Risk models in genetic epidemiology

Advances in the identification and treatment of genetically transmitted diseases have lead to an increased need for reliable estimates of genetic susceptibility risk. These estimates are used in clinic settings to identify individuals at increased risk of being a carrier of a disease susceptibility allele as well as to define the probability of developing a particular disease given one is a carrier. Accurate assessment of these probabilities is extremely important given the implications for medical decision making including the identification of patients who might benefit from genetic counselling or from entry into clinical trials. A wide range of risk models has been proposed including those that utilize logistic regression, Cox proportional hazards regression, log-incidence models, and Bayesian modelling. The specific data used to create the various risk models varies by disease and may include molecular, epidemiologic, and clinical information although, in general, family history remains the primary variable of interest, particularly for those diseases for which a susceptibility allele(s) has yet to be identified. When permitted by sample size, researchers also attempt to measure the effect of any gene-environment interaction. In this paper we give an overview of the various definitions of risk as well as several of the more frequently used methods of risk estimation in genetic epidemiology at present. In addition, the means by which different methods are able to provide a measure of error or uncertainty associated with a given risk estimate will be discussed. Applications to risk modelling for breast cancer are given the disease for which risk assessment has probably been most extensively defined.