Abstract
Background
Most of the genetic variants identified from genome-wide association studies of breast cancer have not been validated in Asian women. No risk assessment model that incorporates both genetic and clinical predictors is currently available to predict breast cancer risk in this population.
Methods
We analyzed 12 single-nucleotide polymorphisms (SNPs) identified in recent genome-wide association studies mostly of women of European ancestry as being associated with the risk of breast cancer in 3039 case patients and 3082 control subjects who participated in the Shanghai Breast Cancer Study. All participants were interviewed in person to obtain information regarding known and suspected risk factors for breast cancer. The c statistic, a measure of discrimination ability with a value ranging from 0.5 (random classification) to 1.0 (perfect classification), was estimated to evaluate the contribution of genetic and established clinical predictors of breast cancer to a newly established risk assessment model for Chinese women. Clinical predictors included in the model were age at menarche, age at first live birth, waist-to-hip ratio, family history of breast cancer, and a previous diagnosis of benign breast disease. The utility of the models in risk stratification was evaluated by estimating the proportion of breast cancer patients in the general population that could be accounted for above a given risk threshold as predicted by the models. All statistical tests were two-sided.
Results
Eight SNPs (rs2046210, rs1219648, rs3817198, rs8051542, rs3803662, rs889312, rs10941679, and rs13281615), each of which reflected a genetically independent locus, were found to be associated with the risk of breast cancer. A dose–response association was observed between the risk of breast cancer and the genetic risk score, which is an aggregate measure of the effect of these eight SNPs (odds ratio for women in the highest quintile of genetic risk score vs those in the lowest = 1.85, 95% confidence interval = 1.58 to 2.18, Ptrend = 2.5 × 10−15). The genetic risk score, the waist-to-hip ratio, and a previous diagnosis of benign breast disease were the top three predictors of the risk of breast cancer, each contributing statistically significantly (P < .001) to the full risk assessment model. The model, with a c statistic of 0.6295 after adjustment for overfitting, showed promise for stratifying women into different risk groups; women in the top 30% risk group accounted for nearly 50% of the breast cancers diagnosed in the general population.
Conclusion
A risk assessment model that includes both genetic markers and clinical predictors may be useful to classify Asian women into relevant risk groups for cost-efficient screening and other prevention programs.
CONTEXT AND CAVEATS
Prior knowledge
Most of the genetic variants identified in genome-wide association studies of breast cancer conducted primarily among women of European ancestry have not been validated in Asian women. Consequently, no risk assessment model that incorporates both genetic and clinical predictors is currently available to predict breast cancer risk in this population.
Study design
Case–control study evaluating associations between the 12 single-nucleotide polymorphisms identified as risk variants and the risk of breast cancer among Chinese women participating in the Shanghai Breast Cancer Study as well as the cumulative risk of breast cancer associated with combinations of these risk variants. A risk assessment model that incorporates both newly identified genetic variants and traditional risk factors was developed, and its performance in risk prediction was evaluated.
Contribution
Eight of the 12 single-nucleotide polymorphisms were also associated with the risk of breast cancer among Chinese women. An aggregate measure of the combined effect of multiple genetic risk variants had moderate discriminatory accuracy by itself but in combination with established risk factors for breast cancer showed promise for stratifying women into high- vs low-risk groups.
Implications
A risk assessment model that includes both genetic markers and clinical predictors may be useful to classify Asian women into relevant risk groups for cost-efficient screening and other prevention programs.
Limitations
The moderate discriminatory accuracy provided by the full risk assessment model established in this study is inadequate for cancer diagnosis and screening. The absolute risk estimates provided by the model would be applicable only to populations with rates comparable to those of Shanghai.
From the Editors
Breast cancer is one of the most common malignancies among women worldwide; it is a complex polygenic disorder for which genetic factors play an important role in disease etiology (1,2). High-penetrance breast cancer susceptibility genes, such as BRCA1 and BRCA2, explain only a small fraction of breast cancers in the general population because of their low mutation rates (3). Since 2007, several genome-wide association studies of breast cancer (4–9), including one among Chinese women (9), have identified a number of genetic susceptibility loci that are associated with the risk of breast cancer. Most of these genome-wide association studies, however, were conducted among women of European ancestry, and many of the genetic risk factors identified to date have not been adequately evaluated in other populations.
Each of the genetic factors identified thus far is associated with only a small to moderate increased risk of breast cancer, and thus, it is unlikely that any single variant will have much meaningful utility for risk prediction. It is possible that a combination of variants may confer substantially increased risk of breast cancer and that such variants in combination with established (ie, traditional) risk factors could potentially be useful for identifying women at high risk of breast cancer who should be targeted for cost-efficient prevention strategies. However, to our knowledge, no study has quantified the risk of breast cancer associated with combinations of newly identified genetic risk variants and traditional risk factors.
The recent identification of multiple common genetic risk variants for breast cancer has raised the hope that these genetic markers could be used to improve the discriminatory accuracy of risk prediction models for this malignancy (10). Although the improvement in risk prediction at the individual level has been shown to be moderate as measured using the c statistic (10), the utility of prediction models that include both genetic and traditional risk factors at the population level to classify women into relevant risk groups for cost-efficient cancer prevention measures has not been adequately quantified (11). Multiple breast cancer risk prediction models have been established for American women (12–19), including one (18) that was evaluated in a small group of Asian Americans. To our knowledge, no risk prediction model exists for women who live in Asia, a population that has experienced a rapid increase in breast cancer incidence over the past few decades (20,21). Because the risk of breast cancer among women who live in most Asian countries remains much lower than that seen among women who live in North America and Western Europe, it is not cost-efficient, and may in fact be financially prohibitive in some Asian countries, to carry out a population-based breast screening program similar to programs that have been implemented in developed countries. Therefore, establishing a breast cancer risk assessment model for Asian women who live in Asia could have substantial public health implications for breast cancer prevention, particularly because more than 60% of the world’s women live in Asia.
In this study, we systematically evaluated associations between the 12 single-nucleotide polymorphisms (SNPs) identified as risk variants in recent genome-wide association studies and the risk of breast cancer among Chinese women. We also investigated the cumulative risk of breast cancer associated with combinations of these risk variants. Furthermore, we established a risk assessment model that incorporates both newly identified genetic variants and traditional risk factors and evaluated the performance of this model in risk prediction at both the individual and the population levels.
