Prospective validation of the BOADICEA multifactorial breast cancer risk prediction model in a large prospective cohort study

Abstract

BackgroundThe multifactorial Breast and Ovarian Analysis of Disease Incidence and Carrier Estimation Algorithm (BOADICEA) breast cancer risk prediction model has been recently extended to consider all established breast cancer risk factors. We assessed the clinical validity of the model in a large independent prospective cohort.MethodsWe validated BOADICEA (V.6) in the Swedish KARolinska Mammography Project for Risk Prediction of Breast Cancer (KARMA) cohort including 66 415 women of European ancestry (median age 54 years, IQR 45–63; 816 incident breast cancers) without previous cancer diagnosis. We calculated 5-year risks on the basis of questionnaire-based risk factors, pedigree-structured first-degree family history, mammographic density (BI-RADS), a validated breast cancer polygenic risk score (PRS) based on 313-SNPs, and pathogenic variant status in 8 breast cancer susceptibility genes: BRCA1, BRCA2, PALB2, CHEK2, ATM, RAD51C, RAD51D and BARD1. Calibration was assessed by comparing observed and expected risks in deciles of predicted risk and the calibration slope. The discriminatory ability was assessed using the area under the curve (AUC).ResultsAmong the individual model components, the PRS contributed most to breast cancer risk stratification. BOADICEA was well calibrated in predicting the risks for low-risk and high-risk women when all, or subsets of risk factors are included in the risk prediction. Discrimination was maximised when all risk factors are considered (AUC=0.70, 95% CI: 0.66 to 0.73; expected-to-observed ratio=0.88, 95% CI: 0.75 to 1.04; calibration slope=0.97, 95% CI: 0.95 to 0.99). The full multifactorial model classified 3.6% women as high risk (5-year risk ≥3%) and 11.1% as very low risk (5-year risk <0.33%).ConclusionThe multifactorial BOADICEA model provides valid breast cancer risk predictions and a basis for personalised decision-making on disease prevention and screening.

Enhancing the BOADICEA cancer risk prediction model to incorporate new data onRAD51C,RAD51D,BARD1updates to tumour pathology and cancer incidence

“…In a separate study, the BOADICEA v6 breast cancer model presented here has been validated in an independent prospective study of 66 415 women attending mammographic screening in Sweden. The full model, including RFs, mammographic density, PRS and PVs in BRCA1 , BRCA2 , PALB2 , CHEK2 , ATM , BARD1 , RAD51C and RAD51D was well calibrated overall (calibration slope 0.97 (95% CI: 0.95 to 0.99)) and in deciles of predicted 5-year risks and had a C-index of 0.71 (95% CI: 0.68 to 0.74) for discriminating between affected and unaffected women 31…”

“…The full model, including RFs, mammographic density, PRS and PVs in BRCA1, BRCA2, PALB2, CHEK2, ATM, BARD1, RAD51C and RAD51D was well calibrated overall (calibration slope 0.97 (95% CI: 0.95 to 0.99)) and in deciles of predicted 5-year risks and had a C-index of 0.71 (95% CI: 0.68 to 0.74) for discriminating between affected and unaffected women. 31 By explicitly modelling the effects of PVs in the new cancer susceptibility genes, the models provide personalised cancer risks of PV carriers when combined with QRFs, MD and PRS. Although the numbers affected by these changes will be small at population level, for individuals with RAD51C, RAD51D and BARD1 PVs and their families, the updated risks will be clinically important.…”

“…The multiplicative model has also been shown to be reasonable for the combined effects of PRS and RFs,35 but there is as yet no large-scale evaluation of the combined effects of PVs and RFs. However, recent prospective validation studies of the current and previous versions of the models suggest that, overall, the models fit well 11 31. Should deviations from the multiplicative model between these PVs and RFs emerge, the model can be updated to take them into account.…”

“…The new model parameters were taken from extensive, well-designed published studies together with existing parameters from model fitting 9 10. We and others have used this approach for developing previous versions of the models,8 11 12 21 40 41 which have been shown to provide clinically valid predictions 42 43 31. As is the case for the previous versions, the updates presented here are primarily based on studies of those of European ancestry in developed countries.…”

UK consensus recommendations for clinical management of cancer risk for women with germline pathogenic variants in cancer predisposition genes:RAD51C,RAD51D,BRIP1andPALB2

“…External, independent validation studies provide useful guidance on these aspects. 74 As such, discussions with patients will require explanation of the variation in risk estimations along with detailed discussions about timing of surgery and potential sequelae. This will then facilitate shared decision making with each individual patient to optimise care as per NICE guidance 197.…”