Breast cancer causes 11,684 deaths per year in the UK; in the age group 35–50 years, it causes more deaths than any other medical condition,1 and the highest number of quality life-years are lost in this age group. In 2011, there were 50,285 breast cancer diagnoses in the UK.1 Currently, screening through the NHS Breast Screening Programme (NHSBSP), with 3-yearly mammography, is targeted at women aged 47–73 years. Women aged 40–49 years with a ≥ 3% risk in that decade are also eligible for annual mammography, but currently these women are assessed by family history alone. A smaller group of women who carry mutations in BRCA1 (breast cancer 1 gene)/BRCA2 (breast cancer 2 gene), or who have an 8% risk aged 30–39 years or a 12% risk aged 40–49 years, can be assessed for magnetic resonance imaging (MRI) screening.2 In the UK, 10–11% of women are projected to develop breast cancer in their lifetime and treatment is becoming increasingly expensive, with newer, complex chemotherapy regimens and drugs such as Herceptin® (Genentech Inc., San Francisco, CA, USA). Although the number of breast cancer deaths has decreased in many Western countries, the incidence of the disease is continuing to rise. In particular, in countries with historically low incidence, breast cancer rates are rising rapidly, making it now the world’s most prevalent cancer.3 The increase in incidence is almost certainly related to dietary and reproductive patterns associated with Western lifestyles. Indeed, there is evidence from genetic studies in the USA, Iceland and the UK of a threefold increased incidence, in the general population and in those at the highest level of risk with BRCA1/BRCA2 mutations, in the past 80 years.2,4,5
BRCA1 and BRCA2 are high-penetrance genes mutations, which are carried by around 1 in 400 women in the outbred UK population but by as many as 1 in 40 women of Ashkenazi Jewish origin.4–6 Women who carry mutations in these genes have a risk of breast cancer to 70 years of up to 85%,4–6 but population-based studies have shown lower risk estimates of 40–50%.5 Targeted screening and prevention strategies would, potentially, create huge savings to the NHS and increase the quality and length of life for many women. Although preventative measures based on chemotherapy and lifestyle change are possible, these are not feasible on a population basis, in part because of the difficulties of identifying women in the general population who are at increased risk. Tamoxifen and raloxifene are now approved by the National Institute for Health and Care Excellence (NICE) as chemoprevention to be offered to high-risk women (10-year risk of ≥ 8%).2
Unlike most screening programmes in other countries, which typically use 1- or 2-yearly intervals, the interval between mammograms in the NHSBSP is 3 years; possibly partly as a result of this, 40% of tumours arise in the interval between mammograms. These cancers have a poorer prognosis and reduce the potential effectiveness of the programme.6,7 Identifying women likely to develop interval cancers and offering them tailored screening and preventative interventions may be a way to reduce the incidence of interval cancers. There is evidence to suggest that women at high risk of breast cancer are more likely to develop interval cancers. The Swedish 2-county study8 showed that women with family history of breast cancer were significantly more likely to develop breast cancer in the interval between 2-yearly screens than equivalent women with no family history. High mammographic breast density (MD) also considerably increases the risk of developing interval breast cancer.9,10 A screening programme adapted to risk may, therefore, improve the effectiveness and efficiency of the NHSBSP. For women at very low risk of developing breast cancer, the screening interval might be extended, thereby potentially safely reducing the number needing to be screened.
Evans and Howell3 state that there are two main types of risk assessment: the chances of developing breast cancer over a given time span, including the lifetime, and the chances of there being a mutation in a known high-risk gene such as BRCA1 or BRCA2. While some risk assessment models are aimed primarily at solving one of these questions, many also have an output for the other. For example, the Tyrer–Cuzick11 model was developed to assess breast cancer risk over time, but does have a read-out for BRCA1/BRCA2 probability for the individual (text reproduced with permission from Evans and Howell.3 © BioMed Central 2007). Breast cancer risk algorithms which aim to predict risk over a given time span generally include a combination of known risk factors, such as a family history of the disease and reproductive and hormonal history. Although current models perform well at predicting the overall number of breast cancer cases arising in a particular population, they are poor at identifying specific individuals.4 To assess breast cancer risks over time as accurately as possible, all known risk factors for breast cancer need to be assessed.
