Estimating the Probability of Carrying a Gene Mutation for Breast Cancer

By Estella Chen, PhD and Tara Friebel, MPH

Genetic counselors may use models to determine whether a patient is likely to have a mutation in BRCA1 or BRCA2. In general, existing models either give the probability of testing positive for a mutation (empirical models) or the probability of carrying a mutation (Mendelian models). Some models, such as BRCAPRO, can predict both. Empirical models were developed by direct use of pedigree (family tree) data from individuals with known mutations. Risk factors were gathered from pedigrees based on clinical and epidemiological data, and those that predicted gene status were identified using statistical techniques. Mendelian models are based on autosomal dominant inheritance of BRCA1 and BRCA2 (i.e. equally inherited by men and women, one mutated gene copy sufficient for predisposition). These models also consider the prevalence of mutations within the population and the age and frequency at which hereditary breast cancer appeared. Explicit genetic models for predicting carrier status for an individual were then formulated.

Since each model does not use the same criteria or methodology, each can give widely differing predictions.The following case illustrates this point(changes have been made to preserve confidentiality):

A patient diagnosed with breast cancer at age 32 has four sisters. One sister is 60 and unaffected (does not have breast cancer). Another was diagnosed with breast cancer at 69. The fourth was diagnosed with breast cancer at 60 and with ovarian cancer at 69. The patient's mother was diagnosed with breast cancer at age 40, and both maternal aunts were diagnosed with ovarian cancer at ages 76 and 54, respectively. No information is available about the paternal side. The family is not of Ashkenazi Jewish descent. The probabilities for this patient testing positive are: 88.5% (Penn Model), 30% (Myriad I for BRCA1 only) and 7.9% (Myriad 2 for both BRCA 1 and 2). Other models calculate the probability that the person is a mutation carrier, which in this case would be 96% (BRCAPRO) and 68% (Yale model).

Because of the variation, genetic counselors use the models with caution and give patients a broad range when estimating the likelihood that the patient carries a mutation. To address this limitation, Dr. Giovanni Parmigiani* is leading a study for the comparison and validation of the five models mentioned above, plus a sixth one. The study is a collaboration with the Cancer Genetics Network (CGN) and allows the use of a large set of patient data combined from the eight sites of the CGN. Probabilities from each model will be calculated on the combined data, and will allow evaluation of the models using the same data set.

The specific models to be tested include: Couch Model (based on empirical data), Frank (Myriad II) Model (empirical), NCI Model (empirical), Penn Model (empirical), Claus (Yale) Model (Mendelian), BRCAPRO (Mendelian). These models use similar, but not identical sets of patient and family history information. It will be valuable to determine which combination of patient data best predicts a patient's likelihood of being a mutation carrier. Furthermore, it is hoped that the direct comparison of these models on the same set of data will improve accuracy and result in practical indications for clinical use.

*Dr. Giovanni Parmigiani, Associate Professor of Biostatistics and Oncology, co-developed BRCAPRO and researches probabilistic modeling and simulation in medical decision-making and cancer genetics.