Part II · Normal biology, carcinogenesis, and inherited susceptibility · Chapter 8

Risk assessment and prevention

The drugs work. Roughly one eligible woman in six takes them, and that is the problem worth solving.

1 · Individual and population risk models and their calibration

A risk model answers a question, and different models answer different questions. Most confusion in this area comes from asking a model something it was not built to estimate.

Two properties of a model have to be judged separately, and they can move in opposite directions.

Calibration asks whether the predicted number of cancers matches the observed number. It is what makes a threshold such as a 1.66% five-year risk mean anything, because eligibility rules are built on absolute numbers.

Discrimination asks whether the model ranks individuals correctly. It is measured by a concordance statistic, and for breast cancer risk models it is modest. In a prospective family cohort of 15,732 women followed for a median of 11.1 years, during which 619 developed breast cancer, concordance statistics ranged from 0.60 to 0.71 across four widely used models1.

Those numbers deserve a plain reading. A model that is well calibrated at the population level can still order individuals only a little better than chance. That is not a reason to abandon models. They succeed at allocating an intervention across a population. They fail at telling one woman what will happen to her.

Population risk and individual risk also diverge in a way that changes policy. Most breast cancers occur in women who were never in a high-risk category, because that category is small. A programme that acts only on the top few percent of predicted risk will therefore prevent a minority of the disease, however well it is targeted.

2 · Gail, Tyrer-Cuzick, BOADICEA and CanRisk compared

Each of these models was built for a specific population and a specific decision. Reading the design intent explains most of the disagreement between them.

The Gail model estimates absolute invasive breast cancer risk from age at menarche, age at first live birth, number of previous breast biopsies and number of affected first-degree relatives2. It was derived from a screening programme cohort, and the original paper says the projections are most reliable for women who are examined about annually. It contains almost no family structure, so it cannot represent a strong pedigree. Its clinical role is eligibility screening for prevention, and that is the role it was built for.

It was also derived in white women, and that limitation was addressed rather than ignored. A model rebuilt on data from 1,607 African American women with invasive breast cancer and 1,647 controls gave higher risk estimates than the original tool3. Applied to screening data from a prevention trial, 30.3% of African American women met a five-year risk of at least 1.66% by the new model against 14.5% by the original. An eligibility threshold is only as good as the model feeding it.

Tyrer-Cuzick was built to combine a detailed pedigree with personal risk factors. It uses Bayesian inference across the family history to estimate the probability of carrying a predisposing variant, then refines the estimate with hormonal and reproductive factors4. It is the appropriate choice when family history is substantial and genetic testing has not been done.

BOADICEA was built as a genetic model and has since absorbed everything else. The current version incorporates truncating variants in BRCA1, BRCA2, PALB2, CHEK2 and ATM, a 313-variant polygenic score, a residual polygenic component, lifestyle and reproductive factors, and mammographic density5. Across the UK population it predicts lifetime risks from 2.8% at the first percentile to 30.6% at the ninety-ninth. It classifies 14.7% of women as moderate risk by the 17% to 30% lifetime band, and 1.1% as high risk at 30% or above.

CanRisk is the delivery mechanism for that model, not a separate model. It is the web interface carrying BOADICEA, with a pedigree builder, and it was the first freely accessible cancer risk prediction tool to carry a CE marking6. The distinction matters when reading a paper, because a study of CanRisk is a study of BOADICEA plus an interface.

Independent validation ranks them consistently. In the prospective family cohort, the ratio of expected to observed cancers was 1.05 for BOADICEA and 1.03 for Tyrer-Cuzick1. Both were well calibrated. The ratio was 0.79 for the Gail-derived tool and 0.59 for BRCAPRO, both of which underpredicted. Concordance statistics were 0.70 for BOADICEA and 0.71 for Tyrer-Cuzick, against 0.60 for the Gail-derived tool. Models that use multigenerational family history performed better, even in women at average risk.

In practice

Pick the model by the question. For eligibility for pharmacologic prevention in a woman without notable family history, a Gail-type five-year absolute risk is the established input and the one prevention trials used7.

