Performance of the Breast Cancer Risk Assessment Tool Among Women Aged 75 Years and Older

JNCI: Journal of the National Cancer Institute, Mar 2016

Schonberg, Mara A., Li, Vicky W., Eliassen, A. Heather, Davis, Roger B., LaCroix, Andrea Z., McCarthy, Ellen P., Rosner, Bernard A., Chlebowski, Rowan T., Rohan, Thomas E., Hankinson, Susan E., et al.

A PDF file should load here. If you do not see its contents the file may be temporarily unavailable at the journal website or you do not have a PDF plug-in installed and enabled in your browser.

Alternatively, you can download the file locally and open with any standalone PDF reader:

Performance of the Breast Cancer Risk Assessment Tool Among Women Aged 75 Years and Older

First published online November Perform ance of the Breast Cancer Risk Assess m ent Tool A m ong W o m en Aged 75 Years and Older Mara A. Schonberg 0 1 2 3 4 5 6 Vicky 0 1 2 3 4 5 6 W. Li 0 1 2 3 4 5 6 A. Heather Eliassen 0 1 2 3 4 5 6 Roger B. Davis 0 1 2 3 4 5 6 Andrea Z. LaCroix 0 1 2 3 4 5 6 Ellen P. McCarthy 0 1 2 3 4 5 6 Bernard A. Rosner 0 1 2 3 4 5 6 Rowan T. Chlebowski 0 1 2 3 4 5 6 Thomas E. Rohan 0 1 2 3 4 5 6 Susan E. Hankinson 0 1 2 3 4 5 6 Edward R. Marcantonio 0 1 2 3 4 5 6 Long H. Ngo 0 1 2 3 4 5 6 0 Medicine, Brigham and Women's Hospital, Harvard Medical School, Harvard School of Public Health , Boston, MA (AHE, BAR , SEH); Division of Epidemiology 1 Boston , MA (MAS, VWL, RBD, EPM, ERM , LHN); Department of Epidemiology, Harvard School of Public Health, Boston, MA and Channing Division of Network 2 Affiliations of authors: Division of General Medicine and Primary Care, Department of Medicine, Harvard Medical School, Beth Israel Deaconess Medical Center 3 Epidemiology, University of Massachusetts , Amherst, MA, SEH , USA 4 Torrance, CA (RTC); Department of Epidemiology & Population Health, Albert Einstein College of Medicine, Bronx, NY (TER); Department of Biostatistics 5 Family and Preventive Medicine, University of California San Diego, La Jolla, CA (AZL); Los Angeles Biomedical Research Institute at Harbor-UCLA Medical Center , USA 6 26. Spiegelman D, Colditz GA, Hunter D, Hertzmark E. Validation of the Gail 29. Cook NR, Rosner BA, Hankinson SE, et al. Mammographic screening Background: The Breast Cancer Risk Assessment Tool (BCRAT, “Gail model”) is commonly used for breast cancer prediction; however, it has not been validated for women age 75 years and older. Methods: We used Nurses' Health Study (NHS) data beginning in 2004 and Women's Health Initiative (WHI) data beginning CI = 1.18 to 1.45, age ≥ 75 years,P = .02), and in WHI 1.03 (95% CI = 0.97 to 1.09, age 55-74 years) and 1.10 (95% CI = 1.00 to 1.21, age ≥ 75 years, P = .21). E/O ratio 95% confidence intervals crossed one among women age 75 years and older when Conclusions: BCRAT accurately predicted breast cancer for women age 75 years and older who underwent mammography © The Author 2015. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: . Older - A W model by age. All statistical tests were two-sided. Results: Seventy-three thousand seventy-two NHS and 97 081 WHI women participated. NHS participants were more likely to be non-Hispanic white (96.2% vs 84.7% in WHIP, < .001) and were less likely to develop breast cancer (1.8% vs 2.0%, P = .02). E/O ratios by age in NHS were 1.16 (95% confidence interval [CI] = 1.09 to 1.23, age 57–74 years) and 1.31 (95% between 0.56 and 0.58 in both cohorts regardless of age. Women age 75 years and older are the fastest growing segment mammography screening included women age 75  years and of the US population, and breast cancer incidence increases older. Therefore, it is not known if screening helps these women with age ( 1 ). However, none of the randomized trials evaluating live longer 2(). Ideally, screening decisions would consider an Received: February 9, 2015;Revised: June 17, 2015; Accepted: October 20, 2015 1 of 11 life breast cancer risk4(). Therefore, we aimed to assess BCRAT’s performance among women age 75 years and older compared with that in postmenopausal women age 55 to 74 years part-ici pating in the Nurses’ Health Study (NHS) and Women’s Health Initiative (WHI). We chose these cohorts because they include many women age 75 years and older and have captured nece-s mine BCRAT’s accuracy for assessing breast cancer risk when