Prevention of breast cancer in the context of a national breast screening programme

Authors


A. Howell, Genesis Prevention Centre and Nightingale Breast Screening Centre, University Hospital of South Manchester, Southmoor Road, Wythenshawe, Manchester, M23 9LT, UK.
(e-mail: anthony.howell@christie.nhs.uk).

Abstract

Abstract.  Howell A, Astley S, Warwick J, Stavrinos P, Sahin S, Ingham S, McBurney H, Eckersley B, Harvie M, Wilson M, Beetles U, Warren R, Hufton A, Sergeant J, Newman W, Buchan I, Cuzick J, Evans DG (Genesis Prevention Centre and Nightingale Breast Screening Centre, University Hospital of South Manchester; School of Cancer and Enabling Sciences, University of Manchester, Manchester; Centre for Cancer Prevention, Wolfson Institute of Preventive Medicine, Queen Mary University of London, London; School of Community Based Medicine, University of Manchester, Manchester; Genetic Medicine, Manchester Academic Health Sciences Centre, University of Manchester and Central Manchester Foundation Trust, Manchester; and Cambridge Breast Unit, Addenbrooke’s Hospital, Cambridge; UK). Prevention of breast cancer in the context of a national breast screening programme (Review). J Intern Med 2012; 271: 321–330.

Breast cancer is not only increasing in the West but also particularly rapidly in Eastern countries where traditionally the incidence has been low. The rise in incidence is mainly related to changes in reproductive patterns and lifestyle. These trends could potentially be reversed by defining women at greatest risk and offering appropriate preventive measures. A model for this approach was the establishment of Family History Clinics (FHCs), which have resulted in improved survival in younger women at high risk. New predictive models of risk that include reproductive and lifestyle factors, mammographic density and measurement of risk-associated single nucleotide polymorphisms (SNPs) may give more precise information concerning risk and enable better targeting for mammographic screening programmes and of preventive measures. Endocrine prevention using anti-oestrogens and aromatase inhibitors is effective, and observational studies suggest lifestyle modification may also be effective. However, referral to FHCs is opportunistic and predominantly includes younger women. A better approach for identifying older women at risk may be to use national breast screening programmes. Here were described pilot studies to assess whether the routine assessment of breast cancer risk is feasible within a population-based screening programme, whether the feedback and advice on risk-reducing interventions would be welcomed and taken up, and to consider whether the screening interval should be modified according to breast cancer risk.

Introduction

Breast cancer is the most common female cancer in many countries. In the UK, the age-adjusted rate for breast cancer rose from 80 to 100 per 100 000 women between 1985 and 2007 [1]. Breast cancer rates are rising particularly rapidly in countries with a traditionally low incidence such as Africa, Asia and parts of South America [2]. In Iceland, long-term precise record keeping has enabled accurate comparisons of breast cancer incidence between 1921–1944 and 1985–2002. During this time period, there was a fourfold increase, not only in the incidence of sporadic breast cancer, but also of the disease in carriers of the single founder Icelandic BRCA2 mutation (population incidence 0.5), indicating that environmental factors not only affect risk in the general population but also affect mutation penetrance [3].

The marked increases in breast cancer incidence are believed to be mainly related to changes in reproductive and lifestyle factors and the use of hormone replacement therapy (HRT). It is estimated that our female ancestors had about 160 ovulations over their reproductive period whereas recently it was estimated that women now have over 450 ovulations during reproductive life; thus, the risk of breast cancer presumed to be associated with cyclical activity of breast epithelial cell proliferation during the menstrual cycle is increased [4]. Lifestyle-related breast cancer risk factors include obesity and lack of physical activity, both of which are becoming increasingly prevalent. Huang et al. [5] reported in the Nurses Health Study that breast cancer risk doubled in women who gained 20 kg or more in weight between the ages of 20 and 50 years compared to women who were essentially weight stable; many other studies report similar relative risks. A recent overview of the contribution of body mass index (BMI) to the incidence of breast cancer estimated that that risk increased approximately 14% for each 5 kg m−2 gain weight in this population [6]. Compared with no physical activity, more than 4 h per week is estimated to reduce the risk of breast cancer by approximately 25% [7]. Thus, although it is not feasible to change reproductive patterns, risks may nevertheless be reduced by avoiding the use of HRT, blocking the tumour-promoting effects of endogenous hormones with agents such as tamoxifen and raloxifene, and sustained lifestyle changes will also probably be effective. Although we know that such measures are likely to be effective, it is not clear how best to identify a target population or whether risk-reducing interventions would be acceptable and sustainable within the population. Preventive approaches have been undertaken for some years in the context of clinics for younger women at risk, mainly because of a family history, and are a paradigm for targeting the much larger population of older women.

