APH 2003, 61, 297-312:

Comparing the performance of family characteristics and predictive models for germline BRCA1/2 mutations in breast cancer families.

G. Goelen, E. Teugels, E. Sermijn, M. Bonduelle, B. Neyns, J. De Greve

Keywords : breast cancer, ovarian cancer, BRCA genes, predictive models, ROC-curve

Aim: The probability of finding a BRCA1/2 mutation is an important factor in the decision to initiate a mutation screen in individual families. This may be based on family characteristics or the probability that a mutation will be identified that may be calculated by various models. The current study compares the relative usefulness of these indicators.

Methods: All 155 Belgian families that had been counselled and accepted for screening for BRCA1/2 germline mutations at the Vrije Universiteit Brussel Family Cancer Clinic and that had enough clinical information on file for at least two of the Couch (1997), Shattuck-Eidens (1997), Frank (1998), BRCAPRO (1998), and Vahteristo (2001) models to be applied post hoc were included in the study sample. The sensitivity and specificity of relevant family characteristics were plotted on the receiver operating characteristic curves of the probability models.

Results: The presence of ovarian cancer in the family is associated with a mutation identification rate of nearly 60%. Breast cancer families without ovarian cancer are a clinically relevant but difficult subgroup for predicting the identification of a BRCA1/2 mutation. BRCAPRO was the only informative predictive model in this subgroup but did not perform better than the number of breast cancers in the family used as the sole criterion for mutation screening.

Conclusion: Two family characteristics, the presence of ovarian cancer and the number of breast cancers in the families without ovarian cancer, are as useful for estimating the probability of finding a BRCA1/2 gene mutation as the available probability models. Such models, such as BRCAPRO, may, however, merit further validation.