Subjects and Methods
Study Population
The study subjects were participants in the Shanghai Breast Cancer Study (SBCS), a population-based case–control study that was conducted among Chinese women living in urban Shanghai, China. Detailed descriptions of the SBCS have been published elsewhere (9,22). Briefly, the SBCS included two recruitment phases. The initial phase of subject recruitment (SBCS-I) was carried out to recruit women who were newly diagnosed with breast cancer between the ages of 25 and 64 years during the period from August 1996 to March 1998. All study participants were permanent residents in urban Shanghai who had no history of cancer and were alive at the time of interview. Through a rapid case ascertainment system and the population-based Shanghai Cancer Registry, 1602 incident breast cancer patients were identified, of whom 1459 (91%) completed an in-person interview using a structured questionnaire to obtain detailed information regarding demographic characteristics and known and suspected risk factors for breast cancer. Cancer diagnoses were systematically reviewed and confirmed by an expert panel for the study that included two senior pathologists. The vast majority of the breast cancer patients (>98%) were diagnosed with invasive cancer. Control subjects were randomly selected from the general female population of urban Shanghai by use of the Shanghai Resident Registry and were frequency matched to the cancer patients by age. In-person interviews were completed for 1556 control subjects (90.3% of those initially selected). Blood samples were obtained from 1193 case patients and 1310 control subjects who completed the interview (81.8% and 84.2% of case patients and control subjects, respectively, who completed the in-person interview). Eligibility criteria for control subjects were identical to those for case patients with the exception of breast cancer diagnosis (ie, 25–64 years old during the period from August 1996 to March 1998, permanent residents in urban Shanghai who had no history of cancer, alive, and completed an in-person interview).
The second phase of subject recruitment (SBCS-II) was conducted between 2002 and 2005 using a protocol similar to the one used during the initial phase. A total of 1989 eligible incident case patients diagnosed with breast cancer between April 2002 and February 2005 and 1989 community-based control subjects were recruited and completed an in-person interview using a structured questionnaire similar to the one used in the SBCS-I, corresponding to 83.7% and 70.4%, respectively, of those identified. The majority of SBCS-II study participants (1932 case patients and 1857 control subjects) provided a blood sample or an exfoliated buccal cell sample that was processed and stored at 70°C. These women, along with those who provided a blood sample in SBCS-I, were included in this study. In total, 3125 case patients and 3167 control subjects were eligible for the current study. All participants were measured for their current weight, height, and waist and hip circumferences. All interviews were tape-recorded and reviewed by the field supervisor and quality control staff to assess the quality of interview data. Institutional review board approval was granted by the Shanghai Cancer Institute, Shanghai Center for Disease Prevention and Control, and Vanderbilt University Medical Center. All study subjects provided written informed consent to participate in the study.
Genotyping Methods
Laboratory protocols for DNA extraction and genotyping in the SBCS have been described in detail elsewhere (9). Briefly, genomic DNA for genotyping assays was extracted from buffy coat fractions of the blood samples or from exfoliated buccal cells. Of the 12 SNPs evaluated in this study, seven (rs2180341, rs3817198, rs3803662, rs2046210, rs1219648, rs2981582, and rs8051542) were included on the Affymetrix Genome-Wide Human SNP Array 6.0 (Santa Clara, CA) or on both the Affymetrix 6.0 and the Affymetrix GeneChip Human Mapping 500K Array Set. Because 4157 samples were previously genotyped using the Affymetrix 6.0 Array (1928 case patients and 1938 control subjects) or the Affymetrix 500K Array (145 case patients and 146 control subjects), genotyping data for these seven SNPs from these 4157 subjects were obtained from these SNP arrays. DNA samples for the remaining participants were genotyped for these seven SNPs by using the iPLEX Sequenom MassARRAY platform (San Diego, CA). For the remaining five SNPs (rs13387042, rs10941679, rs889312, rs13281615, and rs12443621), all samples were genotyped using the TaqMan allelic discrimination assays with an ABI PRISM 7900 Sequence Detection System (Applied Biosystems, Foster City, CA).
Quality control protocols for genotyping assays have also been described elsewhere (9). In the assays using Affymetrix SNP arrays, each 96-well genotyping plate included one negative control (water only) and three positive quality control samples purchased from Coriell Cell Repositories (Camden, NJ). Genotyping data from these three quality control samples, each genotyped approximately 45 times, showed an average concordance rate of 99.85%. In the assays using the Sequenom or the TaqMan assay, each 96-well plate included two negative controls (water only), two sets of blinded study samples, and two genomic DNA samples selected from subjects of European (n = 60) or Chinese (n = 45) ancestry obtained from the HapMap project. The mean concordance rate was 99.6% for the blinded study samples and 100% for the HapMap samples compared with data from the HapMap project in the Sequenom assays and 96.7% for the blinded study samples and 99.3% for the HapMap samples compared with data from the HapMap project in the TaqMan assays.
Statistical Analysis
Included in this study were 3039 case patients and 3082 control subjects whose DNA samples were successfully genotyped. Differences between case patients and control subjects in selected demographic characteristics and established risk factors were compared using Student t tests (for continuous variables) or χ2 tests (for categorical variables). Relative risks were estimated using odds ratios (ORs) and 95% confidence intervals (CIs) that were derived from logistic regression with adjustment for potential confounding factors, including age, education level, and SBCS recruitment phases. A genetic risk score to measure the cumulative effect of multiple genetic risk variants was created using the equation , where k is the number of SNPs replicated in this study and βi is the regression coefficient for SNPi that was derived from a logistic regression model. Nonlinear associations between the genetic risk score and breast cancer risk were evaluated in logistic regression that included nonlinear terms created with the restricted cubic spline function (23). A logistic regression model that included the genetic risk score, age at menarche, age at first live birth, waist-to-hip ratio, a previous diagnosis of benign breast disease, history of breast cancer in a first-degree relative, body mass index, and parity was used to examine independent associations of these factors with breast cancer risk. Potential interactions between any two of the independent variables were evaluated using likelihood ratio tests by comparing models with and without the interaction terms.
A risk assessment model was constructed by including only factors that were independently associated with breast cancer risk to maintain parsimony. The internal validity of the model was evaluated using a bootstrap method (24,25) involving 2000 replications to adjust model regression coefficients for potential overfitting. A detailed description of the bootstrap procedure is provided in the Supplementary Methods (available online). Model performance was evaluated using the area under the receiver operating characteristic curve (also known as a concordance [or c] statistic) (26). The added predictive value of a particular predictor was evaluated by comparing models with and without the predictor with regard to their c statistics (26) and the integrated discrimination improvement (27). The difference in c statistics between two models was tested using a nonparametric Mann–Whitney U test described by DeLong et al. (26). The integrated discrimination improvement was evaluated using a simple asymptotic Z test that was proposed by Pencina et al. (27). These two statistics were calibrated using the same bootstrap procedures described above.
We used an approach similar to that described for the Gail model (12,28) to estimate the absolute risk of breast cancer according to the risk factors that a woman carried. We did not use the Gail model, which was established using data from studies conducted in the United States, because it may not predict breast cancer risk well for women living in Asia given that the strength of the association for breast cancer risk factors may differ by race and the age-specific incidence rates for breast cancer are lower in China than in the United States. Furthermore, because mammographic screening is not as widely used in China as in the United States, data on certain predictors, such as the number of breast biopsies, are not available for most Chinese women. Model parameters included the bootstrap-calibrated relative risk of breast cancer associated with the genetic risk score and traditional risk factors estimated from the SBCS and the age-specific breast cancer incidence and mortality rates in urban Shanghai during 2002 and 2003, the latest available data at the time of the study (Supplementary Methods, available online). Reclassification measures described by Cook and Ridker (29) were used to evaluate the improvement of adding the genetic risk score to a model that included traditional risk factors only.
