Investigation of cohort heterogenerity with Västerbotten data
We have developed a method for cohort studies in epidemiology that is able to identify a sub-structure, or sub-classes in the study population. This method is based on latent class analysis and uses Bayesian model selection to determine the optimal sub-structure of data. In other words, our method takes into account heterogeneity-induced competing risk problems in survival analysis. To this method we have developed a software (written in C). In the current project we aim to use our method in order to investigate cohort heterogeneity of data collected within the Västerbotten Intervention Project (VIP, http://www.biobank.umu.se/for-forskare/nshds/) for different causes of death. The VIP includes data of around 100 000 men and women in Västerbotten, measured at 40, 50 or 60 years of age. The data in VIP includes measurements of lifestyle factors and questionnaire of diet, physical activity and social factors. Due to this large amount of data, and the long follow up in VIP, the calculations are very computer-intensive. This project is a collaboration between Umeå University (Department of Biobank Research) and King's college, London (Cancer epidemiology and population health & Institute of mathematical and molecular biomedicine).