, H.E.J. Meijers-Heijboer; University Hospital Maastricht, the Netherlands: E.B. Gómez-Garcia, M.J. Blok; University Medical Center Groningen, the Netherlands: J.C. Oosterwijk, A.H. van der Hout, M.J. Mourits, G.H. de Bock; The Netherlands Foundation for the Detection of Hereditary Tumours, Leiden, the Netherlands: H.F. Vasen; The Netherlands Comprehensive Cancer Organization (IKNL): S. Siesling, J.Verloop; The nationwide network and registry of histo- and cytopathology in the Netherlands (PALGA): L.I.H. Overbeek. HEBON thanks the registration teams of IKNL and PALGA for part of the data collection. The authors thank Kate Mc Intyre for editorial assistance.

The HEBON study is supported by Dutch Cancer Society grants NKI1998-1854, NKI2004-3088, NKI2007-3756, Netherlands Organization of Scientific Research grant NWO 91109024, Pink Ribbon grants 110005 and 2014-187.WO76, BBMRI grant NWO 184.021.007/CP46, and Transcan grant JTC 2012 Cancer 12-054.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.