|Home | About | Journals | Submit | Contact Us | Français|
Smith et al1 present data showing high rates of breast cancer in first‐degree relatives (FDRs) of BRCA1 or BRCA2 mutation carriers who do not share their relative's mutation. The authors conclude that in high‐risk families, women who test negative for the familial mutation have an increased risk of breast cancer consistent with genetic modifiers and should be offered surveillance. Here, we show by simulation that the observed increased incidence in non‐carriers is no greater than would be expected given the study design, even assuming that such individuals are only at population risk. It is the selection of families with multiple affected individuals for referral to the genetics clinics and the inclusion of cases ascertained before referral that causes this artefact.
The standardised incidence ratio (SIR) in FDRs who tested negative in Smith's study was 5.0 (95% CI 2.9 to 7.8). After adjustment to take account of the fact that women with breast cancer were more likely to have genetic testing than were unaffected family members, the estimated ratio was 3.2 (95% CI 2.0 to 4.9). Looking only at cases diagnosed after mutation testing, the SIR was non‐significantly increased: 2.1 (95% CI 0.4 to 6.2, based on just three cases).
The spurious increase in risk caused by selection of families can be illustrated by an extreme example. Consider a study in which genetic testing is offered to pairs of sisters, both of whom have breast cancer. Suppose there were 100 affected non‐carrier sisters of carriers, and that, based on breast cancer rates in the general population, the expected number of cases in the 100 non‐carriers was 2.5; no one would seriously consider presenting the SIR, but in this example it is 40. (SIR is defined as the observed number of cases divided by the expected number). Suppose we expand the study to include all sisters, whether or not they are affected, as long as there are at least two affected sisters in the family: we might then have 200 non‐carrier sisters of carriers, 100 of whom had breast cancer. Assuming the same mean expected incidence of breast cancer, there would have been 5.0 expected cases, yielding an SIR of 20 (ie, 100/5.0).
To estimate the magnitude of the bias (expressed as an SIR) caused by realistic levels of selection, we simulated data that had features similar to those reported by Smith et al.1 In that study, there were 972 FDRs of 277 index cases (mean 3.51) and 3.6 cases expected in 184 FDRs who underwent mutation testing (2.0% were expected to have had breast cancer). We simulated one million families with an index case who was a carrier and between one and six (female) FDRs: 1/12th of the families had one FDR, 2/12th had two FDRs, 3/12th had three FDRs, 3/12th had four FDR, 2/12th had five FDRs and 1/12th had six FDRs. Thus, the mean number of simulated FDR per case was 3.5. Each FDR had a 50% chance of sharing the index case's mutation. Once the mutation status was randomly assigned, each simulated FDR was randomly labelled as affected (with breast cancer) or (as yet) unaffected. Non‐carriers had a 2% risk of being affected; carriers had a 25% risk.
Families with between one and four FDRs were selected (for referral to a genetic clinic) if at least one simulated FDR had breast cancer. Families with five or six FDRs had to have at least two affected individuals to be included. Larger families were required to have more affected individuals for selection to reflect the reality that larger families typically include older individuals, and a family with cancer in older relatives would require more affected cases to be referred for genetic counselling. This led to 29.1% of families being selected. In the selected families, 37.4% of FDRs had breast cancer (which is similar to the proportion (33.8%=329/972) observed by Smith) and 7.24% of the mutation‐negative relatives had breast cancer. If we assume that 0.25% of the population are carriers,2 so that the general population risk is 2.06% (ie, 99.75%×2.0+0.25%×25), then the SIR in the simulation study was 3.51(ie, 7.24/2.06).
Under realistic assumptions about family size and rates of breast cancer in carriers and non‐carriers, we have estimated that breast cancer would be 3.5 times more common in non‐carrier family members than in unselected women. The SIR of 3.5 based on retrospective cases should be contrasted with the prospective SIR of 0.97 (ie, 2.00/2.06).
These simulations show that selection bias alone is enough to explain the results presented by Smith et al,1 without having to assume that unaffected relatives of gene carriers are at increased risk of breast cancer. We are inclined to believe that there remains a small increase in risk because of shared environmental factors and shared genetic modifiers, but would suggest that the effect is modest (perhaps a 1.25‐fold increase in risk) and does not warrant special surveillance being offered to non‐carriers who are the FDRs of a known carrier.
The issue of phenocopies in families with breast cancer is of real clinical importance and deserves to be studied properly through prospective follow‐up. It is hoped that the genetics community will work together to prospectively document the incidence of breast cancer in non‐carriers who are disease free at testing. In the absence of such prospective data, and taking into account the substantial biases inherent in retrospective studies of cancer in families known to genetics clinics, we see no reason to modify the National Institute for Health and Clinical Excellence guidelines: one should reassure individuals who test negative that their breast cancer risk is not increased.3
Competing interests: None declared.