Risk factors
Family history of breast cancer in relatives12
Age at onset of breast cancer.
Bilateral disease.
Degree of relationship to family member (first or greater).
Multiple cases in the family (particularly on one side).
Other related early-onset tumours (e.g. ovary, sarcoma).
Number of unaffected individuals (large families with many unaffected relatives will be less likely to harbour a high-risk gene mutation).
Hormonal and reproductive risk factors
Hormonal and reproductive factors have been recognised for a long time to have an important role in breast cancer development. Prolonged exposure to endogenous oestrogens is an adverse risk factor for breast cancer. Early menarche and late menopause increase breast cancer risk, as they prolong exposure to oestrogen and progesterone.13–22
Long-term combined hormone replacement therapy (HRT) treatment (> 5 years) after the menopause is associated with a significant increase in risk. However, shorter-term treatments may still be associated with risk to those with a family history of breast cancer.14 In a large meta-analysis, the risk appeared to increase cumulatively by 1–2% per year but disappear within 5 years of cessation.15 Oestrogen-only HRT has a risk that appears much lower, and it may be risk neutral.16–19 A meta-analysis also suggested that both during current use of the combined oral contraceptive and 10 years post use, there may be a 24% increase in risk of breast cancer.13
A woman’s age at first pregnancy influences the relative risk (RR) of breast cancer, as pregnancy transforms breast parenchymal cells into a more stable state, potentially resulting in less proliferation in the second half of the menstrual cycle. As a result, early first pregnancy offers some protection, while women having their first child over the age of 30 years have double the risk of women delivering their first child under the age of 20 years, and these are likely to be similar in those at highest risk from a BRCA1/BRCA2 mutation.20,21
Mammographic breast density
It has been shown that increased breast density not only is associated with an elevated risk of breast cancer but is the largest risk factor after age.23–26 The difference in risk between women with extremely dense, as opposed to predominantly fatty, breasts is approximately four- to sixfold.26 Assuming that the association between breast density and breast cancer risk is causal, MD is the single assessable risk factor with the largest population attributable risk and may also have a substantial heritable component.25,26
If MD is to be used to estimate breast cancer risk, it is necessary to identify the optimal method of MD assessment, in terms of both practicality and feasibility of incorporation into routine practice, and accuracy of risk prediction. The assessment of breast density from mammograms has generally been provided by the subjective visual evaluation of an expert. Computer-based methods have also been developed in an attempt to make the assessment of MD more quantitative; however, many of the older computer-based methods, such as Cumulus (Sunnybrook Health Sciences Centre, Toronto, ON, Canada), still rely on some subjectivity. More recently developed computer-based methods have aimed to determine the true volumes of dense and fatty tissue from digital mammograms. As these methods are automated and require no subjective input, they are by far the most practical methods for wider use.
Genetic factors
Mutations in breast cancer genes such as BRCA1 and BRCA2 are too infrequent to affect risk prediction appreciably in the models for the general population. However, recently identified single nucleotide polymorphisms (SNPs) in many genes and outside genes (n = 77),27 which individually confer small changes in risk, may prove useful in predicting larger differences in risk when considered together. Four Genome-Wide Association Studies (GWASs),8–10,28 published before our programme grant, found common genetic variants (SNPs) each carried by 28–44% of the population were associated with a 1.07–1.26 RR of breast cancer. These variants linked to four genes [FGFR2 (fibroblast growth factor receptor 2), TOX3 (TOX high-mobility group box family member 3), MAP3K1 (mitogen-activated protein kinase kinase kinase 1, E3 ubiquitin protein ligase) and LSP1 (lymphocte-specific protein 1)] confer as much as a 1.17–1.64 risk if two copies are carried. When combined in an individual, they give higher than additive risk of breast cancer.29 Another variant, CASP8 (caspase 8, apoptosis-related cysteine peptidase), is associated with reduced breast cancer risk.30 There are now 77 genetic variants associated with breast cancer risk, but their application requires further validation and assessment of interactions. Therefore, to improve the accuracy of existing risk prediction models, it is necessary to investigate validated SNPs as they are discovered, and, where possible, incorporate these genetic factors into the best performing risk models.
