Mendelian Risk Prediction with Application to Colorectal Cancer

Two major types of modelling approaches for risk prediction are \regression-based " and \Mendelian". Regression-based approaches use regression techniques directly on the pedigree data. The test results and specic pedigree features are extracted to constitute the response variables based on clinical and epidemiological knowledge. There exist potential problems such as incomplete usage of biological evidence, restricted application to families meeting the same requirements, and extrapolation based on dicult-to-test assumptions such as linearity and additivity of the predictor terms. In contrast, Mendelian approaches rst estimate \genetic parameters " of the mutations| population prevalence and penetrance, including modiers; they then apply a well-established Bayesian prediction technique to transform the genetic parameters into carrier probabilities. Hereditary Nonpoliposis Colorectal Cancer (HNPCC), or the Lynch syndrome, can be caused by germline mutations on the DNA mismatch repair genes, mainly MLH1 and MSH2. We present a model called CRCAPRO that evaluates the probability that an individual carries a mutation in the MLH1 or MSH2 genes based on his or her detailed family history of colorectal and endometrial cancers, and results of microsatellite instability testing if available. The model makes use of Bayes rules, is built on the principle of Mendelian inheritance of autosomal dominant genes, and assumes known mutation frequencies and penetrances and known MSI test accuracy. As compared to existing guidelines and empirical models, CRCAPRO provides more precise probabilities and higher resolution that will be especially helpful to families of moderate risks