To evaluate the utility of risk assessment models for stratifying risk at the population level, we estimated the proportion of breast cancers in the general population that could be accounted for above a given risk threshold as predicted by models that included the genetic risk score alone, traditional risk factors alone, or both the genetic risk score and traditional risk factors by using the approach described by Pharoah et al. (11,30).
Detailed descriptions of the methods we used to establish risk prediction models and to evaluate a model's utility in risk stratifications are provided in the Supplementary Methods (available online). All statistical tests were two-sided, and a P value equal to or less than .05 was considered statistically significant.
Results
Case patients and control subjects were comparable with respect to age because of the age frequency–matched design of this study (Table 1). With the exception of hormone replacement therapy use, all established breast cancer risk factors differed statistically significantly between case patients and control subjects. However, very few women in the study population reported ever using hormone replacement therapy. These results were similar for all SBCS subjects, including those who did not provide a genomic DNA sample and thus were not genotyped for the current study (Table 1).
Table 1.
Distribution of demographic characteristics and known breast cancer risk factors for case patients and control subjects, the Shanghai Breast Cancer Study
| Subjects with genotyping data |
All subjects |
|||||
| Variable | Case patients (n = 3039) | Control subjects (n = 3082) | P* | Case patients (n = 3448) | Control subjects (n = 3474) | P* |
| Demographic factors | ||||||
| Mean age, y (SD) | 49.7 (8.4) | 49.9 (8.8) | .3434 | 49.6 (8.3) | 49.7 (8.6) | .7223 |
| Education level, high school or higher, % | 52.2 | 45.8 | <.001 | 52.3 | 46.4 | <.001 |
| Reproductive risk factors | ||||||
| Mean age at menarche, y (SD) | 14.4 (1.7) | 14.7 (1.8) | <.001 | 14.4 (1.6) | 14.7 (1.7) | <.001 |
| Postmenopausal, % | 39.8 | 44.0 | <.001 | 39.7 | 43.4 | .0016 |
| Mean age at menopause†, y (SD) | 48.4 (4.5) | 48.0 (4.7) | .0196 | 48.4 (4.4) | 47.9 (4.7) | .0082 |
| Mean number of live births (SD) | 1.3 (0.8) | 1.4 (0.9) | <.001 | 1.3 (0.8) | 1.4 (0.9) | <.001 |
| Mean age at first live birth‡, y (SD) | 27.3 (3.8) | 26.6 (3.8) | <.001 | 27.2 (3.9) | 26.6 (3.9) | <.001 |
| Ever used hormone replacement therapy†, % | 6.6 | 5.2 | .1096 | 6.5 | 5.0 | .1158 |
| Other risk factors | ||||||
| Family history of breast cancer§, % | 4.6 | 2.8 | <.001 | 4.7 | 2.8 | <.001 |
| Prior breast fibroadenoma, % | 9.8 | 5.5 | <.001 | 9.8 | 5.5 | <.001 |
| Prior breast lobular proliferation, % | 36.2 | 25.1 | <.001 | 35.4 | 24.9 | <.001 |
| Mean body mass index, kg/m2 (SD) | 23.7 (3.3) | 23.4 (3.3) | <.001 | 23.6 (3.3) | 23.3 (3.3) | <.001 |
| Mean waist-to-hip ratio (SD) | 0.82 (0.06) | 0.81 (0.06) | <.001 | 0.82 (0.06) | 0.81 (0.06) | <.001 |
P values (two-sided) were derived from Student t tests (for continuous variables) and χ2 tests (for categorical variables).
Among postmenopausal women.
Among parous women.
Among first-degree relatives.
Six of the 12 breast cancer–associated SNPs identified in previous genome-wide association studies (ie, rs2046210, rs1219648, rs2981582, rs3817198, rs8051542, and rs3803662) were statistically significantly associated with breast cancer risk in this study at P less than or equal to .05 (Table 2). Associations with two additional SNPs were of borderline statistical significance (rs889312: P = .053; rs10941679: P = .097). SNP rs13281615 (P = .145) also showed an association with cancer risk that was consistent with the association reported previously (4). Therefore, we considered these nine SNPs for construction of a genetic risk score. Of the two SNPs located in the FGFR2 gene that are in strong linkage disequilibrium (r2 = .69), we selected rs1219648 for inclusion in the genetic risk score because it showed a stronger association with breast cancer risk than rs2981582 when both SNPs were included in the same model. We selected two of the three SNPs located in the TOX3 gene (ie, rs8051542 and rs3803662) for inclusion in the genetic risk score because each was associated with the risk of breast cancer after adjusting for the other. These two SNPs are located 52 kb apart from each other on chromosome 16q12 and display very weak linkage disequilibrium (r2 = .08). In total, eight SNPs were used to construct the genetic risk score, which had values that ranged from 0 to 1.62 (mean = 0.69). The strongest association was found for rs2046210 (per-allele OR = 1.26, 95% CI = 1.17 to 1.35). Other SNPs, however, were each associated with elevated risks of 16% or lower. When these eight SNPs were combined to form the genetic risk score, we observed a linear association between the score and breast cancer risk (Supplementary Figure 1, available online).
Table 2.
