Independent authentication is essential to justify the use of breast cancer risk models
predict and inform decisions on prevention and screening options. Very independent
Validations were made using cohorts for common breast cancer prediction models,
and those made with small sample sizes and short sequences,
and use earlier versions of the prediction tools. Our aim was to verify the relative
The performance of four models of frequently used breast cancer risk and assessment of the impact
from a limited input of data on each performance.
In this verification study, we used the Breast Cancer Provision Family Studies Team
(ProF-SC), which contains 18 856 women from Australia, Canada, and the USA that made
not having breast cancer when recruiting, between March 17, 1992 and 29 June 2011.
Women of the squad were aged 20-70 and we did not have previous history
o two-side prophylactic mastectomy or ovarian cancer, at least 2 months of sequence
data, and information about the history of the breast cancer family. We used this
selected to calculate 10 year risk scores and compare four breast models
Cancer risk predictions: Breast Analysis and Ovarian of Frequency of Diseases and Transporter
Model Algorithm Estimated (BOADICEA), BRCAPRO, Breast Cancer Risk Assessment
Instrument (BCRAT), and the International Breast Cancer Intervention Study (IBIS) model.
We compared the model calibration based on the ratio of the expected number of breastfeeding
cancer cases to the number observed of breast cancer cases in the cohort, and on the
is the basis of a discriminatory ability to separate those that will and will not
breast cancer has been diagnosed within 10 years as measured with the statistic concordance
(C-statistic). We made sub-group analyzes to compare the performance of the models in Aberystwyth
10 years in BRCA1 or BRCA2 mutation carriers (ie, BRCA-positive women), who are not carriers that have been proven
and unexpected participants (ie, BRCA-negative women), and participants younger than
50 years when recruiting. We also assessed the impact that limited data input (eg,
limiting the amount of family history and anenetic information that has included)
has been on the performance of the models.
After a median sequence of 11 · 1 year (IQR 6 · 0-14 · 4), 619 (4%) of 15 732 women selected
from ProF-SC squad study having been diagnosed with breast cancer in the future
recruited, and 519 (84%) of them had historically confirmed illness. BOADICE a
IBIS was well calibrated in the general validation cohort, while BRCAPRO and BCRAT
presumed risk (ratio of expected cases for cases observed 1 · 05 [95% CI 0·97–1·14] for BOADICEA, 1 · 03 [0·96–1·12] for IBIS, 0 · 59 [0·55–0·64] for BRCAPRO, and 0 · 79 [0·73–0·85] for BRCAT). The estimated C statistics for the complete validation cohort were 0 · 70
(95% CI 0 · 68-0 · 72) for BOADICEA, 0 · 71 (0 · 69-0 · 73) for IBIS, 0 · 68 (0 · 65-0 · 70) for BRCAPRO,
at 0 · 60 (0 · 58-0 · 62) for BCRAT. In sub-group analyzes by BRCA mutation status, the
The ratio of cases expected for observations for BRCA-negative women was 1 · 02 (95% CI 0 · 93-1 · 12)
for BOADICEA, 1 · 00 (0 · 92-1 · 10) for IBIS, 0 · 53 (0 · 49-0 · 58) for BRCAPRO, and 0 · 97 (0 · 89-1 · 06)
for BCRAT. For BRCA's positive participants, BOADICEA and IBIS were well calibrated,
but risk predicted by BRCAPRO (the ratio of expected cases observed 1 · 17 [95% CI
0·99–1·38] for BOADICEA, 1 · 14 [0·96–1·35] for IBIS, and 0 · 80 [0·68–0·95] for BRCAPRO).
We identified similar calibration patterns for girls who are less than 50 years when recruiting.
Finally, valuable restrictions were not affected by predictive scores BOADICEA and IBIS
input data to family history for first degree and second grade relationships.
Our results suggest that models contain multi-productive family history, such as
such as BOADICEA and IBIS, have been better able to predict even the risk of breast cancer
women at an average or below average risk of breast cancer. Although BOADICEA and IBIS
which was performed in the same way, further improvements in the accuracy of predictions could be
possible with hybrid models that incorporate the BOADICEA polygenic risk element
and the non-family history risk factors included in IBIS.
National National Institutions of the United States, National Cancer Foundation, Breast Cancer Research
Foundation Foundation, National Health Authority of Australia and Medical Research Council, Victorian Health
Promotion Foundation, Victorian Breast Cancer Research Consortium, Australian Cancer,
National Breast Cancer Foundation, Queensland Cancer Fund, New Cancer Councils
South Wales, Victoria, Tasmania, South Australia, and Chancery West Cancer