deciding whether or not to screen women age 75 years and older for breast cancer. Methods BCRAT BCRAT estimates the probability that a woman will develop pate in WHI-ES. Participants were between age 55 and 91 years invasive breast cancer in five years. To calculate risk, BCRAT considers a woman’s baseline hazard of developing breast ca-n cer based on her age and race using age (categorized into 5-year age groups) and race/ethnic-specific population breast c-an cer incidence rates from the National Cancer Institute (NCI)’s Surveillance, Epidemiology, and End Results (SEER) program. Outcomes BCRAT uses SEER incidence rates for non-Hispanic whites from 1983 to 1987, for non-Hispanic Blacks from 1994 to 1998, for Hispanics from 1990 to 1996, and for Asians from 1998 to 2002. The model then adjusts a woman’s risk by considering whether entry into the study varied. We excluded women who were alive in 2004 but did not complete a 2004 questionnaire (n = 12 539) because ascertainment of breast cancer may be incomplete for these women. We chose to begin following NHS participants in 2004 because this time period is similar to that of WHI-ES and is two years after publication of WHI’s estrogen plus p-ro increased breast cancer risk1(6). Breast cancer risk associated with E+P was found to rapidly decline within two years of d-is continuation 1(5). Participants ranged in age from 57 to 86 years at study entry. WHI-ES Participants entered our study the day they consented to par-tici at study entry. In primary analyses, we excluded women with a history of cancer (except nonmelanomatous skin cancer) in both cohorts (n = 9394 in NHS, n = 9812 in WHI) because NHS does not confirm second diagnoses of cancer. We followed participants for up to five years or until they de-vel oped invasive breast cancer or died, whichever came first. All WHI breast cancer cases were confirmed by pathology report. older woman’s breast cancer risk, life expectancy, and pref-er women age 50 to 79  years in up to three clinical trials (WHIences (3,4). While there are tools to help estimate life exp-ec CTs) or an observational study (WHI-OS) from 1993 to 1998 and tancy (5) and elicit patient preferences6(), little is known about initially followed women through March 2005. The majority of late-life breast cancer risk factors. participants (82% of WHI-CT participants and 73% of WHI-OS The Breast Cancer Risk Assessment Tool (BCRAT) is the most participants) agreed to an extension study (WHI-ES; n  =  115 commonly used breast cancer risk prediction model in primary 400)  through September 2010. We chose WHI-ES participants care; however, its performance among women age 75 years and for our analyses because many had aged past 75  years and older is not known 7(–9). BCRAT was developed from a statist-i most had stopped using hormone therapy, which is typical of cal model known as the Gail model, which was developed using current practice1(5). (BCDDP), a study that recruited 280 000 women age 35 to 74 between 1973 and 1980 (10,11). BCRAT considers a woman’s age, race/ethnicity, age at menarche, age at first live birth, number of NHS Sample first-degree relatives with breast cancer, number of benign breast Participants entered our study the month they returned biopsies, and history of atypical hyperplasia. However, several of these risk factors (eg, age at menarche) affect women’s estrogen their 2004 questionnaire (NHS measures dates in months). Questionnaires could be returned through May 2006 (4.5% of levels relatively early in life and may not be important for late-participants returned their 2004 questionnaire in 2006), so sary data on breast cancer risk factors. Our goal was to de-ter gesterone (E+P) clinical trial results that found that use of E+P risk factors included in the model are present or absent and the For NHS, we also included self-reported breast cancers (12% of relative risk estimate of each risk factor. The model also con-sid ers the amount of risk that can be explained by the risk factors breast cancers in NHS are accurate (99% are confirmed when cases) because validation studies have found that self-reported included (attributable risk). Finally, BCRAT considers a w-om an’s age based risk of death using National Center for Health Statistics data7(). All women in an age group are considered to have