The Family History Clinic model

Family History Clinics (FHCs) were first set up in the 1980s in response to women’s increasing awareness of their genetic risk of breast cancer [8, 9]. Clinics generally offer risk information, mammographic and other screening and advice concerning the appropriateness of particular preventive interventions for women [8, 9]. Information is given concerning the probability of carrying a mutation in a cancer gene, offering genetic testing in the family if appropriate [10–14] and assessing overall risk of breast cancer for noncarriers produced by combining genetic and other risk factors by the use of, for example, models such as Gail et al. and Tyrer et al. [15, 16].

It is customary to offer annual screening by mammography in this group from age 35 or 40 although the effectiveness of this as regards reducing breast cancer mortality has never been assessed in randomized controlled trials. As a surrogate, we compared the survival of women undergoing 12–18 monthly mammography in our FHC and who developed breast cancer with patients in the same age range who presented symptomatically our surgical clinic over a 10-year period. After correction for lead time bias, survival [HR 0.24 (95% CI 0.08–0.43) P = 0.005] and disease-free survival [HR 0.25 (95% CI 0.11–0.57) P < 0.001] were significantly improved in the FHC compared with women who presented symptomatically to the same breast unit [17]. More recently, in a multicentre study in 76 FHCs in the UK, the outcome of 6710 women who had annual mammograms aged 40–49 were compared with 106 971 women aged between 40 and 42 from the general population at average risk who were followed through their forties in the UK Age Trial and a Dutch study (cancer cases between 25 and 77 years of women with a family history of breast cancer). Compared with the two control groups, screened women were more likely to have small node negative tumours. The predicted 10-year deaths from breast cancer was 1.1% in the FHC controls compared with 1.38% in the controls from the UK Age Trial. The relative risk reduction was 0.80 (95% CI 0.66–0.96, P = 0.022) and suggests that annual mammography in FHCs is likely to prevent deaths from breast cancer [18]. Other studies in BRCA1 and BRCA2 carriers indicate that risk-reducing oophorectomy and mastectomy are also associated with improved survival compared with no surgery groups [18, 19], although recent modelling [20] indicates that focussed screening by mammography and magnetic resonance imaging (MRI) may be as effective as bilateral prophylactic mastectomy. However, there must be some doubt about the efficacy of screening in BRCA1 carriers because even small node negative tumours have a poorer prognosis than other screen detected cancer [21].

Models of risk estimation used in Family History Clinics

Two types of risk are computed in FHCs. One is the probability of being a mutation carrier, which is estimated from family history [10–14] and the other the overall risk of developing breast cancer within a specific time period, which is calculated from family history and other established breast cancer risk factors [15, 16]. For example, the BOADICEA model developed in Cambridge and the Manchester Score predict the probability of a mutation based on the number of breast, ovary and other relevant cancers in the family to determine the need for mutation testing [11, 12]. Mutation prediction of these models have been improved more recently by adding histological details of diagnosed cancers [13, 14] In women without mutations, the model developed by Gail et al. [15] in 1989 and subsequently modified by Constantino et al. [22] is the most widely used model in the United States. The Gail model combines age, number of first degree relatives with breast cancer, number of breast biopsies and whether the biopsies contain epithelial atypia. The model was developed in three hundred thousand women undergoing breast screening in the Breast Cancer Detection Demonstration Project between 1973 and 1980 and validated in the Nurses Health Study [23]. The modified model focuses on invasive breast cancer only and is derived from the USA Surveillance Epidemiology and End Results Database and now includes incidence rates of African American patients [22].