For a woman with a meaningful pedigree, use Tyrer-Cuzick or CanRisk. A Gail estimate in that setting will understate risk because the model has nowhere to put the family history1.

Check whether the model you are using was validated in a population like your patient. The Gail-type tool required rebuilding for African American women, and the rebuilt version doubled the proportion meeting the prevention threshold3.

Record which model produced the number. A risk figure without its model is uninterpretable, and it will be reused by the next clinician who reads the note.

3 · Benign breast disease and risk lesions as risk modifiers

Benign breast disease is common, and its risk implications are graded rather than binary.

A cohort of 9,087 women biopsied for benign disease was followed for a median of 15 years8. Relative risk of subsequent breast cancer against a population registry rose across three histologic groups: 1.27 for nonproliferative lesions (95% CI 1.15 to 1.41), 1.88 for proliferative lesions without atypia (95% CI 1.66 to 2.12), and 4.24 for atypical hyperplasia (95% CI 3.26 to 5.41). Nonproliferative findings accounted for 67% of that cohort. The earlier Nashville series gave the same ordering, with a relative risk of 5.3 for atypia against nonproliferative findings (95% CI 3.1 to 8.8)9.

Atypia is where the risk becomes management-relevant. Among 331 women with atypia followed for a mean of 13.7 years, 66 developed breast cancer, and cumulative incidence approached 35% at 30 years10. Multifocal atypia with calcifications carried a relative risk of 10.35 (95% CI 6.13 to 16.4), and women under 45 at diagnosis carried a relative risk of 6.76 (95% CI 3.24 to 12.4). Risk was similar for atypical ductal and atypical lobular hyperplasia. The management options that follow from this risk level are set out in the accompanying review11.

Caution

Family history did not add significant further risk in women who already had atypia10. This is counterintuitive and it is clinically consequential, because most risk models will add the two. A woman with atypia is already in a high-risk group, and the pedigree does not move her much further within it.

The risk from atypia is bilateral, not confined to the biopsied breast. That is the observation that makes atypia a marker of a susceptible field rather than an obligate precursor, and Hyperplasia, atypia, and the limits of a linear continuum develops the argument. It also means that unilateral surgery is a poor response to it.

4 · Breast density and other risk modifiers

Density is discussed as a property of the breast and measured as a property of an image. Breast density, what it is and why it predicts sets out what the measurement is. This section is about what it does to a risk estimate.

The association is one of the strongest in the field. A meta-analysis of 42 studies aggregated more than 14,000 cases and 226,000 non-cases12. Relative risks of incident breast cancer in the general population rose across four percent density categories, each compared against density below 5%: 1.79 for 5% to 24%, 2.11 for 25% to 49%, 2.92 for 50% to 74%, and 4.64 for 75% or more. The association was much stronger for percent density than for categorical parenchymal pattern scores, which is a statement about measurement rather than about biology.

Masking and independent risk have to be separated, and nested case-control data separate them. Women with density in 75% or more of the mammogram had an odds ratio of 4.7 against women below 10%13. The odds ratio was 3.5 for screen-detected cancers and 17.8 for cancers appearing within 12 months of a negative screen. The excess risk persisted for at least 8 years, which masking alone cannot explain.

Supplemental imaging follows from the masking component rather than from the risk component, and that policy question belongs in Dense breasts, supplemental imaging, and notification policy with the DENSE data14.

Two other modifiers belong in a risk model rather than in a lifestyle discussion. Endogenous hormone levels predict postmenopausal risk, with a relative risk of 2.00 in the highest quintile of serum oestradiol against the lowest (95% CI 1.47 to 2.71) in a reanalysis of nine prospective studies15. And the genetic architecture of an individual tumour relates to how it was detected, so mode of detection is not independent of biology16.

The modern models absorb these rather than leaving them to judgement. BOADICEA takes mammographic density and a polygenic score as inputs alongside family history5, which is why a density result now changes a computed risk rather than a clinical impression.