the same risk of death regardless of their health. The source data S(upplementary Table  ,1available online) and fo-r mula for BCRAT may be found in theSupplementary Materials (available online). Data medical records are obtained)1(7). Detailed definitions of covar-i ates and outcome variables are in thSeupplementary Materials We describe briefly below each cohort used in our ana-ly baseline breast cancer incidence for Asians we randomly ses (12–14); detailed descriptions are in theSupplementary assigned Asian ethnicity for the 0.83% of NHS participants that Materials(available online). NHS is a longitudinal study of were Asian using 1970 Census data (NHS began in 1976). While 121 700 female nurses, age 30 to 55  years in 1976, from 11 of BCRAT follows women through 89.999 years, it does not co-m the most populous US states1(4). At baseline and in biennial pute five-year probabilities for women older than age 85 years follow-ups, participants provide detailed lifestyle and medical at entry. Therefore, we reclassified women older than age history information through mailed questionnaires. WHI is a 85 years at study entry as being 85 (17 NHS and 1253 WHI-ES multicenter study that recruited 161 808 postmenopausal US participants were reclassified). Downloaded from by guest To further compare NHS and WHI-ES participants, we used (perfect discrimination)1(8). To test whether E/O ratio estimates data on participant mammography use in the past two years, and c-statistics differed by age within cohort, we used the n-or body mass index, oophorectomies, hormone therapy use, and mal approximation z-test. Our primary analyses were limited to history of significant illness (including diabetes, myocardial participants with complete data on BCRAT risk factors. infarction, emphysema, congestive heart failure, stroke, or peripheral artery disease). In NHS, all of these conditions were their medical records except for emphysema and heart failure. Because previous studies found that BCRAT performs be-t In WHI-ES, all conditions were physician-adjudicated with me-d ical records except for emphysema and diabetes. WHI-ES does not assess reasons for undergoing mammography; therefore, we present receipt of any mammogram in the past two years. Statistical Analyses ter among women who undergo mammography, we repeated our analyses limiting our sample to women who underwent mammography in the past two years1(0,22,23). Because we were interested in BCRAT’s performance among women age 75  years and older and comorbidity increases with age, we repeated our analyses excluding women with significant -ill nesses (defined above). In addition, we repeated our analyses We used chi-square tests to compare characteristics between limiting our sample to women who were recently screened and NHS and WHI-ES.Tests of statistical significance were two-sided, were without significant illness. We also repeated our analyses and a P value of less than .05 was considered to be statistically using SEER incidence rates from 2006 to 2010 for whites, blacks, significant. To examine BCRAT’s performance, we measured and Hispanics because these data matched our study period. BCRAT’s calibration and discrimination within our cohorts st-rat Because previous studies found that BCRAT performs better ified by age (55–74 vs 75+) (18). Calibration assesses whether a when predicting estrogen receptor–positive (ER+) breast cancer; model’s predicted probabilities are accurate. Discrimination we reassessed BCRAT’s discrimination using ER+ breast cancer assesses how well a model distinguishes between individuas our outcome (24). We also repeated our analyses including all als who do or do not experience the event of interest. A model participants regardless of missing data. that discriminates well will assign higher risk values to those In addition, we examined BCRAT’s performance among who develop the outcome of interest1(8). To assess calibration, women who originally participated in WHI-OS and WHI-CT we compared the expected (E) number of breast cancers based separately. Although BCRAT considers history of atypical hy-per on BCRAT estimates (calculated using BCRAT’s SAS macro7))( plasia, information on atypia was only available for 1155 NHS to the observed number (O) in each cohort overall and within participants. However, atypia was