Other risk factors
A number of other risk factors for breast cancer are being further validated. Obesity, diet and exercise are probably interlinked.31,32 Other risk factors such as alcohol intake have a fairly small effect, and protective factors such as breastfeeding are also of small effect unless a number of years of total feeding have taken place. None of these factors is currently incorporated into available risk assessment models.
Risk models
In a comparison of the Gail (National Cancer Institute; www.cancer.gov/bcrisktool), Claus (www4.utsouthwestern.edu/breasthealth/cagene/default.asp), Ford (BRCAPRO) (BayesMendel Lab, Harvard University, Boston, MA, USA; http://bcb.dfci.harvard.edu/bayesmendel/index.php) and Tyrer–Cuzick (TC) (version 6; Professor Jack Cuzick, Centre for Cancer Prevention, London, UK) risk prediction models, using observed data from 3170 women with a family history of breast cancer in the UK, we showed that the TC model performed best.13 We identified a need to reassess these models in larger numbers of women and also to compare the BOADICEA (Breast and Ovarian Analysis of Disease Incidence and Carrier Estimation Algorithm; Cambridge University, http://ccge.medschl.cam.ac.uk/boadicea/) model alongside these models in the family history population. There has not been a large-scale study comparing risk models in the general population, and, therefore, we aimed to assess the TC and Gail models in the general population. The Claus and BRCAPRO models are unsuitable for population prediction, as they are entirely based on family history risk factors.
Improving the risk models
Current risk prediction models are based on combinations of risk factors and have good overall predictive power, but are still weak at predicting which particular women will develop the disease. New risk prediction methods are likely to come from examination of a range of high-risk genes as well as SNPs in several genes associated with lower risks.8 This was married in a prediction programme with other known risk factors to provide a far more accurate individual prediction.
The incorporation of density into the standard risk prediction models is associated with some improvement in risk prediction,33,34 and three publications suggest that adding breast density improves the Gail risk model.35–37 Therefore, to improve the accuracy of the best performing risk model in each population of women (family history and general population), it is necessary to incorporate MD. We used a number of different density assessments to determine which adds most to the precision of breast cancer risk estimation.
Economic evidence supporting the incorporation of risk prediction into the NHS Breast Screening Programme
There is a substantial economic evidence base that has been generated to support the introduction of breast screening programmes, in general. To date, however, there is no economic evidence to support that using a risk-based screening programme will be an effective use of limited health-care resources. To understand the potential impact of introducing a risk-based screening programme, preliminary economic analysis was conducted to identify relevant costs and benefits. A decision-analytic model was constructed to represent the options for screening strategies for women at increased risk of breast cancer who have been identified using the best performing risk model. A systematic review summarised existing economic modelling research in this clinical area and used to inform the development of a model structure.38–41 Earlier work of the programme informed necessary changes to the model structure, clinical pathways, data sources, etc. An expert panel (project leads, geneticists, oncologists, patient representatives) refined the modelling structure. Data from PROCAS (Predicting the Risk of Cancer At Screening) [UK Clinical Research Network identification number (UKCRN-ID) 8080] and systematic reviews and assimilation of published data were used to inform the model inputs.
Overarching research questions and aims
The overall aim of this project was to improve risk prediction and early detection of breast cancer, for women who have a family history of the disease and for those in the NHSBSP. To achieve this, it is necessary to first identify the best performing model in each population and then improve the precision of the best performing model by incorporating MD (assessed using the optimal method) and new genetic modifiers of risk, SNPs (where possible). This will enable better individualised risk prediction and allow women access to appropriate risk-management strategies and screening intervals. We also aimed to conduct a preliminary economic analysis to identify the relevant incremental costs and benefits associated with introducing a risk-based screening programme on a national level. As the report is largely based on two cohort studies, we have used STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guideline reporting recommendations where possible.