5 · Lifestyle, metabolic, and environmental prevention

This section is short because the evidence is thinner than the volume of advice suggests.

Alcohol has the clearest dose relationship. A collaborative reanalysis pooled 53 studies including 58,515 women with breast cancer and 95,067 without17. Relative risk rose by 7.1% for each additional 10 g of alcohol per day (95% CI 5.5 to 8.7). Relative risk was 1.32 at an intake of 35 to 44 g per day and 1.46 at 45 g or more, each against no alcohol. The same analysis estimated that about 4% of breast cancers in developed countries are attributable to alcohol. Restricting the analysis to 22,255 women with breast cancer who reported no alcohol showed no independent effect of smoking.

Dietary intervention has been tested prospectively and the result is modest. Long-term follow-up of the Women's Health Initiative dietary modification trial found a reduction in breast cancer mortality with a low-fat dietary pattern18. That is an effect on mortality after diagnosis rather than a demonstrated large effect on incidence.

Metabolic intervention has been tested directly and failed. Metformin did not improve invasive disease-free survival against placebo in early breast cancer19. Obesity remains associated with worse outcomes, including differential benefit from endocrine agents by body mass index20, but an association of that kind does not establish that intervening on weight reverses it.

Environmental exposures are where claims most outrun data. Night shift work is a worked example of how the evidence moves. Analysis of three prospective cohorts, combined with a meta-analysis of prior publications, found no association between night shift work and breast cancer incidence21. Earlier positive findings had come largely from retrospective designs.

Caution

The honest summary is that lifestyle modification has a real but small effect on breast cancer incidence, and the interventions with large effects are pharmacologic and surgical. Presenting lifestyle advice as the main preventive option is comfortable and it misallocates the conversation. The options with the largest demonstrated effect are in Pharmacologic risk reduction and its persistent underuse and Risk-reducing surgery and the decision calculus.

6 · Pharmacologic risk reduction and its persistent underuse

The evidence here is unusually strong for a preventive intervention, and it is almost unused. The 2009 ASCO update found that tamoxifen and, after the menopause, raloxifene reduce risk, and did not recommend aromatase inhibitors or fenretinide outside a trial22.

Tamoxifen was tested in 13,388 women at increased risk. It reduced invasive breast cancer by 49%, with cumulative incidence through 69 months of follow-up of 43.4 per 1,000 women on placebo against 22.0 per 1,000 on tamoxifen7. Oestrogen receptor positive tumours fell by 69%, and receptor-negative tumours did not fall at all. Endometrial cancer rose, with a risk ratio of 2.53 (95% CI 1.35 to 4.97), and thromboembolic events rose, both predominantly in women aged 50 or older.

The durability is what makes the benefit-to-harm calculation work. In 7,154 women followed for a median of 16.0 years, breast cancer occurred in 7.0% of the tamoxifen group against 9.8% of the placebo group23. The hazard ratio was 0.71 (95% CI 0.60 to 0.83). Protection in the second decade was as large as in the first, with a hazard ratio of 0.69 after 10 years. Five years of treatment buys at least sixteen years of protection, while the excess of serious adverse events is concentrated during treatment.

Aromatase inhibitors are more effective and avoid the endometrial and thrombotic signal. In 3,864 postmenopausal women at increased risk, anastrozole for 5 years reduced breast cancer with a hazard ratio of 0.47 at a median of 5.0 years (95% CI 0.32 to 0.68)24. Long-term follow-up at a median of 131 months showed 85 cancers on anastrozole against 165 on placebo, a hazard ratio of 0.51 (95% CI 0.39 to 0.66)25. Benefit continued after treatment stopped, with no excess of fractures or cardiovascular disease. Exemestane gave a consistent result in 4,560 women, with 11 invasive cancers against 32 on placebo and a hazard ratio of 0.35 (95% CI 0.18 to 0.70)26.

