Umeå universitets logga

In this paper a method for predicting future number of cancer diagnoses is derived. The method is based on estimation of the cumulative hazard of cancer diagnosis and cumulative hazard of population mortality. The estimation of cancer hazard is done non-parametrically, while the population death hazard is assumed to follow the Gompertz-Makeham distribution. The prediction is based on the assumption that cancer incidence and population mortality in the prediction intervals are derived. Also, prediction intervals, based on non-parametric bootstrap, are presented. The method is applied to predict number of colon cancer diagnoses among females in the northern part of Sweden. It has shown to detect and adjust for changes in population age structure, and to provide good predictions in situation where the cancer incidence and population mortality are stable during the prediction period.

A method to estimate cancer risks, when a recessive genetic defect is assumed to influence the risk to develop cancer, is presented. The proportion of the recessive defect is also estimated. The method uses cancer diagnosis data from a large sample of biological families. A family is here parents and two children. The information used is the frequencies of cancer diagnoses among siblings. Estimations are performed with the method of moments. The work is methodological, meaning that no specific type of cancer is considered.

The subject considered is estimation of cancer risks assuming occurrence of recessive defects in some large population. Inheritance of the recessive defect is discussed for two generations, parents and children. A hierarchic model representing the distribution of the number of cancer diagnoses among children in a family is presented. The aim is to estimate two different cancer risks, one caused by environmental factors and the other by inherited genetical defects. The proportion of individuals in the population carrying the recessive defect is also of interest. The data assumed is diagnoses recorded in a database of familial cancer. To fit the model an application to MCMC is proposed. The idea is to simulate values from the posterior distribution of the model parameters when the posterior is defined as proportional to the likelihood function. By visualizing the simulated values we find parameter values supported by the observed data.