The Tyrer–Cuzick Risk Prediction Model was developed to incorporate additional risk factors compared with Gail in an attempt to predict individual risk more accurately and to be more relevant to European populations [16]. Based on information from the IBIS-1 Breast Cancer Prevention Trial and additional epidemiological data, the model includes age of breast cancer in a relative, ovarian cancer in the family, second-degree relatives with breast and/or ovarian cancer, BMI, age at menopause and use of HRT in addition to factors used in the Gail 1 and 2 models [15] (Fig. 1).

Figure 1.

Lifetime risk of breast cancer and risk during the period of screening in women with increased risk factors, population risk factors and minimal risk factors.

To choose a model for use in our FHC, we assessed how well each model predicted 52 cancers detected in 1933 women in our FHC with a mean follow-up of 5.27 years. The ratios (with 95% confidence intervals) of expected to observed cancers were 0.48 (0.37–0.64) for Gail, 0.56 (0.43–0.75) for Claus, 0.49 (0.37–0.65) for Ford and 0.81 (0.62–1.08) for Tyrer–Cuzick [24]. Thus, the Gail model underpredicted in our population as has been reported by other studies in Europe and the USA [25–27]. In our clinic, we found that the Tyrer–Cuzick model performs optimally but this result requires further validation in other high-risk populations and also in the general population. Further discussion of the merits of these and other risk prediction models may be found in two recent reviews [28, 29].

Addition of mammographic density to risk models

Mammographic breast density is a major risk factor for breast cancer. Density may be subjectively quantified visually by assessing the proportion of dense versus fatty tissue and is variously subdivided into between four and 21 categories (Fig. 2). Two recent meta-analyses reported a four- to five-fold increased risk of invasive cancer for women in the highest category of breast density compared with those in the lowest [30, 31] (Fig. 1). Cummings et al. [31] performed a meta-analysis of how much breast density added to the discriminative capacity of breast cancer risk models, based on standard risk factors, in studies involving a total of 12 754 women. The area under (AUC) the receiver operator characteristic curve (ROC. c-statistic) ranged from 0.62 to 0.66 compared with the widely reported AUC for the Gail model of between 0.58 and 0.62. In a study from California, a model incorporating Gail [32] and additional risk factors reported an AUC of 0.605 without density and 0.62 with density (measured visually by the four category BIRADS system). Tice et al. [33] used a simplified Gail model and reported that the AUC increased from 0.61 to 0.66 also using BIRADS density. Chen et al. [34] reported improved estimates of absolute risk when density assessed by the CUMULUS system of computer-aided dense area calculation was added to the assessment of the Gail model. Thus, these studies suggest that combining classical breast cancer risk factors and mammographic density in an appropriate model increases the discriminatory power of risk prediction, but by a relatively small magnitude.

Figure 2.

Screening interval related to risk assuming a constant pick up rate of four breast cancers per 1000 women screened. The interval for high risk is estimated at 15 months, for population risk is 36 months and for low risk, 70 months.

Addition of single nucleotide polymorphism results to risk models

Recent genome-wide association studies (GWAS) have identified various novel breast cancer susceptibility variants [35–38]. Each single nucleotide polymorphisms (SNP) is associated with a small increase or reduction in risk, and it is hoped that by combining individual SNP risks and as more SNPs are discovered, the overall estimates of breast cancer risk will be improved. Van Zitteren et al. [39] performed a meta-analysis of known breast cancer risk SNPs and modelled the effect on the AUC of ROC curves in a simulated population of 10 000 women. They identified 41 polymorphisms that were significantly associated with breast cancer risk and computed an AUC of 0.67, which is similar to the best risk assessment model with density as outlined above. Others have modelled how much small numbers of validated SNPs add to risk prediction models, mainly using the Gail model. For example, by adding seven SNPs, Gail reported an increase in the AUC from 0.607 to 0.632 [40]. Other studies using from 7 to 11 SNPs have also demonstrated small increases in the AUC compared with risk models alone [40–43]. Park et al. [44] modelled the effect of 67 susceptibility SNPs used alone and estimated AUC of 0.635 and commented that risk models based on common variants appear to have modest discriminatory power. These estimates are made assuming no genetic interaction between SNPs, which may or may not be true. If there is no interaction and the fact that there are now a reported additional 30 new SNPs (see Easton et al.– this issue), there is a potential to use these and other known SNPs to improve prediction further. A major question is whether combining the current standard risk models with density and also SNPs will improve prediction further with the hope that more precise prediction will allow us to target preventive measures more accurately (See reviews by Chatterjee et al. [45] ‘Predictng the future of genetic risk prediction’, and by Pharaoh et al. [46]). Ultimately, it is essential that any combined models are validated properly in appropriate populations of women before their widespread use.