Dose has been addressed, because toxicity is the stated barrier. In 500 women with breast intraepithelial neoplasia, tamoxifen at 5 mg daily for 3 years halved neoplastic events against placebo27. There were 14 events against 28, a hazard ratio of 0.48 (95% CI 0.26 to 0.92). The five-year number needed to treat was 22 (95% CI 20 to 27). Contralateral events fell by 75%. Patient-reported outcomes did not differ between arms except for slightly more frequent daily hot flushes.

Now the failure. A systematic review and meta-analysis of 26 studies covering 21,423 women at increased risk found a pooled uptake of preventive therapy of 16.3% (95% CI 13.6 to 19.0)28. Uptake was 25.2% within trials and 8.7% outside them. Uptake was higher after an abnormal biopsy, at higher objective risk, and where a physician had recommended it. Only one study reported persistence of at least 80% at 5 years.

In practice

Physician recommendation is among the strongest modifiable determinants of uptake28. The decision is often framed as a patient preference problem. It is at least as much an offer problem.

Offer the aromatase inhibitor to postmenopausal women. It reduces incidence more than tamoxifen does and carries neither the endometrial nor the thrombotic excess25.

Quote the benefit as absolute numbers over a defined horizon, and quote the harms the same way. A 49% relative reduction and a fall from 43.4 to 22.0 cancers per 1,000 women are the same finding, and they are heard differently7.

For a woman who declines because of toxicity, low-dose tamoxifen is an evidence-based alternative rather than a compromise, at least in intraepithelial neoplasia27.

Say that receptor-negative disease is not prevented by any of these agents7. A woman whose family history suggests triple-negative disease should hear that before she starts.

7 · Risk-reducing surgery and the decision calculus

Bilateral risk-reducing mastectomy produces the largest risk reduction available, and the size of the effect is not in doubt.

One series followed 639 women with a family history who underwent bilateral prophylactic mastectomy at a median age of 42, for a median of 14 years29. In the moderate-risk group, 4 breast cancers occurred against 37.4 predicted by a risk model, a reduction of 89.5%. Among high-risk probands, breast cancer occurred in 3 of 214, which is 1.4%. It occurred in 156 of their 403 sisters who had not had the surgery.

In carriers specifically, breast cancer occurred in 2 of 105 women after bilateral prophylactic mastectomy, which is 1.9%30. It occurred in 184 of 378 matched carriers who did not have the operation, which is 48.7%, over a mean follow-up of 6.4 years. Risk reduction was approximately 90% in women with intact ovaries and approximately 95% in women with prior or concurrent oophorectomy. A larger prospective cohort of 2,482 carriers found no breast cancers among 247 women after risk-reducing mastectomy31.

Salpingo-oophorectomy is the decision with a mortality signal behind it. In the same cohort, risk-reducing salpingo-oophorectomy was associated with lower ovarian cancer risk31. It was associated with lower risk of a first breast cancer in BRCA1 carriers (hazard ratio 0.63, 95% CI 0.41 to 0.96) and in BRCA2 carriers (hazard ratio 0.36, 95% CI 0.16 to 0.82). All-cause mortality was lower (hazard ratio 0.40, 95% CI 0.26 to 0.61), as were breast cancer specific mortality and ovarian cancer specific mortality.

The asymmetry between the two operations is the part to carry into the consultation. Mastectomy has the larger effect on breast cancer incidence and no demonstrated effect on overall mortality in these cohorts. Oophorectomy has a smaller effect on breast cancer incidence and an association with lower all-cause mortality. They are not interchangeable, and a patient who understands only the incidence figures will weigh them wrongly.

Three cautions belong with any of these numbers. All of these data are observational, because no randomised trial of risk-reducing mastectomy exists or will. Women who choose surgery differ from women who do not. And the follow-up in the carrier series is short relative to a lifetime of risk, which matters because the comparator group's risk keeps accruing.

Caution

Contralateral risk-reducing mastectomy after a unilateral cancer is a different decision from bilateral surgery in an unaffected carrier, and the evidence above does not transfer to it. In a woman whose prognosis is governed by the cancer she already has, removing the other breast reduces a second cancer without changing the risk that is actually driving her outcome. The exception is the carrier with early-stage disease and a long expected survival, where the contralateral risk is high and durable32.

