Purpose: Carrier prediction models estimate the probability that a person has a BRCA mutation. We evaluated the accuracy of the BOADICEA model and compared its performance with that of other models (BRCAPRO, Myriad I and II, Couch, and Manchester Scoring System). We also studied the effect of extended family information on risk estimation using BOADICEA.
Methods: We compared the area under receiver operating characteristic curves generated from 472 families with one member tested for BRCA mutations. We calculated sensitivity, specificity, and predictive values at an estimated probability of 10% and explored the biases of carrier prediction.
Results: BOADICEA performed better than the other models in Ashkenazi Jewish (AJ) families, BRCAPRO performed slightly better in non-AJ families, and Myriad II performed comparably well in both groups. Including extended family information in BOADICEA yielded slightly better performance than did limiting the information to second-degree relatives. Using a 10% cutoff point, BOADICEA and Myriad II were most sensitive in predicting BRCA1/2 mutations in AJ families, and Myriad II was most sensitive in non-AJ families. The Manchester Scoring System was the most sensitive and least specific in a subgroup of non-AJ families. BOADICEA and BRCAPRO tended to underestimate the observed risk at low estimated probabilities and overestimate it at higher probabilities.
Conclusion: The BOADICEA, BRCAPRO, and Myriad II models performed similarly. Including second-degree relatives slightly improved carrier prediction by BOADICEA. The Myriad II model was the easiest to implement.