Association between risk of breast cancer and 12 single-nucleotide polymorphisms (SNPs) identified from recent genome-wide association studies, the Shanghai Breast Cancer Study (SBCS)*
| First author, year of initial publication (reference) | SNP | Chromosome/gene | Position† | Allele‡ | Risk allele frequency, % |
OR (95% CI)§ |
OR (95% CI) per risk allele§ | Ptrend‖ | ||
| Case patients | Control subjects | Heterozygote | Homozygote | |||||||
| Stacey, 2007 (6) | rs13387042 | 2q35/unknown | 217614077 | A/G | 11.5 | 11.3 | 1.04 (0.91 to 1.18) | 1.07 (0.69 to 1.66) | 1.03 (0.92 to 1.16) | .554 |
| Stacey, 2008 (7) | rs10941679 | 5p12/MRPS30 | 44742255 | G/A | 51.7 | 50.1 | 1.03 (0.90 to 1.17) | 1.13 (0.98 to 1.31) | 1.07 (0.99 to 1.15) | .097 |
| Easton, 2007 (4) | rs889312 | 5q11.2/MAP3K1 | 56067641 | C/A | 53.5 | 51.7 | 1.02 (0.89 to 1.16) | 1.15 (0.99 to 1.33) | 1.07 (0.99 to 1.15) | .053 |
| Gold, 2008 (8) | rs2180341 | 6q22.33/ECHDC1 | 127642323 | G/A | 24.7 | 25.9 | 0.90 (0.81 to 1.01) | 0.95 (0.77 to 1.17) | 0.93 (0.86 to 1.02) | .157 |
| Zheng, 2009 (9) | rs2046210 | 6q25.1/unknown | 151990059 | A/G | 42.2 | 36.8 | 1.26 (1.13 to 1.41) | 1.58 (1.35 to 1.84) | 1.26 (1.17 to 1.35) | <.001 |
| Easton, 2007 (4) | rs13281615 | 8q24.21/unknown | 128424800 | G/A | 51.5 | 50.2 | 1.07 (0.94 to 1.22) | 1.12 (0.96 to 1.29) | 1.06 (0.98 to 1.14) | .145 |
| Hunter, 2007 (5) | rs1219648 | 10q26.13/FGFR2 | 123336180 | G/A | 42.5 | 39.1 | 1.17 (1.04 to 1.31) | 1.31 (1.12 to 1.54) | 1.15 (1.07 to 1.24) | <.001 |
| Easton, 2007 (4) | rs2981582 | 10q26.13/FGFR2 | 123342307 | A/G | 35.4 | 32.0 | 1.14 (1.02 to 1.27) | 1.36 (1.14 to 1.62) | 1.15 (1.07 to 1.25) | <.001 |
| Easton, 2007 (4) | rs3817198 | 11p15.5/LSP1 | 1865582 | C/T | 13.6 | 12.3 | 1.11 (0.98 to 1.26) | 1.24 (0.85 to 1.80) | 1.12 (1.00 to 1.24) | .049 |
| Easton, 2007 (4) | rs8051542 | 16q12.1/TOX3¶ | 51091668 | T/C | 19.7 | 17.7 | 1.11 (0.98 to 1.24) | 1.48 (1.11 to 1.98) | 1.15 (1.04 to 1.26) | .005 |
| Easton, 2007 (4) | rs12443621 | 16q12.1/TOX3 | 51105538 | A/G | 42.6 | 42.8 | 0.98 (0.87 to 1.09) | 0.98 (0.85 to 1.14) | 0.99 (0.92 to 1.06) | .747 |
| Easton, 2007 (4); Stacey, 2007 (6) | rs3803662 | 16q12.1/TOX3 | 51143842 | A/G | 67.9 | 64.6 | 1.30 (1.09 to 1.54) | 1.42 (1.20 to 1.69) | 1.16 (1.07 to 1.25) | <.001 |
CI = confidence interval; OR = odds ratio.
Chromosome base position according to the National Center for Biotechnology Information database, build 36.
Risk allele/reference allele, as initially reported (based on the forward strand).
Odds ratios for each SNP were estimated separately from a logistic regression model with adjustment for age, education level, and recruitment phases in the SBCS (I and II).
P values for trend (two-sided) were derived from trend tests (df = 1).
Also known as TNRC9, as reported initially.
We next evaluated associations between breast cancer risk and the genetic risk score (based on its quintile distribution in the control subjects) and established risk factors (Table 3). Compared with women who were in the lowest quintile of the genetic risk score, women in the highest quintile had a 1.85-fold (95% confidence interval = 1.58- to 2.18-fold) elevated risk of breast cancer (Ptrend = 2.5 × 10−15). The magnitude of the association with the genetic risk score was virtually unchanged after adjusting for five established risk factors that each also showed a highly statistically significant association with the risk of breast cancer. There was no apparent interaction between any two of the six variables shown in Table 3 or between any one of these variables and age or menopausal status, nor was there an interaction between body mass index and menopausal status (data not shown). Furthermore, body mass index was not associated with the risk of breast cancer after adjusting for waist-to-hip ratio. Parity was also not associated with the risk of breast cancer after adjusting for age at first live birth. Therefore, neither body mass index nor parity was further evaluated in the risk assessment model.
Table 3.
Associations between risk of breast cancer and the genetic risk score and established risk factors, the Shanghai Breast Cancer Study (SBCS)*
| Predictor (code) | Case patients (n = 3039) | Control subjects (n = 3082) | OR (95% CI)† | OR (95% CI)‡ |
| Genetic risk score, quintiles | ||||
| 1 (0 [low]) | 431 | 621 | 1.00 (referent) | 1.00 (referent) |
| 2 (1) | 506 | 613 | 1.19 (1.00 to 1.41) | 1.19 (1.00 to 1.42) |
| 3 (2) | 665 | 616 | 1.57 (1.33 to 1.85) | 1.53 (1.29 to 1.81) |
| 4 (3) | 652 | 618 | 1.52 (1.29 to 1.80) | 1.52 (1.29 to 1.80) |
| 5 (4 [high]) | 785 | 614 | 1.85 (1.58 to 2.18) | 1.83 (1.55 to 2.16) |
| Ptrend | 2.5 × 10−15 | 1.4 × 10−14 | ||
| Age at menarche, y | ||||
| >15 (0) | 780 | 995 | 1.00 (referent) | 1.00 (referent) |
| 14–15 (1) | 1259 | 1226 | 1.26 (1.12 to 1.43) | 1.28 (1.13 to 1.46) |
| ≤13 (2) | 1000 | 861 | 1.37 (1.21 to 1.57) | 1.44 (1.26 to 1.65) |
| Ptrend | 2.8 × 10−6 | 1.1 × 10−7 | ||
| Age at first live birth, y | ||||
| <25 (0) | 647 | 801 | 1.00 (referent) | 1.00 (referent) |
| 25–29 (1) | 1631 | 1675 | 1.07 (0.94 to 1.23) | 1.19 (1.05 to 1.36) |
| ≥30 or nulliparous (2) | 761 | 606 | 1.37 (1.17 to 1.61) | 1.56 (1.34 to 1.82) |
| Ptrend | 6.6 × 10−5 | 1.1 × 10−8 | ||
| Waist-to-hip ratio, tertiles | ||||
| <0.783 (0) | 703 | 1003 | 1.00 (referent) | 1.00 (referent) |
| 0.783–0.829 (1) | 986 | 1045 | 1.41 (1.23 to 1.61) | 1.40 (1.23 to 1.60) |
| ≥0.830 (2) | 1350 | 1034 | 2.06 (1.80 to 2.35) | 2.06 (1.81 to 2.35) |
| Ptrend | 1.1 × 10−16 | 2.1 × 10−28 | ||
| Family history of breast cancer | ||||
| No (0) | 2898 | 2995 | 1.00 (referent) | 1.00 (referent) |
| Yes (1) | 141 | 87 | 1.61 (1.22 to 2.12) | 1.52 (1.15 to 2.00) |
| Prior BBD diagnosis§ | ||||
| No (0) | 1987 | 2346 | 1.00 (referent) | 1.00 (referent) |
| Yes (1) | 1052 | 736 | 1.60 (1.43 to 1.79) | 1.71 (1.53 to 1.92) |
BBD = benign breast disease; CI = confidence interval; OR = odds ratio.