8 · Prevention implementation and unequal access

Everything above works. The gap between efficacy and delivered benefit is where the preventable burden actually sits, and it is not distributed evenly.

Start with the size of the gap. Preventive endocrine therapy reduces breast cancer incidence by roughly half in the trials, and about one in six eligible women takes it28. No biologically plausible improvement in the drugs would produce the benefit that closing that gap would.

The gap widens along familiar lines, and each mechanism has been measured separately.

Risk assessment itself is unequal. The standard risk model required rebuilding for African American women, and the rebuilt version identified roughly twice the proportion meeting the prevention eligibility threshold3. Women assessed with a miscalibrated model are not offered what they qualify for.

Genomic risk stratification is unequal. A polygenic score derived in European-ancestry populations discriminated substantially worse in women of African ancestry, with an area under the curve of 0.571 against 0.63033,34. Expanding gene panels widens the disparity in uninterpretable results rather than narrowing it35.

Access to the pathway is unequal at every subsequent step. Time from an abnormal screening mammogram to biopsy differs by multiple factors operating at patient, facility and area level36. Time from diagnosis to surgery differs by race even within an equal-access health system, and the difference tracks survival37. Residence in historically redlined districts is associated with worse breast cancer outcomes38, and cumulative physiological stress burden is associated with all-cause mortality after a breast cancer diagnosis39. Structural racism is a measured exposure in this literature rather than a framing device40.

Trial participation is unequal, which feeds back into the evidence base. Structural, clinical and physician-level barriers account for most of the shortfall in enrolment among under-represented groups, rather than patient refusal41.

In practice

Audit the offer, not the uptake. A clinic that records how many eligible women were offered preventive endocrine therapy, rather than how many accepted it, will find the modifiable step28.

Check which risk model your service uses and in whom it was validated. A single default model applied to every patient is a silent source of inequity3.

Treat the polygenic component of a risk estimate as ancestry-dependent, and say so when the estimate is given33. Disparities across the breast cancer continuum develops why this matters beyond prevention.

Measure the intervals your service controls, from abnormal screen to biopsy and from diagnosis to treatment. Those intervals differ by group within single institutions, and they are visible to audit36,37.

The concluding claim of this chapter is narrow and should be read strictly. Breast cancer prevention does not currently need a better drug. It needs the drugs that exist to be offered, and the models that select for them to work in everyone.

