Probability estimation models for prediction of BRCA1 and BRCA2 mutation carriers: COS compares favourably with other models

Background Accurate risk assessment is essential to genetic counselling for a family history of cancer. Several empiric and computer-based risk assessment models have been developed to estimate a counselee's probability of being a carrier of mutation in BRCA1 and/or 2 genes, and to predict the risk of developing breast cancer. The COS model was developed from the better-known BRCAPro model to estimate risk of carriage of BRCA1 or 2 mutation. The COS model remains to be validated in a population discrete from that used for its development. Methods Four probability estimation models including COS, Manchester scoring system (MSS), BOADICEA and Tyrer-Cuzick (T-C) were applied to 275 Scottish families tested for BRCA1/2 mutations ascertained through regional genetics centres to ascertain models' sensitivity, specificity and accuracy. A subset of 130 families from Grampian (North and Northeast Scotland) was used to assess the models' ability to estimate the prevalence of BRCA1/2 mutation carriers. Sensitivity, specificity and ROC plots were used to ascertain models' individual performance, in terms of number of cancer cases, type of cancer and age of diagnosis of breast cancer. Results The COS and MSS models demonstrated the greatest sensitivities and area under ROC curves for the majority of family structures. They also showed the highest sensitivities (91-92%) and AUCs (76-78%) for the entire dataset overall. However, BOADICEA and T-C had the highest specificities for the majority of the family structures. BOADICEA and T-C generated the best estimates for the prevalence of mutations in the population; BOADICEA was more accurate for BRCA1 and T-C for BRCA2. Conclusion The COS and MSS models are the most effective models for use in clinical practice to select families for mutation analysis, but BOADICEA and T-C are more accurate for estimating mutation prevalence within a population.