Evaluation of mathematical models for breast cancer risk assessment in routine clinical use

Abstract

Chemoprevention, prophylactic surgery, and intensified screening can be offered to patients with an increased lifetime risk, p(life), for breast cancer. Estimation of p(life) includes BRCA analysis and risk estimation based on individual risk factors and family history. MENDEL and BRCAPRO are models that estimate the probability of BRCA1/2-mutations, p(mut), and p(life). In this study, the models are compared with Ford and Claus penetrance/frequency functions. The results were compared with the Tyrer-Cuzick model. Genetic analysis of 111 breast cancer-affected patients from 103 kindreds with a family history of breast and/or ovarian cancer (German Consortium for Hereditary Breast and Ovarian Cancer) was carried out by sequencing BRCA1 and BRCA2. p(life) and p(mut) were calculated with MENDEL, BRCAPRO(Claus), BRCAPRO(Ford), as well as the Tyrer-Cuzick model. The accuracy of p(mut) was analyzed by receiver operating characteristics, and p(life) of each model was compared. The strongest correlation of p(life) was shown by BRCAPRO(Ford)/MENDEL, at r=0.69; no correlation was shown by BRCAPRO(Claus)/MENDEL, at r=0.018. The Tyrer-Cuzick model had the strongest correlations with MENDEL and BRCAPRO(Ford). For MENDEL and BRCAPRO, low correlation or p(mut)-prediction was improved by excluding kindreds with ovarian cancer. p(mut) showed the best accuracy for BRCAPRO(Ford) and MENDEL. BRCAPRO and MENDEL are useful tools for calculating p(mut). They can provide support in decision-making for or against genetic analysis. Estimations of p(life) and p(mut) based on a mathematical model should use algorithms and penetrance/frequency data appropriate to the population counseled. Reproductive/hormonal data, should be incorporated as Tyrer-Cuzick does.