References

All 41 references
  1. Terry MB, Liao Y, Whittemore AS, et al. 10-year performance of four models of breast cancer risk: a validation study. Lancet Oncol 2019 20:504-517. PMID 30799262
  2. 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 81:1879-1886. PMID 2593165
  3. Gail MH, Costantino JP, Pee D, et al. Projecting individualized absolute invasive breast cancer risk in African American women. J Natl Cancer Inst 2007 99:1782-1792. PMID 18042936
  4. Tyrer J, Duffy SW, Cuzick J. A breast cancer prediction model incorporating familial and personal risk factors. Stat Med 2004 23:1111-1130. PMID 15057881
  5. Lee A, Mavaddat N, Wilcox AN, et al. BOADICEA: a comprehensive breast cancer risk prediction model incorporating genetic and nongenetic risk factors. Genet Med 2019 21:1708-1718. PMID 30643217
  6. Carver T, Hartley S, Lee A, et al. CanRisk Tool: a web interface for the prediction of breast and ovarian cancer risk and the likelihood of carrying genetic pathogenic variants. Cancer Epidemiol Biomarkers Prev 2021 30:469-473. PMID 33335023
  7. Fisher B, Costantino JP, Wickerham DL, et al. Tamoxifen for prevention of breast cancer: report of the National Surgical Adjuvant Breast and Bowel Project P-1 Study. J Natl Cancer Inst 1998 90:1371-1388. PMID 9747868
  8. Hartmann LC, Sellers TA, Frost MH, et al. Benign breast disease and the risk of breast cancer. N Engl J Med 2005 353:229-237. PMID 16034008
  9. Dupont WD, Page DL. Risk factors for breast cancer in women with proliferative breast disease. N Engl J Med 1985 312:146-151. PMID 3965932
  10. Degnim AC, Visscher DW, Berman HK, et al. Stratification of breast cancer risk in women with atypia: a Mayo cohort study. J Clin Oncol 2007 25:2671-2677. PMID 17563394
  11. Hartmann LC, Degnim AC, Santen RJ, Dupont WD, Ghosh K. Atypical hyperplasia of the breast, risk assessment and management options. N Engl J Med 2015 372:78-89. PMID 25551530
  12. McCormack VA, dos Santos Silva I. Breast density and parenchymal patterns as markers of breast cancer risk: a meta-analysis. Cancer Epidemiol Biomarkers Prev 2006 15:1159-1169. PMID 16775176
  13. Boyd NF, Guo H, Martin LJ, et al. Mammographic density and the risk and detection of breast cancer. N Engl J Med 2007 356:227-236. PMID 17229950
  14. Bakker MF, de Lange SV, Pijnappel RM, et al. Supplemental MRI screening for women with extremely dense breast tissue. N Engl J Med 2019 381:2091-2102. PMID 31774954
  15. Key T, Appleby P, Barnes I, Reeves G. Endogenous sex hormones and breast cancer in postmenopausal women: reanalysis of nine prospective studies. J Natl Cancer Inst 2002 94:606-616. PMID 11959894
  16. Li J, Ugalde-Morales E, Wen WX, et al. Differential burden of rare and common variants on tumor characteristics, survival, and mode of detection in breast cancer. Cancer Res 2018 78:6329-6338. PMID 30385609
  17. Collaborative Group on Hormonal Factors in Breast Cancer. Alcohol, tobacco and breast cancer: collaborative reanalysis of individual data from 53 epidemiological studies, including 58,515 women with breast cancer and 95,067 women without the disease. Br J Cancer 2002 87:1234-1245. PMID 12439712
  18. Chlebowski RT, Aragaki AK, Anderson GL, et al. Dietary modification and breast cancer mortality: long-term follow-up of the Women's Health Initiative randomized trial. J Clin Oncol 2020 38:1419-1428. PMID 32031879
  19. Goodwin PJ, Chen BE, Gelmon KA, et al. Effect of metformin vs placebo on invasive disease-free survival in patients with breast cancer: the MA.32 randomized clinical trial. JAMA 2022 327:1963-1973. PMID 35608580
  20. Sestak I, Distler W, Forbes JF, Dowsett M, Howell A, Cuzick J. Effect of body mass index on recurrences in tamoxifen and anastrozole treated women: an exploratory analysis from the ATAC trial. J Clin Oncol 2010 28:3411-3415. PMID 20547990
  21. Travis RC, Balkwill A, Fensom GK, et al. Night shift work and breast cancer incidence: three prospective studies and meta-analysis of published studies. J Natl Cancer Inst 2016 108:djw169. PMID 27758828