Encouragingly, recent studies indicate no interaction between SNPs and other standard risk factors. Milne et al. [47] found no interaction between 12 SNPs and age at menarche, ever having a live birth, number of live births, age at first birth and BMI in data from 26 349 invasive breast cancer cases and up to 32 208 controls. In the UK Million Women Study, Travis et al. [48] reported no interaction between 12 SNPs and ten established breast cancer risk factors (age at menarche, parity, age at first birth, breastfeeding, menopausal status, age at menopause, use of HRT, BMI, height and alcohol consumption) after correction for multiple testing. In the Predicting Risk Of Cancer At Screening (PROCAS) study (see later), little or no interaction was found between high-risk women detected by the Tyer-Cuzick model, mammographic density and 18 SNPs (Evans et al. submitted). Clearly more work is needed to resolve the paradox that mammographic density and SNPs appear to add only a little to the discriminatory power of the Gail model whereas other studies, including our own, show little interaction between standard risk factors, mammographic density and SNPs.

Breast cancer prevention

The question arises about the effectiveness of currently available preventive measures and how we might target the appropriate at risk population more accurately. Two main preventive avenues of investigation have been undertaken: the use of endocrine blocking agents (mainly anti-oestrogen) and lifestyle change, particularly weight loss and exercise.

Endocrine prevention

Recent reviews summarize the data indicating the effectiveness of preventive therapy (chemo-prevention) of breast cancer [31, 49, 50]. The agents already shown to be effective include tamoxifen and raloxifene [51–58]. More recently, the aromatase inhibitor exemestane was reported to give greater risk reduction than placebo [53]; a further study of anastrozole versus placebo (IBIS II) is in progress [54]. Most prevention studies are performed in women at increased risk of breast cancer. For example, the NSABP1 (tamoxifen v placebo [55]) and STAR (tamoxifen v raloxifene [52]) trials stipulate that only women with a 5-year Gail risk of breast cancer of >1.66% should be entered into these studies. However, raloxifene was shown to be effective in women with population breast cancer risk but who were treated with raloxifene for osteoporosis [The MORE trial [56]) and for cardiovascular disease (The RUTH trial [57]).

Cuzick et al. [51] performed an overview of the four randomized, placebo-controlled trials of tamoxifen and reported an overall risk reduction of 38% [51]. In the IBIS I trial, the long-term effect of 5 years of tamoxifen and a further 5 years of follow-up showed that the curves for placebo and treated groups continued to diverge so that there was a doubling of the preventive effect at 10 years compared with the effect determined at 5 years [58]. In the STAR trial, women were randomly assigned to receive either tamoxifen or raloxifene for 5 years [52]. The risk ratio (RR: raloxifene versus tamoxifen) for invasive breast cancer was 1.24 (95% CI, 1.05–1.47). The greater effectiveness of tamoxifen was associated with a higher incidence of side effects so that the risk benefit ratio favoured raloxifene in women with a uterus and was equivalent to tamoxifen in women without a uterus, indicating the important negative effect of tamoxifen on the endometrium [59].

Aromatase inhibitors are more effective for the prevention of relapse after breast cancer diagnosis compared with tamoxifen, and the early results of a comparison between exemestane versus placebo for prevention of breast cancer in women at high risk show a 62% reduction in risk [OR 0.38 (95% CI)] with little differences in the side effect profiles between exemestane and placebo [53]. However, the estimated number needed to treat (NNT) to prevent one breast cancer was projected to be 25. These and other NNT data indicate the need for more precise prediction of risk, for more effective agents and for biomarkers to predict women most likely to benefit from preventive therapy [60]. Currently tamoxifen is the preventive treatment of choice for premenopausal women and raloxifene for postmenopausal women. Aromatase inhibitors, although promising, require further data on their risk benefit ratio.