Adjusted for age, education level, and study stage.
Adjusted for age, education level, SBCS recruitment phases, and all variables listed in the table.
Includes breast fibroadenoma and breast lobular proliferation diagnosed any time from birth to the time point 2 years before breast cancer diagnosis (for case patients) or interview (for control subjects).
To estimate the relative risk of breast cancer for a woman with a given combination of risk factors, we constructed a logistic regression model that included the genetic risk score and the five established risk factors, which was expressed as follows: OR = exp (0.14463 × genetic risk score + 0.36447 × waist-to-hip ratio + 0.53218 × benign breast disease + 0.18190 × age at menarche + 0.22379 × age at first live birth + 0.42243 × family history of breast cancer). To correct potential overfitting, we adjusted model parameters using bootstrapping (Table 4). The values of these adjusted parameters were almost the same as those estimated from the unadjusted analysis, indicating that there was little overfitting of the model.
Table 4.
Log odds ratio estimates (βi ) and covariances for a logistic regression model that included all variables*
| Statistic | GRS | WHR | BBD | AGEMEN | AGEFLB | BCFH |
| Parameter (βi )† | 0.13075 | 0.36325 | 0.53379 | 0.18226 | 0.22191 | 0.41784 |
| Covariance† | 3.53424 | 0.25528 | −0.13522 | 0.03719 | 0.07164 | −0.56488 |
| 10.86988 | 1.79466 | 0.15770 | 1.34940 | −0.12958 | ||
| 34.17435 | −0.12633 | −1.25011 | −1.11121 | |||
| 11.70274 | −0.58407 | −0.64148 | ||||
| 15.34808 | −1.31858 | |||||
| 201.58482 |
Codes for predictors included in the model are genetic risk score (GRS) (in quintiles, with values ranging from 0 to 4), waist-to-hip ratio (WHR) (<0.783 [0], 0.783–0.829 [1], ≥0.830 [2]), benign breast disease (BBD) (no [0], yes [1]), age at menarche (AGEMEN) (>15 years [0], 14–15 years [1], ≤13 years [2]), age at first live birth (AGEFLB) (<25 years [0], 25–29 years [1], ≥30 years or nulliparous [2]), and family history of breast cancer (BCFH) (no [0], yes [1]). GRS is calculated as follows: GRS = 0.13111 × rs1219648 + 0.13927 × rs3803662 + 0.10133 × rs3817198 + 0.073668 × rs8051542 + 0.22184 × rs2046210 + 0.058222 × rs13281615 + 0.055379 × rs889312 + 0.066955 × rs10941679. For example, to estimate the relative risk of breast cancer for a women with an average GRS (GRS = 0.69), a WHR of 0.80 (WHR = 1), without BBD (BBD = 0), AGEMEN = 13 years (AGEMEN = 2), AGEMEN = 25 years (AGEFLB = 1), and a BCHF (BCHF = 1): OR = exp (0.13075 × 0.69 + 0.36235 × 1 + 0.53379 × 0 + 0.18226 × 2 + 0.22191 × 1 + 0.41784 × 1) = exp (1.4568) = 4.29.
Parameters (βi) were adjusted for overfitting using bootstrapping. The covariance was derived from the original logistic regression model. The covariance values shown in the table were multiplied by 10 000.
The adjusted c statistic for the full model described above was 0.6295, which was slightly lower than that derived from unadjusted analyses (0.6335). The contribution of the genetic risk score to the model, as measured by the reduction in adjusted c statistic from the full model, was 0.0117 (Table 5), which, along with a previous diagnosis of benign breast disease (reduction in adjusted c statistic = 0.0175) and waist-to-hip ratio (reduction in adjusted c statistic = 0.0276), was among the three strongest predictors in the model. The corresponding reduction in c statistics for the other three predictors included in the model were 0.0070 for age at first live birth, 0.0052 for age at menarche, and 0.0016 for a family history of breast cancer (Table 5). Analyses using integrated discrimination improvement provided results consistent with those using decreases in the c statistics.
Table 5.
Changes in the c statistic and integrated discrimination improvement (IDI) for each predictor when it was removed from the full model, the Shanghai Breast Cancer Study*
| Predictor | Decrease in c statistic (P) | IDI (P) |
| Waist-to-hip ratio | 0.0276 (<.0001) | 0.0195 (<.0001) |
| Previous benign breast disease diagnosis | 0.0175 (<.0001) | 0.0131 (<.0001) |
| Genetic risk score | 0.0117 (<.0001) | 0.0084 (<.0001) |
| Age at first live birth | 0.0070 (.0058) | 0.0050 (<.0001) |
| Age at menarche | 0.0052 (.0124) | 0.0043 (<.0001) |
| Breast cancer family history | 0.0016 (.1584) | 0.0013 (.0029) |
All estimates were adjusted for overfitting by bootstrapping. The adjusted c statistic for the full model that included all predictors listed in this table was 0.6295. All P values are two-sided and derived from the nonparametric Mann–Whitney U test for c statistics (26) or the Z test for IDIs (27).
Using the relative risks estimated in this study and the age-specific breast cancer incidence and non-breast cancer mortality rates for urban Shanghai, we estimated the probabilities of developing breast cancer over a specific time period for women at a given age and relative risk of breast cancer (Supplementary Table 2, available online). For example, the probability of developing breast cancer for a woman aged 50 years with a relative risk of 5 would be 1.58% for the next 10 years and 2.97% for the next 20 years.
We also examined the improvement in risk classification that would result from adding the genetic risk score to a model that included the established risk factors only (Table 6). Assuming that women with a 10-year risk of breast cancer of 1.5% or higher would be recommended for a cancer screening, the full model would reclassify 4.8% [3.9/(47.2 + 33.6)] of women to receive a screening and 22.9% [(4.3 + 0.1)/(14.1 + 3.0 + 2.1)] of women not to receive a screening compared with the model with established risk factors only. The improvement in risk classification of the full model would vary with the threshold defined for cancer screening.
Table 6.
