Modelling high dimensional sets of binary co-morbidities

The construction of classical co-morbidity indices is described. When the co-morbidities are binary we advocate the use of log-linear models which better capture the dependence structure in the data. We use R to implement new search strategies which enable us to analyse, sparse, high dimensional contingency tables rapidly and hence identify the best fitting models. We apply our new algorithms to a set of real medical data.