Chapter 2 concentrates on the population with a family history of breast cancer. Our original proposal was to update our earlier study, based on just over 3000 women screened for breast cancer between 1987 and 2002, which had shown that the TC model and a Manual approach were most predictive. As outlined in the application, it has been possible to rerun these models in nearly 10,000 women, with almost an eightfold increased power and with over 400 detected breast cancers. In addition to the Gail, TC and Manual models, we have been able to assess the newer BOADICEA model in a subset of women. In addition, as proposed, MD has been assessed in a case–control study (see Chapter 4). This was able to assess a visual assessment score (VAS) in all patients for whom mammograms were available. In addition, two automatic models were assessable on those who had digital mammograms.
In Chapter 3, we report on a large-scale analysis of SNPs undertaken in a number of populations, as proposed in the application. At the time of application, only eight or nine SNPS were assessable, but in 2010 this rose to 18, and these 18 were used in all the analyses using a PRS. We have assessed the SNP18 in four populations:
We proposed to extended our analysis to include all female BRCA1/BRCA2 carriers (n = 850 at that time) and assess the SNP18 for interactions and modifier effects. As above, the proposed research was extended to test 925 proven BRCA1 or BRCA2 female carriers to assess impact on breast cancer risk.
The second proposed analysis was to test the SNPs in 1400 high-risk BRCA1-/BRCA2-negative breast cancer cases and compare this with 500 matched control samples. The variants will be weighted for their individual effects in heterozygous and homozygous states and assessed for interactions with other variants. This analysis was deemed unnecessary, as we obtained funding (from the Genesis Breast Cancer Prevention Appeal) for obtaining saliva deoxyribonucleic acid (DNA) in 10,000 women from our population-based PROCAS study. We carried out testing of > 5000 women in PROCAS and have extended the modelling to assess the effects of using 67 SNPs.
The testing of DNA from 58 breast cancer phenocopies (women who are negative for the family BRCA1/BRCA2 mutation) was likewise proposed to be assessed for the variants in a weighted analysis to identify whether or not significantly more of the high-risk alleles are carried by those women to account for their increased risk of breast cancer. SNP18 was run on these women.
We proposed using data from the first three analyses to develop a weighted score to assess the predictive value of the combined group of validated variants in predicting which women developed breast cancer in our familial screening programme. We proposed using the incident cancers and matching with controls on a 1 : 3 basis randomly selected from our family history clinic (FHC) for the presence or absence of familial BRCA1/BRCA2 mutation plus all other currently used risk factors (menarche, parity, age, family history and breast disease). Women were recruited from our FHCs to provide blood samples for DNA analysis or to give their permission to use stored DNA. The matching of controls was changed (with statistical advice from Professor Cuzick) to matching for just age and type of mammogram.
Chapter 4 concentrates on our large-population PROCAS study. In this study, which was the largest recruiting portfolio study nationally, we recruited > 53,000 women and reached our target of 600 prospective breast cancers. As outlined in the application, we have shown that it was feasible to collect risk information from women attending mammography through the NHSBSP and have already given risk feedback to nearly 800 women at high and low risk. We have shown that, as anticipated, MD adds substantially to the predictive precision of the risk models TC and Gail. We were hopeful that automatic measures of density carried out on digital mammograms would be at least equivalent to the gold-standard Cumulus or VAS, but this was not the case, and more work is therefore required to develop a better digital method. A VAS-adjusted TC model was shown to improve risk prediction and identified 70% of the population at < 3.5% 10-year risk who were at very low risk of a high-stage cancer.
Finally, in Chapter 5, three studies are reported. The first study was a systematic review of published economic evaluations relevant to breast screening and summarised the current economic evidence base. Study 2 explored the potential out-of-pocket expenses incurred by women attending a national breast screening programme. Study 3 structured a decision-analytic model to conduct a preliminary cost-effectiveness analysis to identify the relevant costs and patient benefits if a risk-based breast screening programme was introduced into clinical practice.