Distribution (%) of Chinese women aged 50–59 years by joint risk categories (10-year probabilities of developing breast cancer) according to models that include traditional risk factors only or both genetic and traditional risk factors
| Risk categories defined by a model that includes established risk factors only | Risk categories defined by a model that includes genetic and established risk factors |
|||||
| <1.0% | 1.0% to <1.5% | 1.5% to <2.0% | 2.0% to <2.5% | ≥2.5% | Total | |
| <1.0% | 42.2 | 5.0 | 0.0 | 0.0 | 0.0 | 47.2 |
| 1.0% to <1.5% | 10.2 | 19.6 | 3.9 | 0.0 | 0.0 | 33.6 |
| 1.5% to <2.0% | 0.0 | 4.3 | 6.9 | 2.8 | 0.0 | 14.1 |
| 2.0% to <2.5% | 0.0 | 0.1 | 1.1 | 1.0 | 0.8 | 3.0 |
| ≥2.5% | 0.0 | 0.0 | 0.0 | 0.7 | 1.4 | 2.1 |
| Total | 52.4 | 29.0 | 11.8 | 4.5 | 2.2 | 100.0 |
The distribution of the log relative risks of breast cancer estimated using the model that included the genetic risk score and the established risk factors was approximately normal, with an SD of 0.45 (Supplementary Figure 2, available online). Women in the highest 20% of this risk distribution had a relative risk of 3.54 compared with women in the lowest 20% of the risk distribution. We also found approximately normal distributions for models that included the genetic risk score only (SD = 0.21) or the established risk factors only (SD = 0.41) (data not shown). On the basis of these distributions, we estimated that approximately 37.7% of breast cancer cases in the general population would be found among those at the top 30% risk threshold as defined by the genetic risk score (Model A) and 45.6% of breast cancer cases in the general population would be found among those at the top 30% risk threshold as defined by established risk factors (Model B) (Table 7). The prediction accuracy would be further improved using the full model that included both the genetic risk score and established risk factors, and approximately 47.0% of breast cancer cases in the general population could be found in the top 30% risk group as defined by the full model (Model C). In other words, a screening program that targets women who are in the top 30% risk group as predicted by the full model would capture approximately 47% of breast cancer cases in the general population, a 56.7% improvement in case finding compared with a screening program that did not include risk assessment.
Table 7.
Proportion of breast cancer case patients in the general population with risks exceeding the risk thresholds predicted by risk assessment models*
| Proportion (%) of the general population with risk exceeding threshold predicted by models | Proportion (%) of breast cancer patients with risk exceeding the risk threshold predicted by the model | ||
| Model A | Model B | Model C | |
| 10 (high risk)† | 14.2 | 19.2 | 20.3 |
| 20 | 26.4 | 33.3 | 34.8 |
| 30 | 37.7 | 45.6 | 47.0 |
| 40 | 48.3 | 56.2 | 57.8 |
| 50 | 58.3 | 65.9 | 67.4 |
| 60 | 67.8 | 74.6 | 75.9 |
| 70 | 76.7 | 82.5 | 83.5 |
| 80 | 85.4 | 89.5 | 90.2 |
| 90 | 93.2 | 95.5 | 95.8 |
Model A includes the genetic risk score derived from eight single-nucleotide polymorphisms (rs2046210, rs1219648, rs3817198, rs8051542, rs3803662, rs889312, rs10941679, and rs13281615). Model B includes five traditional risk factors. Model C includes both the genetic risk score and traditional risk factors.
Women in the top 10% risk group predicted by the model.
Discussion
To our knowledge, this is the first study to systematically evaluate the association between 12 SNPs identified in recent genome-wide association studies of breast cancer and the risk of breast cancer. Using data from the SBCS, a large population-based case–control study that included 6122 participants, we found that most of the SNPs identified in previous genome-wide association studies conducted among women of European ancestry are also associated with the risk of breast cancer among Chinese women. These results extend findings from previous genome-wide association studies to another ethnic group and provide additional support for associations between these SNPs and the risk of breast cancer. Furthermore, data from this study may help to guide fine mapping efforts to identify causal variants for breast cancer. We have also shown that the genetic risk score, an aggregate measure of the combined effect of multiple genetic risk variants, is among the strongest predictors identified thus far for breast cancer. Although the genetic risk score had moderate discriminatory accuracy, it in combination with established risk factors for breast cancer showed promise for risk assessments to stratify women into high- vs low-risk groups, which could then be targeted for cost-efficient cancer prevention.
Asian women have been underrepresented in previous genome-wide association studies. Two previous studies (4,6) included some Asian women in their replication stages, and results for three (rs2981582, rs3803662, and rs889312) of the five SNPs evaluated in Asian women were replicated (4). The sample size for Asian women included in these studies, however, was small, and the results were not conclusive. We have shown for the first time, to our knowledge, that three other SNPs (rs10941679, rs3817198, and rs8051542) reported initially as being associated with the risk of breast cancer among women of European ancestry are also associated with breast cancer risk among Asian women. Although results for eight of the 11 SNPs initially identified in women of European ancestry were replicated in our study, the strength of the associations in Asian women was, in general, weaker than those reported previously. Furthermore, results for three SNPs previously found to be associated with breast cancer risk among women of European ancestry were not replicated in this study. One of these SNPs, rs13387042, was also not replicated in the Asian samples included in the original study (6), and the other two SNPs (rs218034 and rs12443621) have not been evaluated previously in any other Asian populations. The lack of an association with some SNPs in Asian women and the weaker association with some other SNPs in Asians compared with Europeans are perhaps not surprising given that most, if not all, of the SNPs identified by genome-wide association studies are associated with breast cancer risk through their strong linkage disequilibrium with causal variants and there are differences in the genetic architecture between women of Chinese and European descents.
Of the 12 SNPs evaluated in this study, the strongest association was found for rs2046210, for which there was a 26% elevated risk of breast cancer with each risk allele. However, the other SNPs were, in general, weakly associated with the risk of breast cancer, with each allele conferring elevated risks of 16% or lower. An SNP that confers such a small to moderate risk has limited utility in risk assessment when it is considered alone. By contrast, the genetic risk score, an aggregate measure of the risk associated with multiple risk variants, was one of the strongest predictors included in our risk assessment model. Addition of the genetic risk score to the risk assessment model provided a small but meaningful improvement in the discriminatory accuracy as measured by using the c statistic, which is an insensitive measure of the improvement in risk prediction when a new predictor is considered (27,31,32), as was the case when any other predictors, such as a family history of breast cancer and a previous diagnosis of benign breast disease, were added. Nevertheless, this score, in combination with established risk factors, appeared to be promising in identifying high-risk women who should be targeted for prevention programs.
We found that a model that included the genetic risk score alone did not provide a better risk prediction than the model that included all established risk factors. However, with the exception of rs2046210, all other breast cancer–associated SNPs were initially identified in studies conducted in women of European ancestry, and these SNPs, in general, showed a weaker association with breast cancer risk in Asians than in Europeans. It is likely that these breast cancer–associated SNPs may perform better for risk prediction in European populations than in Asian populations. Furthermore, the genetic risk variants included in this study were identified after 2007, and additional genetic risk factors, particularly those unique for Asian women, are likely to be discovered, which could further improve the discriminatory accuracy of breast cancer prediction models. An obvious advantage of including genetic variants in risk assessments is that they can be accurately measured at any time during a person's life. By contrast, some established risk factors, such as waist-to-hip ratio and reproductive history, are prone to measurement error. Furthermore, the data for certain established risk factors may vary with the time window of measurement. For example, the probability of being diagnosed with benign breast disease or of having a family history of breast cancer may increase as a woman ages; thus, these predictors need to be constantly updated. Although the cost to generate SNP data has decreased dramatically over the past few years and further cost reduction is expected in the coming years, the costs of genotyping as well as costs related to sample collection and processing should be considered in any program that will incorporate genetic markers in risk prediction. Furthermore, given the substantial role of traditional risk factors in risk assessment, it will be important for risk assessment programs to obtain information on these predictors.