captured for 38 218 WHI-CT deciles of risk stratified by age (55–74, 75+ years). To determine participants during the trial and we repeated our analyses deciles of risk, we ordered the probabilities given by BCRAT for among these women adding history of atypia2(5). In addition, each age group within a cohort and categorized these pro-b we repeated our analyses including women in WHI-ES with a abilities into deciles. Within each decile, we took the average history of cancer (excluding breast cancer). probability of risk and multiplied this probability by the sample To examine if qualitative differences in E/O ratios and c-st-a size within each decile to determine the expected number of tistics resulting from our sensitivity analyses were statistically breast cancers (E). We calculated 95% confidence intervals (CIs) significant, we used bootstrapping to estimate the standard of E/O ratios using the Poisson variance for the logarithm of error of the difference and the z-statistics from which we co-m the observed number of cases (19). We further assessed calibr-a puted theP values. tion using the Hosmer-Lemeshow (HL) chi-square test. An E/O the actual risk) and a nonsignificant HL-test statistic indicate ratio close to 1 (meaning the model’s estimates of risk matches Relative Risks good calibration2(0). To assess BCRAT’s discrimination, we used We present the relative risks associated with each BCRAT risk Rosner and Glynn’s methods to determine the c-statistic or area factor from NHS and WHI-ES calculated using multivariable under the receiver operating characteristic curve and its sta-nd ard error2( 1 ). This area ranges from 0.5 (no discrimination) to 1.0 logistic regression. All analyses were completed using SAS s-ta tistical software, version 9.3 (SAS Institute Inc., NC). Women aged 55+ in the Nurses’ Health Study (NHS) or Women’s Health Initiative (WHI) Downloaded from by guest Breast cancer risk assessment tool risk factors Stroke, % Congestive heart failure, % Number of significant illnesses from above 0, % 1, % 2+, % Outcomes¶ Breast cancer diagnosed during study, % Died during study, % † All comparisons between NHS and WHI-ES overall were statistically significanPt<at.001 using chi-square statistics, except thPe value for the difference between breast cancer incidence wasP = .02. ‡ The youngest women in NHS at study entry were age 57 years. § Body mass index was based on nurse self-report in NHS and was measured in WHI. || In NHS, diabetes, myocardial infarction, peripheral artery disease, and stroke were confirmed by participants and/or adjudicated by review of their medical records. Congestive heart failure, myocardial infarction, peripheral artery disease, and stroke were physician adjudicated with medical records in WHI-ES. ¶ Participants were followed for five years. Results for the Hosmer-Lemeshow test among women age 75 years and In both cohorts, the 95% confidence intervals of the E/O ratios crossed one for women age 75 years and older when we limited missing data on age at first birth than WHI-ES participants our sample to women who underwent mammography and were (17.2%, P < .001). More NHS participants (17.4%) were missing without significant illness (E/O ratios decreased from 1.30, 95% data on recent mammography use than WHI-ES participants CI = 1.17 to 1.45, to 1.13, 95% CI = 0.98 to 1.29,P = .12 in NHS and (0.2%, P < .001); these NHS participants completed a shorter v-er sion of the 2004 questionnaire that did not assess mammogr-a phy use. Otherwise, there were small differences in missing data between cohorts on sample characteristics. WHI-ES participants dence rates led to statistically significantly increased E/O ratios were more racially/ethnically diverse (96.2% vs 84.7%P ,< .001), among women age 55 to 74 years in both cohorts but did not s-ta younger at first birth, more likely to have had a breast biopsy, tistically change E/O ratios among women age 75 years and older. mone replacement therapy. NHS participants were more likely able online). Among WHI-CT participants, including inform- a and to have a BMI of 30kg/m2 or more; they were less likely to have a family history of breast cancer and to have ever used h-or to have significant illness (22.1% vs 20.5%P, < .001) and to die during follow-up (6.8% vs 5.4%,P < .001; 15.3% vs 11.0% among women ≥75 