Abstract
Health care providers need simple tools to identify patients at genetic risk of breast and ovarian cancers. Genetic risk prediction models such as BRCAPRO could fill this gap if incorporated into Electronic Medical Records or other Health Information Technology solutions. However, BRCAPRO requires potentially extensive information on the counselee and her family history. Thus, it may be useful to provide simplified version(s) of BRCAPRO for use in settings that do not require exhaustive genetic counseling. We explore four simplified versions of BRCAPRO, each using less complete information than the original model. BRCAPROLYTE uses information on affected relatives only up to second degree. It is in clinical use but has not been evaluated. BRCAPROLYTE-Plus extends BRCAPROLYTE by imputing the ages of unaffected relatives. BRCAPROLYTE-Simple reduces the data collection burden associated with BRCAPROLYTE and BRCAPROLYTE-Plus by not collecting the family structure. BRCAPRO-1Degree only uses first-degree affected relatives. We use data on 2,713 individuals from seven sites of the Cancer Genetics Network and MD Anderson Cancer Center to compare these simplified tools with the Family History Assessment Tool (FHAT) and BRCAPRO, with the latter serving as the benchmark. BRCAPROLYTE retains high discrimination; however, because it ignores information on unaffected relatives, it overestimates carrier probabilities. BRCAPROLYTE-Plus and BRCAPROLYTE-Simple provide better calibration than BRCAPROLYTE, so they have higher specificity for similar values of sensitivity. BRCAPROLYTE-Plus performs slightly better than BRCAPROLYTE-Simple. The Areas Under the ROC curve are 0.783 (BRCAPRO), 0.763 (BRCAPROLYTE), 0.772 (BRCAPROLYTE-Plus), 0.773 (BRCAPROLYTE-Simple), 0.728 (BRCAPRO-1Degree), and 0.745 (FHAT). The simpler versions, especially BRCAPROLYTE-Plus and BRCAPROLYTE-Simple, lead to only modest loss in overall discrimination compared to BRCAPRO in this dataset. Thus, we conclude that simplified implementations of BRCAPRO can be used for genetic risk prediction in settings where collection of complete pedigree information is impractical.