Incremental impact of breast cancer SNP panel on risk classification in a screening population of white and African American women

Introduction

Accurate risk assessment has the potential to decrease morbidity and mortality from breast cancer by facilitating individualized prevention strategies. Genome-wide association studies (GWAS) have identified several single-nucleotide polymorphisms (SNPs) that increase the risk of breast cancer,[1–12] and panels of SNP markers are now commercially marketed as a way to improve breast cancer risk assessment. SNP panel risk estimates can be combined with existing risk models such as the Breast Cancer Risk Assessment Tool (BCRAT, also known as the Gail model), which uses individual risk factors such as age, family history, reproductive history, and history of breast biopsy or atypical hyperplasia to estimate a woman’s absolute risk of breast cancer.[13] The combination of SNP panels with the BCRAT has been shown to at most modestly improve risk prediction.^14–18 SNP panels reclassify some women across risk categories, which may potentially change clinical management.

The BCRAT was originally developed using data from white women, but was subsequently updated and validated for use in African American women. However, the discriminatory accuracy among African American women is lower than among white women.¹⁹ Thus, improving risk prediction among African-American women is a particularly important goal. Although SNP panel markers were identified and validated primarily in white European populations, it is likely that the distribution of risk alleles will vary across populations and lead to differential contributions of SNP risk prediction by race or ethnicity.

In this study we evaluated how the combination of the BCRAT with the 12 SNP panel changed risk stratification in both white and African American women undergoing screening mammography. Although studies attempting to validate these risk variants in women of African descent have yielded mixed results,^20–26 there is insufficient data to estimate race specific effects. Thus, we applied the population level effects for the SNP panel results from published data to both African American and white women to assess the potential reclassification from the current use of SNP panels.

Methods

Results

The descriptive characteristics of the study population are shown in Table 1. Overall, the mean age of study participants was 52 years. Approximately 28% of participants reported a prior biopsy, but only 6 women reported atypical hyperplasia on the biopsy. Fifty-five percent of women were white, 39% were African American, 1% was Hispanic, 1% was Asian, and 5% identified as other race/ethnicity. With the exception of first degree relatives with breast cancer, all baseline characteristics differed significantly between whites and African Americans. The majority of women had mammograms that were BIRADS 1 or 2 (89%). Very few women had results of BIRADS 3 or greater (<1%). Ten percent of women had BIRADS 0 mammograms, requiring additional imaging, and 22 women (3%) had a biopsy. One woman was diagnosed with atypical hyperplasia, three were diagnosed with intraductal carcinoma, and three were diagnosed with invasive breast cancer. Mammogram results did not differ significantly by race.

Each of the 12 SNPs was tested for Hardy-Weinberg Equilibrium, and there were no significant deviations for the total population or by race (data not shown). We compared the frequencies of high-risk alleles for our study population to the deCODE expected frequencies (Table 2). In white women, 3 SNPs had significantly higher than expected frequencies of high-risk alleles (rs13281615, rs3803662, rs4973768). In African American women, seven high-risk alleles had significantly higher than expected frequencies (rs1045485, rs1219648, rs13387042, rs3803662, rs4415084, rs889312, rs999737), and four had significantly lower than expected frequencies (rs11249433, rs3817198, rs4973768, rs6504950).

Table 2.

Expected and Observed Allele frequencies

SNP	Risk Allele	Homozygote Relative Risk	Expected Allele Frequency	Total	White		African American
				N=814	N=449	p-value*	N=316	p-value*
rs1045485 (CASP8)	G	1.03	0.870	0.899	0.872	0.888	0.937	<0.001
rs11249433 (1p11.2)	C	1.18	0.390	0.298	0.412	0.183	0.144	<0.001
rs1219648 (10q26, FGFR2)	G	1.31	0.380	0.402	0.396	0.317	0.421	0.030
rs13281615 (8q24.21)	G	1.10	0.400	0.446	0.462	<0.001	0.418	0.364
rs13387042 (2q35)	A	1.19	0.497	0.592	0.525	0.113	0.703	<0.001
rs3803662 (16q12, TOX3)	T	1.42	0.269	0.402	0.313	0.006	0.535	<0.001
rs3817198 (11p15, LSP1)	C	1.10	0.300	0.258	0.321	0.172	0.161	<0.001
rs4415084 (5p12, FGF10)	T	1.19	0.396	0.487	0.398	0.881	0.604	<0.001
rs4973768 (3p24, SLC4A7)	T	1.12	0.460	0.464	0.532	<0.001	0.369	<0.001
rs6504950 (17q23.2, STXBP4)	G	1.03	0.720	0.717	0.742	0.126	0.671	0.013
rs889312 (5q11.2, MAP3K1)	C	1.19	0.280	0.324	0.305	0.101	0.340	0.001
rs999737 (14q24.1, RAD51B)	C	1.06	0.760	0.853	0.778	0.204	0.957	<0.001

