gms | German Medical Science

Aims: About 5% of all colorectal cancer (CRC) cases are hereditary due to a specific genetic conditions. About 25% of all CRC cases are related to familial clusters without clear genetic background. It is of interest to identify these families for specific screening. The Amsterdam and Bethesda criteria [1] were introduced to detect hereditary risk, but are also commonly used to predict familial cancer risk. For example, offspring of a CRC case have an approximately doubled risk of also getting CRC [2]. Hence, it is of interest to determine the information content of these criteria with respect to familial CRC risk. The currently active study "Familien Schützen und stärken - der Umgang mit dem familiären Darmkrebsrisiko" will create family data of cancer histories for patients diagnosed with CRC during 2012 and 2013. The talk will present a statsitical analysis strategy for the upcoming data.

Methods: A Bayesian approach is used to quantify the posterior of a family to carry familial CRC risk given its own cancer history. Several statistical models for familial risk are introduced. The penetrance model for familial CRC risk is based on an increased hazard ratio, accelerated time, or time shift for the general age dependent CRC incidence. Bayesian model mixing is used to incorporate the uncertainty with regard to the relevant genetic mechanisms. Simulation studies allow assessing the quality of the inference for the parameters needed in the models considered.

Results: An MCMC approach was taken to estimate the model parameters from the simulated data. Simple families consisting of parents and three children were constructed. Sex and age specific incidence of CRC was modelled. The parameters could be estimated by sufficient a fair precision. In the most informative situation (dominant, risk given to all children) the mean posterior for a family with two CRC cases to carry familial risk was 37.0about 40%. If risk is transmitted to children by more uncertain mechanisms (modelled by random effects), the mean posterior for a family with two CRC cases to carry familial risk was 28.4about 30%. Extension to model mixing is possible by using posterior probabilities for the genetic mechanisms studied.

Conclusion: In our artificial population made up by uniform families with three children, the information content of the direct relative Amsterdam and Bethesda criteria to detect familial CRC constellations is very restricted. In further work, family trees will be extended to more generations and varying number of offspring. It will also be of interest to do similar calculations for more complex Amsterdam and Bethesda criteria. Finally we use our strategy to analyse data of an ongoing study on familial CRC risk [3].