Lifestyle change

There are few randomized data to support a positive effect of lifestyle change in relation to breast cancer prevention. However, observational data indicate that lifestyle, mainly caloric excess and exercise deprivation, increases the risk of breast cancer and that breast cancer risk is reduced by decreasing weight and increasing physical activity. Two large prospective studies [61, 62] demonstrate that weight reduction in midlife or after the menopause decrease the risk of postmenopausal breast cancer by approximately 25–50% as does weight reduction related to bariatric surgery [63]. The results of other observational studies of weight reduction are mixed, possibly reflecting the small size of some and lack of data on maintained weight loss in the reported studies (summarized in [64, 65]). Reduction in fat intake without appreciable calorie restriction has only a minor affect on risk as shown in the Women’s Health Initiative large randomized trial [65]. This study also demonstrated that increased intake of vegetables, fruit and grain does not appear to reduce breast cancer risk.

A meta-analysis of 93 studies of exercise and breast cancer incidence reported an overall risk reduction of about 25% in both pre- and postmenopausal women [9]. The risk reduction was greatest in women with a normal BMI, suggesting that the optimal approach to lifestyle reduction of risk of breast cancer is to combine weight control and appropriate physical activity.

Use of screening programmes to focus risk reduction

Most Western and many other countries fund mammographic screening programmes designed to detect breast cancer early with aim of increasing cure rates. Beyond the system of FHCs that tends to focus on younger women with a family history, there are few systematic attempts to offer risk estimation and preventive measures in the context of screening [32, 33]. If such an approach is advisable, the question arises How might such a system be instituted? The rise in incidence of breast cancer, the trauma of breast cancer diagnosis and treatment and indications of the effectiveness of endocrine and lifestyle prevention make it appropriate to explore measures to reduce breast cancer risk by introducing risk prediction and preventive measures in the context of mammographic screening programmes.

Family History Clinics depend upon women presenting to their family doctors and being referred if above a certain risk threshold. The emphasis on family history tends to focus the service on younger women ignoring other important risk factors such as age of first pregnancy. Most breast cancer is diagnosed in older women during the screening period [[31, 47–72] in the UK National Health Service Breast Screening Programme (NHSBSP) Fig. 1]. The majority of older women do not have a family history but have other risk factors. Assessing risk at screening is potentially attractive because a large proportion of the female population is screened on a regular basis, it is not opportunistic and would involve a group of older women largely not seen in the FHC system. However, because screening large numbers of women effectively requires high throughput, it would be vitally important that adding risk prediction did not interfere with the mechanics of the screening process or reduce in the numbers of women attending for mammography.

The PROCAS study

We have initiated the PROCAS study to test the feasibility and acceptability to women of collecting breast cancer risk information (standard risk factors, mammographic density and SNPs) during the routine mammographic screening process in the UK NHSBSP. The aim of the study is attempt to improve risk prediction and introduce preventive strategies in women at high risk. Unlike most other screening programmes, the interval between mammograms in the NHSBSP is 3 years leading to high rates of interval cancers [66, 67] and thus a longer term aim of PROCAS is to investigate the possibility of adjusting the screening interval based on risk.

The study was initiated in October 2009 in fifteen screening sites in Manchester, mainly on mobile screening vans. By August 2011, 30 000 of the projected 60 000 women were recruited. Women are mailed a risk questionnaire (see PROCAS website http://www.uhsm.nhs.uk/research/Pages/PROCASstudy.aspx) in the interval between the invitation to screening and attending for a mammogram, because this was found to be the optimal approach in a previous large-scale study (CADET [68]). Approximately 70% of women invited attend for screening, and, to date, 40% of them have agreed to enter PROCAS. The risk questionnaire is completed prior to the mammography appointment and returned at the time of mammography where informed consent is obtained. The questionnaire is scanned into the database later and the Tyrer–Cuzick risk automatically computed. Mammographic density is assessed on a visual analogue scale (VAS) by 11 specially trained radiologists and radiographers. Three volumetric methods of density calculation (Volpara, Quantra and Stepwedge) are being compared with CUMULUS [69] and VAS, but the ultimate aim is to automate measurement so that the risk from density can be read out with standard risk factors automatically [70]. Ten per cent of the population give buccal smear samples for SNP estimation (currently 18 SNPs are tested).