Open in a new tab

^*

p-value from one-sample t-test comparing observed to expected frequency

Table 3 displays the average relative risk from the SNP panel, the average absolute risk from the BCRAT, and average absolute risk from the combined BCRAT model and SNP panel (Combined Model). Based on the SNP panel, the average relative risk of breast cancer in the study population was 1.17, meaning a 17% increased risk based on genetic profile. Using the BCRAT, the average 5-year breast cancer risk was 1.47% and the average lifetime risk was 10.56%. When the BCRAT and SNP panel risk estimates were combined, the average 5-year risk of breast cancer increased to 1.72% and the lifetime risk increased to 12.31%. Risk estimates differed significantly by race (p<0.001). White women had higher average risk estimates from the BCRAT model, while African American women had a higher average relative risk from the SNP panel. In the Combined Model, white women had higher average risk than African American women.

Elevated risk of breast cancer was defined as a 5-year risk greater than or equal to 1.7% or a lifetime risk greater than or equal to 20%. Among white women, the proportion with elevated 5-year risk was similar using the BCRAT and the Combined Model (Table 4, 41.5% vs. 41.3%). Among African American women, the percentage with elevated 5-year risk increased from 12.7% using the BRCAT to 32.9% using the Combined Model. For both white and African American women the lifetime risk estimates were higher using the Combined Model than the BCRAT alone. Six percent of white women had elevated lifetime risk using the BCRAT model, compared with 12.5% using the Combined Model. Only one African American woman was classified as having elevated lifetime risk using the BCRAT, compared with 5.7% using the Combined Model. Twenty-one percent of African American women were reclassified as having elevated 5-year risk of breast cancer with the Combined Model compared to only 10% of white women (p<0.001). There was no difference in the percent of women reclassified as having high lifetime breast cancer risk between African American and white women (5.4% vs. 7.8%, p=0.195)

In the total study population, there was fair agreement between the BCRAT and the Combined Model for 5-year risk (κ=0.543) and poorer agreement for lifetime risk (κ=0.372), and in both cases the Combined Model was more likely to classify individuals as having elevated risk than the BCRAT alone (p<0.001). Among whites there was fair agreement for both 5-year (κ=0.582) and lifetime risk (κ=0.462), and the Combined Model was more likely to classify individuals as having elevated lifetime risk than the BCRAT alone (p<0.001). Among African Americans, there was fair agreement between the models for elevated 5-year risk (κ=0.432) but poor agreement for lifetime risk (κ=0.010). For both 5-year and lifetime risk, the Combined Model was significantly more likely to classify individuals as having elevated risk compared with the BCRAT alone (p<0.001). The results were similar in sensitivity analyses excluding women age 60 and older. We also calculated percent positive agreement, due to concern about small cell sizes²⁷, however the interpretations were similar to those using Kappa.

Discussion

Our results highlight that the addition of SNP information to the BCRAT reclassifies the risk status of some women in a screening mammography population. In particular, a significantly larger percentage of African American women were re-classified as high-risk when the 12 SNP panel was combined with the BCRAT. For 11 of the 12 SNPs in the panel, allelic frequencies for African American women were significantly different than the reported frequencies in white women. These results suggest that SNP panels have the potential to play a more important role in risk prediction among African American than white women and highlight the pressing need for studies of breast cancer SNP panels in minority populations to determine if this reclassification is clinically valid.

Though the BCRAT is widely used to estimate individual breast cancer risk, its discriminatory accuracy is limited, particularly in African American women. When the BCRAT was recalibrated for African Americans, the relative risk estimates for most risk factors were lower for African American than for white women.¹⁹ In other words, the risk factors explained a smaller proportion of the cancer burden in African American women than white women. The average age-specific AUC, or probability that a randomly selected case had a higher predicted risk than a randomly selected control, was 0.555 for African American women, which was lower than the discriminatory accuracy of the original BCRAT model (0.596)¹⁹, and highlights significant room for improvement in breast cancer risk prediction, particularly for African American women.

