To the editor: The suspected association between breast cancer risk and MMR genes MSH6 and PMS2 has been a frequent topic of discussion in the clinical cancer genetic field. In our cancer genetic counseling and testing clinic, we frequently observe that MSH6/PMS2 gene mutation carriers are affected with breast cancer or have a very significant family history of breast cancer. The article published by Roberts et al. detailed the findings in more than 50,000 women who underwent germline genetic testing through the GeneDx laboratory for the purpose of cancer predisposition assessment.1 The study reached the conclusion that MSH6 and PMS2 mutation is associated with 2–3 fold increased breast cancer occurrence. However, the method of statistical analysis on the standardized incidence ratio (SIR) calculation seems to be flawed. In the field of clinical cancer genetic testing, the population of the tested is significantly skewed as the majority of candidates are women who were affected with breast cancer or concerned about breast cancer in the family, and thus belong to a high-risk group. The cohort Roberts studied has already been significantly enriched with breast cancer and thus brought with it a significant ascertainment bias. The SIR analysis utilized in this study compared the breast cancer incidence in these mutation carriers with the population-based incidence in Surveillance, Epidemiology, and End Results (SEER) data. As reported by Roberts et al., the cohort of this study includes 25.3% cases with a personal history of breast cancer and the average age of the first breast cancer diagnosis was 50.2 years. While in the general population, lifetime risk is 12.4% and the average age of diagnosis is 60 years. Even without a lifetime risk assessment, the cohort’s incidence rate for breast cancer has already doubled that of the general population. If a lifetime risk is computed, it could be even higher. In the Introduction and Discussion sections of the article, the authors did consider the ascertainment bias but failed to recognize that the strong bias effect may render the conclusion invalid. In this cohort, based on random association by chance, we would expect to see the breast cancer cases reach 25% in MSH6 mutation or PMS2 mutation carriers. Alternatively, a proper analysis for such subjects can be done by comparing the breast cancer cases in MSH6 or PMS2 mutation carriers and non–mutation carriers within the same cohort. In statistical terms, to assess the question of whether the relationship between two variables is independent or associated, i.e., to compare proportions of a categorical outcome (e.g., breast cancer versus no breast cancer) according to different independent groups (e.g., PMS2/MSH6 mutation carrier or noncarrier), chi-squared test or Fisher’s exact test can be used. The chi-squared test applies to a large sample and Fisher’s exact test to small-sized samples. An alternative analytical method regarding mutation and cancer association was described by Couch et al.2 In this article, the observed frequency of pathogenic variants within the gene was compared between patients with and without breast cancer, i.e., ExAC-NFE non-TCGA general population reference controls. The frequency should be statistically higher in the breast cancer group if the variant is associated with breast cancer. This study found slightly less than two folds of MSH6 mutations in the breast cancer group (odds ratio [OR] 95% confidence interval [CI] 1.93 [1.16–3.27], p = 0.01). It could not find evidence to support association with PMS2 mutation (OR [95%CI] 0.82 [0.44–1.47], p = 0.56).

In their response to Roberts et al., ten Broeke et al.3 also delineated the possible reasons for a misguided conclusion of association between mutations and breast cancer. They raised the question of whether “the breast cancer may truly be caused by MSH6/PMS2 variant due to its high prevalence in the population.” We believe a high population prevalence will not result in higher association between the variants and breast cancer if the SIR calculation is conducted between comparable cohorts/populations. We agree with ten Broeke et al.3 that when Roberts et al. attempted to include other cancers in addition to Lynch syndrome families to counteract the ascertainment bias, it actually resulted in overestimation of the breast cancer incidence in this cohort.

At the Moffitt Cancer Center, we have also explored the breast cancer risk with MSH6/PMS2 mutations. We have analyzed a similar but smaller cohort of individuals who underwent cancer predisposition genetic testing. The majority of the patients in our center are females affected with breast cancer or with a family history of breast cancer. We compared the cases with PMS2 and MSH6 mutations and without mutations, and found no significant increase of breast cancer. Among 1656 consecutive cases who underwent panel genetic testing, 90% were females, 10% were males, 882 (53.3%) had a history of breast cancer, 64 (3.9%) had colorectal cancer, 41 (2.5%) had uterine or endometrial cancer, and 109 (6.6%) had ovarian, peritoneal, or fallopian tube cancer. PMS2 mutations were discovered in 18 cases (1.1%), MSH6 in 13 (0.8%), MSH2 in 6 (0.4%), and MLH1 in 3 (0.2%). Breast cancers were found in 4/18 (22.2%) PMS2 mutation carriers, 4/13 (30.7%) MSH6 mutation carriers, 1/6 (16.7%) MSH2 mutation carriers, and 0/3 (0%) MLH1 mutation carriers. If PMS2 mutation is associated with breast cancer, the breast cancer should have occurred in more than 53.3% (n = 9) of the PMS2 mutation carriers, or more than 53.3% (n = 7) of MSH6 mutation carriers. On the contrary, using Fisher’s exact analysis on the Moffitt cohort, PMS2 mutation is statistically significant for being less associated with breast cancer (OR 0.247 [95% CI 0.059–0.792], p = 0.0084), while the association between breast cancer and MSH6 mutation carriers appears to be inconclusive (OR 0.387 [95% CI 0.087–1.400], p = 0.16).

To summarize, when studying the cause–effect relationship of germline gene mutations and cancer, the conclusions can be very different based on different cohorts or methods of analysis. Only the answers derived from the population-based unselected cohort may be appropriate to direct cancer risk assessment in the general population.