Assessment of Tyer-Cuzick risk and VAS mammographic density have been reported for the first 10 000 women and SNP information available on 983 women (Evans et al. submitted). The median 10-year breast cancer risk was 2.65%, and the median VAS score was approximately 25%. Interestingly, when the top 5% of women for Tyrer–Cuzick risk, VAS density and SNPs were compared, there was little overlap in the populations identified, suggesting that the three methods detect different at risk populations.

Women with a 10-year risk of ≥8% or with a 10-year risk of 5–7.9% and with a mammographic density of ≥60% (VAS) were invited to attend or be telephoned to be counselled concerning their risk in our FHC. Over 80% have been counselled to date and 18.8% of 85 eligible women at high risk entered a randomized prevention study. Thus, to date, results from the PROCAS study indicate that it is feasible to assess breast cancer risk and offer risk information and risk-reducing advice within the context of the population mammographic screening. The utility for improved risk calculation by combining standard risk factors in the Tyrer–Cuzick model, mammographic density and SNP estimations will be assessed after each individual’s second mammogram at 3 years when we expect about 600 tumours in the population of 60 000 women enrolled. A similar programme to PROCAS called KARMA involving 100 000 women has been initiated in Sweden by Hall et al. (personal communication).

Prospects for risk-adapted mammographic screening

Unlike most other screening programmes, the interval between mammograms in the NHSBSP is 3 years. A recent report of the NHSBSP indicated that 38% of invasive tumours detected in the context of the programme were found in the interval between mammograms. Interval cancers have a poorer prognosis and reduce the potential effectiveness of the programme [64]. Identification of women likely to develop interval cancers and offering them tailored screening and preventive interventions may be a way to improve the effectiveness of the NHSBSP. There is evidence to suggest that women at high risk of breast cancer are more likely to develop interval cancers. The Swedish two-county study showed that women with a family history of breast cancer were significantly more likely to develop breast cancer in the interval between screens than equivalent women with no family history [72]. High mammographic breast density also increases the risk of developing interval breast cancer [71]. Thus, it may be appropriate to offer women at high risk and with high density a shorter screening interval. Women at very low risk of developing breast cancer may require screening less frequently and thereby safely reduce the numbers needing to be screened. In Fig. 3, we model the period between screens assuming a constant breast cancer detection rate of 4/1000. The model predicts that high-risk women should be screened every 15 months to achieve this rate of detection whereas intermediate risk women would need 3 years and low risk 6 yearly screens.

Figure 3.

The range of mammographic density.

Risk-adapted screening would require all women to complete a risk assessment form. The PROCAS study outlined here indicates that only 40% of women who attend for screening complete the form. Currently, there is no incentive for women to join the study except goodwill and the knowledge that they will be told their risks. However, in the future, if women were aware that the screening interval depended on risk, this may be a greater incentive. Women not completing the risk form could be maintained on the 3 years interval.

Summary

The increasing incidence of breast cancer focuses on the need for prevention and improved early detection of the disease. FHCs are models for management of younger women at risk but could potentially be used for older women determined to be at high risk in the UK NHSBSP.

Models such as Gail and Tyrer–Cuzick predict general risk well but have low discriminatory power for individuals [73]. The models may (or may not) be improved by adding other risk factors such as mammographic density and measurement of breast cancer risk–associated SNPs. Clinical trials of endocrine agents have demonstrated that it is possible to prevent breast cancer using preventive therapies, and observational studies suggest that lifestyle changes may also reduce risk; however, ideally we need appropriate randomized trials to test these assumptions.

The early results of the Manchester PROCAS Study suggests that it may be feasible to introduce risk prediction and prevention strategies in the context of a population-based mammographic screening programme and thus to focus preventive approaches and possibly introduce risk-adapted screening.

Conflict of internet statement

No conflict of interest was declared.

Acknowledgements

We acknowledge the support of the Manchester Biomedical Research Centre and funding from the National Institute of Health Research (NIHR) and the Genesis Breast Cancer Prevention Appeal. We would like to thank the study radiologists and radiographers, the whole PROCAS team, Dr Stephen Eyre, University of Manchester for advice with genotyping and Professor Stephen Duffy for reviewing the manuscript. This article presents, in part, independent research commissioned by the National Institute for Health Research (NIHR) under its Programme Grant (Reference Number RP-PG-0707-10031). The views expressed in this article are those of the author(s) and not necessarily those of the NHS, the NIHR or the UK Department of Health.

Ancillary