years). Fewer NHS participants were diagnosed with breast cancer than WHI-ES participants (1.8% vs 2.0 %P ,= .02). Calibration Calibration graphsF(igure 2) demonstrate that BCRAT accurately categorized as being at the highest risk, where BCRAT ove-r predicted breast cancer riskS.upplementary Table  2(available online) presents E/O ratios and their 95% confidence intervals for each decile of risk stratified by age and cohort. BCRAT was more likely to overpredict breast cancer risk in NHS, part-icu larly for women categorized as being at higher risk and among women age 75  years and older T(able  2). E/O ratios on average by age in WHI-ES were 1.03 (95% CI  =  0.97 to 1.09) for women age 55 to 74 years and 1.10 (95% CI = 1.00 to 1.21) for women age Within WHI-ES, BCRAT’s calibration was similar between WHI-OS and WHI-CT participantsS(upplementary Table ,4availtion on history of atypia did not change BCRAT’s calibration (Supplementary Table ,4available online). Including women with a history of cancer also did not change calibration in WHI-ES 75 years and older P( = .21). In NHS, E/O ratios on average by age RRs for the interaction between family history and age at first were 1.16 (95% CI = 1.09 to 1.23) for women age 57 to 74 years birth tended to be higher in the BCRAT development cohort Downloaded from by guest a statistically significant breast cancer risk factor for WHI-ES orby BCDDP participants than NHS participants. Because ma-m for NHS participants age 75 years and older. Discussion mography may find breast cancer before symptoms develop and may even find some cancers that would never have caused problems, mammography may lead to an increased estimated risk among those who are screened 2(8,29). In the third analysis, In WHI-ES, BCRAT accurately predicted five-year breast cancer BCRAT was found to underestimate breast cancer by 6%, which risk for women age 55 to 74 years and overpredicted breast ca-n was attributed to greater use of mammography by NHS part-ici cer risk by 10% on average among participants age 75 years and pants than among SEER participants. Similar to our findings, older. In NHS, BCRAT overpredicted breast cancer risk by 16% on BRCAT’s c-statistic in the third analysis was 0.58 (95% CI = 0.56 average among women age 57 to 74 years and by 31% on average was most likely to overpredict breast cancer among women c-at participants beginning at enrollment (between 1993–19982)4(). egorized as being at higher risk. BCRAT’s prediction accuracy In that study, BCRAT was found to underestimate breast cancer improved among women age 75 years and older when the samby 20%, which was attributed to greater use of mammography ples were limited to women who underwent mammography and and breast biopsies among WHI participants than in SEER. That were without significant illness, suggesting that BCRAT may be study also found that BCRAT’s c-statistic was 0.58 and improved most appropriate to use among older women with these ch-ar to 0.60 when the outcome was limited to ER+ breast cancer. This acteristics. BCRAT’s discrimination was modest in both cohorts finding was attributed to the fact that older women are more and age groups, with c-statistics ranging between 0.56 and 0.58. likely to develop ER+ breast cancer30() and may explain why BCRAT’s performance has previously been tested among we also found that BCRAT tended to show better discrimi-na NHS participants using data from: 1)  197626(), 2)  1982 (26), tion among women age 75  years and older when predicting and 3) 1992 (27) (when participants were age 30–55, 36–61, and ER+ breast cancer. BCRAT’s performance has also been tested 45–71  years, respectively). In the first two analyses, the model in other cohorts of older women. In the National Institutes was used to predict invasive and noninvasive breast cancer and of Health–AARP Diet and Health Study (recruited women baseline breast cancer incidence was estimated using BCDDP age 50–71  years between 1995–1996) and the Prostate, Lung, data. In the third analysis, BCRAT was used to predict only in-va Colorectal, and Ovarian Cancer Screening trial (recruited women sive breast cancer and 1987 SEER incidence rates were used age 55–74  years between 1993–2001), BCRAT underpredicted to estimate baseline breast cancer incidence. In the first two breast cancer by 13% to 