Evaluation of the Tyrer-Cuzick v8 Model for Breast Cancer Risk Stratification in a UK Family History Clinic

Emily Berry¹ , Adam Brentnall¹, Sacha Howell², D. Gareth Evans²

¹Queen Mary University of London, ²The University of Manchester

 

1. Background:

-        Breast cancer is the most common cancer among women in the UK and the fourth leading cause of cancer death¹.

-        Accurate risk prediction helps identify individuals at highest risk who may benefit from targeted interventions.

-        Tyrer-Cuzick (TCv8) model:

• Widely used in Family History, Risk and Prevention Clinics (FHRPCs) for risk prediction².
• Uses detailed questionnaire data on family history and reproductive, hormonal and environmental factors.
• Optional inputs: mammographic density (MD) and polygenic risk scores (PRS)³.

Objective: Evaluate risk stratification performance of the TC model in a FHRPC in Manchester, UK. 

2. Methodology:

Cohort:

• Breast cancer-free women who consented between 2010-2012 at the FHRPC, Manchester.
• Follow-up: until January 2024; censored at breast cancer diagnosis, risk reducing mastectomy, non-breast-cancer-related death, migration or end of follow-up.

Data :

• Questionnaire
• Saliva for genetic testing:
  • BRCA1/2 variants
  • PRS (SNP313 / SNP143 / SNP18)
  • Digital mammograms analysed with Volpara for breast density

Tyrer-Cuzick risk models:

1. base (questionnaire only)
2. + PRS (questionnaire + non-BRCA PRS)
3. + MD (questionnaire + mammographic density)
4. full (questionnaire + PRS + MD)

Analysis:

• Stratified by BRCA status (BRCA1, BRCA2, non-carrier) and 10-year risk groups (<3%, 3-5%, 5-8%, 8+%), corresponding to current UK risk-based screening guidance.
• Calibration Assessment:
  • Absolute: observed-to-expected (O/E) ratios to compare observed breast cancer diagnoses (O) with expected (E) from each Tyrer-Cuzick model.
  • Relative: Cox models of log(10-year risk)

3. Results

a) Baseline characteristics:               

Characteristics:                                                                     

-        Age: 49 (mean); 8.2 (sd)

-        55% premenopausal

-        80% parous

-        6% on endocrine therapy

-        10% enrolled in IBIS-1/11 trials

PRS availability: Table 1: Number of patients with each PRS type used in the analysis.

b) Breast Cancer, Surgery and Mortality:

Women carrying variants in moderate- or high-penetrance genes have an increased risk of breast cancer (moderate: 20-50%; high: >50% lifetime risk).

Table 2: Number of women, breast cancers, non-breast-cancer-related deaths and risk-reducing mastectomies in carriers and non-carriers of breast cancer mutations.

Key observations:

• Median follow-up: 10 years
• Breast cancer incidence was low in non-BRCA (36/868; 4%) and high in BRCA carriers (12/64; 19%)
• Risk-reducing mastectomies (RRMs) and non-breast cancer-related deaths were more common in BRCA carriers.

c) Risk Group Reclassification:

Figure 1: Reclassification of 10-year risk estimated by the Tyrer-Cuzick (TC) model when mammographic density (MD) and polygenic risk scores (PRS) are included in the model.

MD and PRS flatten risk distribution:

-        Net increase of +148 individuals into the highest (8+%) category and of +88 into the lowest (<3%) category.

-        Net decrease in the middle two categories (3-5%: -5; 5-8%: -198).

d) Calibration Assessment – Absolute:

BRCA1/2 calibration (Table 3):

-        Table 3: Number of cases observed divided by number expected by base Tyrer-Cuzick model and TC + Mammographic Density (MD) in BRCA1/2 carriers.

 

• BRCA1: Observed < Expected => overestimation by the model
• BRCA2: Observed > Expected => underestimation by the model

Non-BRCA subgroup calibration (Figure 2):

• Calibration slope < 1 for all models => overestimating high risk and underestimating low risk.
• Figure 2: Number of cases observed divided by number expected within each 10-year risk group predicted by the four Tyrer-Cuzick model variants. Regression line with 95% confidence interval is shown in black and perfect calibration is shown at O/E = 1 with a blue dashed line.

e) Calibration Assessment – Relative:

-        Figure 3: Relative risk calibration of Tyrer–Cuzick (TC) 10-year risk for each model variant  (PRS: Polygenic Risk Score; MD: Mammographic Density). Cox regression coefficients for log(10-year TC risk) with 95% confidence intervals are shown.

-        β = 1 => perfect relative calibration;

-        β = 0 => no association with breast cancer.

-        All models have good relative risk calibration.

 

4. Interpretation & Future Work

-        Tyrer-Cuzick v8 model provided reasonable risk stratification in a UK FH clinic, supporting the potential of current risk models.

-        Incorporating breast density and PRS substantially changed the risk profile of the cohort, resulting in a flatter distribution with more individuals put in the extreme risk categories.

-        However, precision of calibration estimates was limited by sample size.

-        Consistency of PRS to be improved by using SNP313 for everybody.

-        There is a need for further validation in larger cohorts.

References

1. Cancer Research UK.  [cited 2025 20/05/2025].

2. Evans, D.G., et al., Breast cancer incidence and early diagnosis in a family history risk and prevention clinic: 33-year experience in 14,311 women. Breast Cancer Research and Treatment, 2021. 189(3): p. 677-687.

3. Evans, D.G.R., et al., Breast cancer risk stratification in women of screening age: Incremental effects of adding mammographic density, polygenic risk, and a gene panel. Genetics in Medicine, 2022. 24(7): p. 1485-1494.