This study has several limitations. With a c statistic of 0.6295, the full risk assessment model established in this study provided only moderate discriminatory accuracy, which would be inadequate for cancer diagnosis and screening. Furthermore, the absolute risk provided by the model was based on breast cancer incidence and total mortality data from Shanghai, which would be applicable only to populations with comparable rates. However, this study demonstrates the value of using both genetic and traditional predictors in risk stratification at the population level. For example, a screening program that targets women in the top 30% of risk as predicted by the full model could identify approximately 47% of breast cancer cases in the general population and would increase the number detected by 56.7% compared with a program with no risk assessment. This improvement could have substantial public health implications, particularly for a population in which no population-based screening program is available. Nevertheless, further improvement in model discriminatory accuracy will be needed to enhance the utility of the model in population risk stratification.
There has been a rapid increase in breast cancer incidence over the past few decades in many Asian countries (20), including China (33), Japan (34), Korea (35), and Singapore (36). Breast cancer is now one of the most common malignancies diagnosed among women in many Asian countries (20,33–36). It is predicted that this upward trend will continue with the economic expansion and lifestyle changes that are occurring in China (21) and other Asian countries, such as Singapore (37). Although breast cancer screening programs are still in their infancy in most Asian countries, cost-efficient breast cancer screening programs are being explored in several countries (38,39). Given that the incidence of female breast cancer in many Asian countries remains less than a one-third that observed in most North American and European countries (20), it would be difficult to justify the implementation of annual mammographic screening for all women at age 40 years, as recommended by the American Cancer Society, in Asian countries. Instead, a viable alternative approach may be to identify high-risk women who should be targeted for such screenings, particularly given the financial constraints in some Asian countries that prohibit implementing full-scale population-based breast cancer screening programs.
In summary, we have shown in this study that the majority of the genetic risk variants identified in genome-wide association studies of breast cancer conducted among women of European ancestry are also associated with breast cancer in Asian women. Although each of these genetic variants confers only a small to moderate increased risk of breast cancer, they are among the strongest predictors identified to date when considered in combination. Risk assessment models that incorporate both a genetic risk score based on these SNPs and the established risk factors for breast cancer may be useful for identifying high-risk women for targeted cancer prevention. As additional genetic variants are identified, we anticipate that the precision of risk assessments that use a combination of the genetic risk score and traditional risk factors will be further improved.
Supplementary Data
Supplementary data can be found at http://www.jnci.oxfordjournals.org/.
Funding
National Institutes of Health (R01CA124558, R01CA64277, R37CA70867, and R01CA90899).
Supplementary Material
Footnotes
The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The study sponsor had no role in the design of the study; the collection, analysis, or interpretation of the data; the writing of the manuscript; or the decision to submit the manuscript for publication.
None of the authors have any conflict of interest.
The authors wish to thank the study participants and research staff for their contributions and commitment to this project. Genotyping assays were carried out at Proactive Genomics (Winston-Salem, NC) as well as at the Survey and Biospecimen and Functional Genomics Shared Resources Facility (Vanderbilt University School of Medicine, Nashville, TN), which is supported in part by the Vanderbilt-Ingram Cancer Center (P30 CA68485).
References
- 1.Nathanson KL, Wooster R, Weber BL. Breast cancer genetics: what we know and what we need. Nat Med. 2001;7(5):552–556. doi: 10.1038/87876. [DOI] [PubMed] [Google Scholar]
- 2.Balmain A, Gray J, Ponder B. The genetics and genomics of cancer. Nat Genet. 2003;33(suppl):238–244. doi: 10.1038/ng1107. [DOI] [PubMed] [Google Scholar]
- 3.Walsh T, Casadei S, Coats KH, et al. Spectrum of mutations in BRCA1, BRCA2, CHEK2, and TP53 in families at high risk of breast cancer. JAMA. 2006;295(12):1379–1388. doi: 10.1001/jama.295.12.1379. [DOI] [PubMed] [Google Scholar]
- 4.Easton DF, Pooley KA, Dunning AM, et al. Genome-wide association study identifies novel breast cancer susceptibility loci. Nature. 2007;447(7148):1087–1093. doi: 10.1038/nature05887. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Hunter DJ, Kraft P, Jacobs KB, et al. A genome-wide association study identifies alleles in FGFR2 associated with risk of sporadic postmenopausal breast cancer. Nat Genet. 2007;39(7):870–874. doi: 10.1038/ng2075. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Stacey SN, Manolescu A, Sulem P, et al. Common variants on chromosomes 2q35 and 16q12 confer susceptibility to estrogen receptor-positive breast cancer. Nat Genet. 2007;39(7):865–869. doi: 10.1038/ng2064. [DOI] [PubMed] [Google Scholar]
- 7.Stacey SN, Manolescu A, Sulem P, et al. Common variants on chromosome 5p12 confer susceptibility to estrogen receptor-positive breast cancer. Nat Genet. 2008;40(6):703–706. doi: 10.1038/ng.131. [DOI] [PubMed] [Google Scholar]
- 8.Gold B, Kirchhoff T, Stefanov S, et al. Genome-wide association study provides evidence for a breast cancer risk locus at 6q22.33. Proc Natl Acad Sci U S A. 2008;105(11):4340–4345. doi: 10.1073/pnas.0800441105. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Zheng W, Long J, Gao YT, et al. Genome-wide association study identifies a new breast cancer susceptibility locus at 6q25.1. Nat Genet. 2009;41(3):324–328. doi: 10.1038/ng.318. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Gail MH. Discriminatory accuracy from single-nucleotide polymorphisms in models to predict breast cancer risk. J Natl Cancer Inst. 2008;100(14):1037–1041. doi: 10.1093/jnci/djn180. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Pharoah PD, Antoniou AC, Easton DF, Ponder BA. Polygenes, risk prediction, and targeted prevention of breast cancer. N Engl J Med. 2008;358(26):2796–2803. doi: 10.1056/NEJMsa0708739. [DOI] [PubMed] [Google Scholar]