14%, which was attributed to use of Downloaded from by guest Age at first birth, y X 0,1,2+ first degree relatives (1.45 to 2.10) (1.41 to 1.84) (1.43 to 2.18) (1.33 to 1.82) (1.13 to 2.46) (1.38 to 2.26) * Nurses’ Health Study began in 2004. Women’s Health Initiative Extension Study began in 2005. BCDDP = Breast Cancer Demonstration and Detection Pro3j1e)c; t ( ‡ The youngest women in NHS in 2004 were age 57 years. § No separate indicator was used for nulliparous women. SEER incidence rates from 1983 to 19872( 1 ). BCRAT’s calibration have significant illness and to die. Competing mortality risks improved when 1995 to 2003 SEER incidence rates were used, may have prevented these women from undergoing mammogwhich corresponded to the time period of that study. BCRAT’s raphy and as a result these women would have been less likely c-statistic in these cohorts was 0.58 to 0.5292(,24). These studies to have an early-stage breast cancer detecteSdu(pplementary predicting breast cancer among postmenopausal women. our sample) remained alive during our study but did not co-m Interestingly, we found BCRAT’s calibration to be better in plete a follow-up questionnaire. Breast cancer may have been WHI-ES than NHS. This is likely because breast cancer incidence missed among these women. If we assume that 2.0% of these in WHI-ES (2.0%) was more similar to SEER S(upplementary women were diagnosed with breast cancer, then our overall lower in NHS (1.8%) because NHS participants were less likely participants that remained alive during our study period did not to undergo breast biopsy, were less likely to be obese, and were complete a follow-up questionnaire. more likely to undergo oophorectomy. Also, NHS participants, While we found that BCRAT provides accurate probab-ili particularly those age 75  years and older, were more likely to ties of five-year breast cancer risk for older women without Downloaded from by guest on late-life breast cancer risk. Our study has limitations. Family history of breast cancer and diagnosis of emphysema (for WHI-CT participants only) were obtained on average 8.5  years before start of the WHI-ES and therefore may be underestimated in WHI-ES. Although 17% when we repeated our analyses including these women our results did not changeS(upplementary Table ,8available online). In summary, BCRAT provides accurate five-year probabi-li ties of breast cancer among women age 75  years and older who undergo mammography and are without significant -ill ness but tends to overpredict breast cancer among postme-no pausal women categorized at the highest risk of breast cancer. Discrimination of the model is modest among all postmenop-au sal women. Incorporating other aspects of personalized breast cancer risk including lifelong estrogen exposure (eg, from obesity) Funding Notes and competing mortality risks may improve model performance. 9. Amir E, Freedman OC, Seruga B, et  al. Assessing women at high risk of This work was supported by the National Institute on Aging at the an NHS program project grant (P01 CA87969). 3. Breast cancer screening in older women. American Geriatrics Society Cl-ini cal Practice CommitteeJ.Am Geriatr Soc. 2000;48(7):842–844. 4. Walter LC, Schonberg MA. Screening mammography in older women: a breast cancer: a review of risk assessment modelsJ. Natl Cancer Inst. 2010;102(10):680–691. 10. Gail MH, Brinton LA, Byar DP, et al. Projecting individualized probabilities of developing breast cancer for white females who are being examined ann-u significant illness who undergo mammography, it has modest CA) Marcia L.  Stefanick; (The Ohio State University, Columbus, in BCRAT such as breast density has been shown to improve discrimination modestly 3( 1 ); unfortunately, our data do not life breast cancer (eg, obesity) may also be needed to improve BCRAT’s discrimination among older women. Also, the RRs ass-o ciated with some of the risk factors in BCRAT, such as family h-is The sponsor had no role in the design of the study; the co-l tory, may no long be accurate. Broad use of mammography has lection, analysis, or interpretation of the data; the writing of led to increased detection of early-stage breast cancers and may the manuscript; or the decision to submit the manuscript for have attenuated the impact of a family history of breast cancer publication. of WHI-ES