  22. Visvanathan K, Chlebowski RT, Hurley P, et al. American society of clinical oncology clinical practice guideline update on the use of pharmacologic interventions including tamoxifen, raloxifene, and aromatase inhibition for breast cancer risk reduction. J Clin Oncol 2009 27:3235-58. PMID 19470930
  23. Cuzick J, Sestak I, Cawthorn S, et al. Tamoxifen for prevention of breast cancer: extended long-term follow-up of the IBIS-I breast cancer prevention trial. Lancet Oncol 2015 16:67-75. PMID 25497694
  24. Cuzick J, Sestak I, Forbes JF, et al. Anastrozole for prevention of breast cancer in high-risk postmenopausal women (IBIS-II): an international, double-blind, randomised placebo-controlled trial. Lancet 2014 383:1041-1048. PMID 24333009
  25. Cuzick J, Sestak I, Forbes JF, et al. Use of anastrozole for breast cancer prevention (IBIS-II): long-term results of a randomised controlled trial. Lancet 2020 395:117-122. PMID 31839281
  26. Goss PE, Ingle JN, Ales-Martinez JE, et al. Exemestane for breast-cancer prevention in postmenopausal women. N Engl J Med 2011 364:2381-2391. PMID 21639806
  27. DeCensi A, Puntoni M, Guerrieri-Gonzaga A, et al. Randomized placebo controlled trial of low-dose tamoxifen to prevent local and contralateral recurrence in breast intraepithelial neoplasia. J Clin Oncol 2019 37:1629-1637. PMID 30973790
  28. Smith SG, Sestak I, Forster A, et al. Factors affecting uptake and adherence to breast cancer chemoprevention: a systematic review and meta-analysis. Ann Oncol 2016 27:575-590. PMID 26646754
  29. Hartmann LC, Schaid DJ, Woods JE, et al. Efficacy of bilateral prophylactic mastectomy in women with a family history of breast cancer. N Engl J Med 1999 340:77-84. PMID 9887158
  30. Rebbeck TR, Friebel T, Lynch HT, et al. Bilateral prophylactic mastectomy reduces breast cancer risk in BRCA1 and BRCA2 mutation carriers: the PROSE Study Group. J Clin Oncol 2004 22:1055-1062. PMID 14981104
  31. Domchek SM, Friebel TM, Singer CF, et al. Association of risk-reducing surgery in BRCA1 or BRCA2 mutation carriers with cancer risk and mortality. JAMA 2010 304:967-975. PMID 20810374
  32. Kuchenbaecker KB, Hopper JL, Barnes DR, et al. Risks of breast, ovarian, and contralateral breast cancer for BRCA1 and BRCA2 mutation carriers. JAMA 2017 317:2402-2416. PMID 28632866
  33. Du Z, Gao G, Adedokun B, et al. Evaluating polygenic risk scores for breast cancer in women of African ancestry. J Natl Cancer Inst 2021 113:1168-1176. PMID 33769540
  34. Mavaddat N, Michailidou K, Dennis J, et al. Polygenic risk scores for prediction of breast cancer and breast cancer subtypes. Am J Hum Genet 2019 104:21-34. PMID 30554720
  35. Caswell-Jin JL, Gupta T, Hall E, et al. Racial/ethnic differences in multiple-gene sequencing results for hereditary cancer risk. Genet Med 2018 20:234-239. PMID 28749474
  36. Lawson MB, Bissell MCS, Miglioretti DL, et al. Multilevel factors associated with time to biopsy after abnormal screening mammography results by race and ethnicity. JAMA Oncol 2022 8:1115-1126. PMID 35737381
  37. Eaglehouse YL, Georg MW, Shriver CD, Zhu K. Racial differences in time to breast cancer surgery and overall survival in the US Military Health System. JAMA Surg 2019 154:e185113. PMID 30673075
  38. Plascak JJ, Beyer K, Xu X, Stroup AM, Jacob G, Llanos AAM. Association between residence in historically redlined districts indicative of structural racism and racial and ethnic disparities in breast cancer outcomes. JAMA Netw Open 2022 5:e2220908. PMID 35802373
  39. Obeng-Gyasi S, Elsaid MI, Lu Y, et al. Association of allostatic load with all-cause mortality in patients with breast cancer. JAMA Netw Open 2023 6:e2313989. PMID 37200034
  40. Bailey ZD, Krieger N, Agenor M, Graves J, Linos N, Bassett MT. Structural racism and health inequities in the USA: evidence and interventions. Lancet 2017 389:1453-1463. PMID 28402827
  41. Unger JM, Vaidya R, Hershman DL, Minasian LM, Fleury ME. Systematic review and meta-analysis of the magnitude of structural, clinical, and physician and patient barriers to cancer clinical trial participation. J Natl Cancer Inst 2019 111:245-255. PMID 30856272