- 12.Gail MH, Brinton LA, Byar DP, et al. Projecting individualized probabilities of developing breast cancer for white females who are being examined annually. J Natl Cancer Inst. 1989;81(24):1879–1886. doi: 10.1093/jnci/81.24.1879. [DOI] [PubMed] [Google Scholar]
- 13.Rosner B, Colditz GA. Nurses’ health study: log-incidence mathematical model of breast cancer incidence. J Natl Cancer Inst. 1996;88(6):359–364. doi: 10.1093/jnci/88.6.359. [DOI] [PubMed] [Google Scholar]
- 14.Costantino JP, Gail MH, Pee D, et al. Validation studies for models projecting the risk of invasive and total breast cancer incidence. J Natl Cancer Inst. 1999;91(18):1541–1548. doi: 10.1093/jnci/91.18.1541. [DOI] [PubMed] [Google Scholar]
- 15.Gail MH, Costantino JP, Pee D, et al. Projecting individualized absolute invasive breast cancer risk in African American women. J Natl Cancer Inst. 2007;99(23):1782–1792. doi: 10.1093/jnci/djm223. [DOI] [PubMed] [Google Scholar]
- 16.Chlebowski RT, Anderson GL, Lane DS, et al. Predicting risk of breast cancer in postmenopausal women by hormone receptor status. J Natl Cancer Inst. 2007;99(22):1695–1705. doi: 10.1093/jnci/djm224. [DOI] [PubMed] [Google Scholar]
- 17.Barlow WE, White E, Ballard-Barbash R, et al. Prospective breast cancer risk prediction model for women undergoing screening mammography. J Natl Cancer Inst. 2006;98(17):1204–1214. doi: 10.1093/jnci/djj331. [DOI] [PubMed] [Google Scholar]
- 18.Tice JA, Cummings SR, Smith-Bindman R, Ichikawa L, Barlow WE, Kerlikowske K. Using clinical factors and mammographic breast density to estimate breast cancer risk: development and validation of a new predictive model. Ann Intern Med. 2008;148(5):337–347. doi: 10.7326/0003-4819-148-5-200803040-00004. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Chen J, Pee D, Ayyagari R, et al. Projecting absolute invasive breast cancer risk in white women with a model that includes mammographic density. J Natl Cancer Inst. 2006;98(17):1215–1226. doi: 10.1093/jnci/djj332. [DOI] [PubMed] [Google Scholar]
- 20.Parkin DM, Whelan SL, Ferlay J, Storm H. Cancer Incidence in Five Continents. 7th ed. Vol. 1–8. Lyon: France; 2005. IARC CancerBase No. 7. [Google Scholar]
- 21.Linos E, Spanos D, Rosner BA, et al. Effects of reproductive and demographic changes on breast cancer incidence in China: a modeling analysis. J Natl Cancer Inst. 2008;100(19):1352–1360. doi: 10.1093/jnci/djn305. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Gao YT, Shu XO, Dai Q, et al. Association of menstrual and reproductive factors with breast cancer risk: results from the Shanghai Breast Cancer Study. Int J Cancer. 2000;87(2):295–300. doi: 10.1002/1097-0215(20000715)87:2<295::aid-ijc23>3.0.co;2-7. [DOI] [PubMed] [Google Scholar]
- 23.Harrell FE., Jr. Regression modeling strategies: with applications to linear models, logistic regression, and survival analysis. Springer. 2001 [Google Scholar]
- 24.Steyerberg EW, Harrell FE, Jr, Borsboom GJ, Eijkemans MJ, Vergouwe Y, Habbema JD. Internal validation of predictive models: efficiency of some procedures for logistic regression analysis. J Clin Epidemiol. 2001;54(8):774–781. doi: 10.1016/s0895-4356(01)00341-9. [DOI] [PubMed] [Google Scholar]
- 25.Efron B, Tibshirani R. Improvements on cross-validation: The .632+ bootstrap method. J Amer Statist Assoc. 1997;92(438):548–560. [Google Scholar]
- 26.DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44(3):837–845. [PubMed] [Google Scholar]
- 27.Pencina MJ, D’Agostino RB, Sr, D’Agostino RB, Jr, Vasan RS. Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond. Stat Med. 2008;27(2):157–172. doi: 10.1002/sim.2929. [DOI] [PubMed] [Google Scholar]
- 28.Benichou J, Gail MH. Methods of inference for estimates of absolute risk derived from population-based case-control studies. Biometrics. 1995;51(1):182–194. [PubMed] [Google Scholar]
- 29.Cook NR, Ridker PM. Advances in measuring the effect of individual predictors of cardiovascular risk: the role of reclassification measures. Ann Intern Med. 2009;150(11):795–802. doi: 10.7326/0003-4819-150-11-200906020-00007. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Pharoah PD, Antoniou A, Bobrow M, Zimmern RL, Easton DF, Ponder BA. Polygenic susceptibility to breast cancer and implications for prevention. Nat Genet. 2002;31(1):33–36. doi: 10.1038/ng853. [DOI] [PubMed] [Google Scholar]
- 31.Pepe MS, Janes HE. Gauging the performance of SNPs, biomarkers, and clinical factors for predicting risk of breast cancer. J Natl Cancer Inst. 2008;100(14):978–979. doi: 10.1093/jnci/djn215. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Gail MH. Value of adding single-nucleotide polymorphism genotypes to a breast cancer risk model. J Natl Cancer Inst. 2009;101(13):959–963. doi: 10.1093/jnci/djp130. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Zheng W, Chow WH, Yang G, et al. The Shanghai Women's Health Study: rationale, study design, and baseline characteristics. Am J Epidemiol. 2005;162(11):1123–1131. doi: 10.1093/aje/kwi322. [DOI] [PubMed] [Google Scholar]
- 34.Minami Y, Tsubono Y, Nishino Y, Ohuchi N, Shibuya D, Hisamichi S. The increase of female breast cancer incidence in Japan: emergence of birth cohort effect. Int J Cancer. 2004;108(6):901–906. doi: 10.1002/ijc.11661. [DOI] [PubMed] [Google Scholar]
- 35.Yoo KY, Kang D, Park SK, et al. Epidemiology of breast cancer in Korea: occurrence, high-risk groups, and prevention. J Korean Med Sci. 2002;17(1):1–6. doi: 10.3346/jkms.2002.17.1.1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Seow A, Duffy SW, McGee MA, Lee J, Lee HP. Breast cancer in Singapore: trends in incidence 19681992. Int J Epidemiol. 1996;25(1):40–45. doi: 10.1093/ije/25.1.40. [DOI] [PubMed] [Google Scholar]
- 37.Chia KS, Reilly M, Tan CS, et al. Profound changes in breast cancer incidence may reflect changes into a Westernized lifestyle: a comparative population-based study in Singapore and Sweden. Int J Cancer. 2005;113(2):302–306. doi: 10.1002/ijc.20561. [DOI] [PubMed] [Google Scholar]
- 38.Tan SM, Evans AJ, Lam TP, Cheung KL. How relevant is breast cancer screening in the Asia/Pacific region? Breast. 2007;16(2):113–119. doi: 10.1016/j.breast.2006.08.005. [DOI] [PubMed] [Google Scholar]
- 39.Okonkwo QL, Draisma G, der KA, Brown ML, de Koning HJ. Breast cancer screening policies in developing countries: a cost-effectiveness analysis for India. J Natl Cancer Inst. 2008;100(18):1290–1300. doi: 10.1093/jnci/djn292. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.