participants were missing data on age at first birth, 2. The benefits and harms of breast cancer screening: an independent review. Geller; Clinical Coordinating Center: (Fred Hutchinson Cancer 23. Bondy ML, Lustbader ED, Halabi S, Ross E, Vogel VG. Validation of a breast 24. Chlebowski RT, Anderson GL, Lane DS, et al. Predicting risk of breast cancer Downloaded from by guest We would like to thank the participants and staff of the Nurses’ Health Study for their valuable contributions as well as th-e fol lowing state cancer registries for their help: AL, AZ, AR, CA, CO, 587. CT, DE, FL, GA, ID, IL, IN, IA, KY, LA, ME, MD, MA, MI, NE, NH, NJ, NY, NC, ND, OH, OK, OR, PA, RI, SC, TN, TX, VA, WA, WY. In addition, this study was approved by the Connecticut Department of 2002;288(3):321–333. 13. Hays J, Hunt JR, Hubbell FA, et al. The Women’s Health Initiative recruitment 14. Colditz GA, Manson JE, Hankinson SE. The Nurses’ Health Study: 20-year contribution to the understanding of health among womeJnW. omens Health. 15. Chlebowski RT, Kuller LH, Prentice RL, et al. Breast cancer after use of es-tro gen plus progestin in postmenopausal womenN. Engl J Med. 2009;360(6):573– 16. Rossouw JE, Anderson GL, Prentice RL, et  al. Risks and benefits of estr-o gen plus progestin in healthy postmenopausal women: principal results From the Women’s Health Initiative randomized controlled triJaAlM.A. data used in this publication were obtained from the DPH. 18. Justice AC, Covinsky KE, Berlin JA. Assessing the generalizability of progn-os 19. Daly L. Simple SAS macros for the calculation of exact binomial and Poisson 20. Hosmer DW, Hosmer T, Le Cessie S, Lemeshow S. A comparison of good ness-of-fit tests for the logistic regression Med. 1997;16(9):965– 21. Rosner B, Glynn RJ. Power and sample size estimation for the Wilcoxon rank sum test with application to comparisons of C statistics from alternative prediction models.Biometrics. 2009;65( 1 ):188–197. 22. Schonfeld SJ, Pee D, Greenlee RT, et al. Effect of changing breast cancer in-ci dence rates on the calibration of the Gail modeJlC.lin Oncol. 2010;28(14):2411– et  al. model for predicting individual breast cancer risJkN. atl Cancer Inst. 1432. by 1. Data on Seer (Surveillance, Epidemiology and End Results) cancer statistics . http://seer.cancer. gov. Accessed on September 10 , 2014 . 5. Yourman LC , Lee SJ , Schonberg MA , Widera EW , Smith AK . Prognostic indices for older adults: a systematic reviewJA . MA . 2012 ; 307 ( 2 ): 182 - 192 . 6. Schonberg MA , Hamel MB , Davis RB , et al. Development and evaluation of a decision aid on mammography screening for women 75 years and older . JAMA Intern Med . 2014 ; 174 ( 3 ): 417 - 424 . 7. Breast Cancer Risk Assessment SAS Macro (Gail Model) . http://dceg.cancer. gov/tools/risk-assessment/bcrasasmacro . Accessed September 10 , 2014 . 8. Elmore JG , Fletcher SW . The risk of cancer risk prediction: “What is my risk of getting breast cancer”?JNatl Cancer Inst . 2006 ; 98 ( 23 ): 1673 - 1675 . 25. Rohan TE , Negassa A , Chlebowski RT , et al. Conjugated equine estrogen and Natl Cancer Inst . 2008 ; 100 ( 8 ): 563 - 571 . and risk factors for breast cancerA . m J Epidemiol . 2009 ; 170 ( 11 ): 1422 - prevention. J Natl Cancer Inst . 2001 ; 93 ( 5 ): 358 - 366 . 2010 ; 28 ( 12 ): 2038 - 2045 . 27. Rockhill B , Spiegelman D , Byrne C , Hunter DJ , Colditz GA . Validation of the 30 . Schonberg MA , Marcantonio ER , Li D , et  al. Breast cancer among the o-ld est old: tumor characteristics, treatment choices, and survivJaClli . n Oncol . 28 . Schonberg MA , McCarthy EP , Davis RB , Phillips RS , Hamel MB . Breast cancer 31. Chen J , Pee D , Ayyagari R et al. Projecting absolute invasive breast cancer risk screening in women aged 80 and older: results from a national surveJyA . m Geriatr Soc . 2004 ; 52 ( 10 ): 1688 - 1695 . Cancer Inst . 2006 ; 98 ( 17 ): 1215 - 1225 .

This is a preview of a remote PDF:

Schonberg, Mara A., Li, Vicky W., Eliassen, A. Heather, Davis, Roger B., LaCroix, Andrea Z., McCarthy, Ellen P., Rosner, Bernard A., Chlebowski, Rowan T., Rohan, Thomas E., Hankinson, Susan E., Marcantonio, Edward R., Ngo, Long H.. Performance of the Breast Cancer Risk Assessment Tool Among Women Aged 75 Years and Older, JNCI: Journal of the National Cancer Institute, 2016, DOI: 10.1093/jnci/djv348