Our results suggest that the use of SNP panels may increase the proportion of African American women classified as high-risk of breast cancer compared to the BCRAT alone, but future validation is needed to determine whether this re-classification is correct. Validation studies of the 12 SNP panel in African American women are ongoing, but several studies have examined the associations of individual SNPs with breast cancer risk, with mixed results.^{20,22–24,26,28–30} Relative risk estimates for rs13387042 (2q35) in African American women were similar to relative risk estimates used for the SNP panel in two studies.^21,22 An additional two studies found rs1219648 (FGFR2) to be significantly associated with increased breast cancer risk in African American women, with higher relative risk estimates than the SNP panel.^20,22 The association of rs4415084 (5p12, FGF10) with increased breast cancer risk was of borderline significance overall, but strongly associated with ER positive breast cancer in African American women.²⁸ If these higher relative risks represent the true association in African American women, risk stratification for African American women from the SNP panel would be underestimated. The T allele of rs3803662 (16q12, TOX3) was significantly associated with a decreased breast cancer risk in African American women, while this allele has been associated with increased risk in white women²³. If this association were true, then our risk stratification from the 12 SNP panel may be overestimated. Several studies have failed to detect significant associations with rs13387042^20,23,26, rs1219648 ^23,24,26, rs3803662^20,21,26,29, and rs4415084^21,26. The eight remaining SNPs in the panel have not been replicated as significantly associated with breast cancer risk in any studies of African American populations (rs1045485²⁰, rs11249433^21,23,26, rs13281615^20–23,26, rs3817198^20,21,23,26, rs4973768^21,23,26, rs6504950^23,26, rs889312^{20,21,23,24,26}, rs999737^21,26). A recent study examining 19 loci in women of African ancestry failed to validate any high-risk SNPs from previous GWAS.²⁶

Several challenges to genetic susceptibility studies in African American women complicate the validation of the 12 SNP panel. First, given that the 12 SNPs in the panel have modest effect sizes, a large sample size of minority women is needed to have sufficient statistical power to detect associations. Second, the SNPs in the panel may be causal variants or they may be in linkage disequilibrium with causal variants. When SNP analysis is shifted to an African American population, the tagging relationship between the genotyped SNP and the risk variant may be disrupted, as a result of differing patterns of linkage disequilibrium between ancestral populations. Third, disease heterogeneity likely complicates both risk prediction and SNP validation. African American women are more likely than white women to develop triple negative cancers, which are believed to be etiologically different from hormone receptor positive cancers.

Population stratification and admixture are additional challenges when attempting to validate breast cancer risk variants in African Americans. We used self-reported race/ethnicity in our study, which may not sufficiently capture an individual’s ancestral background. For example, the degree of West African versus European ancestry varies across individuals and across subpopulations of African Americans³¹, and such population stratification can bias associations between SNPs and disease outcomes.³² Ancestral informative makers (AIMs) can be used to identify individual ancestry and can be controlled for in studies of disease susceptibility SNPs.³³ Most of the studies referenced above adjusted for AIMs in the analysis. It is unclear whether the inclusion of AIMs improves prediction of breast cancer risk amongst African Americans.

For all of these reasons, future research is needed to clarify the clinical validity and utility of breast cancer associated SNPs in African American women, and SNP panels predictive of breast cancer risk in African American women should be explored.

A strength of our study is the inclusion of large number of African American participants. In addition, we obtained detailed information on breast cancer risk factors allowing us to calculate risk estimates using the BCRAT for each participant. Several limitations of our study should be noted. First, we used a simple multiplication of the BCRAT absolute risk and the total genetic relative risk ratio, which has been shown to be valid for 5-year risk, but may overestimate lifetime risk estimates at high risk levels.{Mealiffe, 2010 #51} Therefore the lifetime risk estimates and proportions of individuals with high lifetime risk may be slightly elevated, and this may partly explain the poorer agreement between lifetime BCRAT and Combined models as compared to 5-year risks. Few African American women were classified as high-risk by either the BCRAT or Combined Models, and therefore our estimates of agreement are based on small numbers. In addition, because this study cohort was recently enrolled, we do not yet have information on long term cancer outcomes and therefore cannot validate the SNP panel. We plan to continue prospective follow-up of this cohort in order to evaluate and develop better risk stratification tools.

Identifying women with elevated risk for breast cancer can facilitate tailored preventive interventions^34–38, and the use of novel genetic markers has the potential to improve such risk stratification. Given the gaps in knowledge about breast cancer associated SNPs among African Americans and the burden of disease in this population, validation and identification of breast cancer associated SNP panels in African Americans should be a priority. This will require large pooled analyses of multiple GWAS studies in order to achieve sample sizes large enough to detect SNPs with moderate to small relative risks.