Abstract

Studies focusing on mortality data use a wide variation of strategies for data analyses, making comparison between studies difficult. The research problems focus upon different statistical analyses of mortality among patients and matched controls regarding clustered data and relations over different levels of age and follow-up time. Four hundred and twenty (420) treated female alcoholics were compared to 2,036 matched controls and public register data for a follow-up period of 27 years were used. The statistical analyses are multilevel, structural equation modeling (SEM) level-and-difference analyses, multilevel Cox regression analysis, interaction Cox models, time-dependent Cox survival models, proportional and non-proportional latent discrete-time survival models. The multilevel analyses confirm the success of the matching procedure. The interaction model adds more information to the main effect model and shows the mortality estimate to be dependent on age. Continuous time-dependent Cox regression models and latent discrete-time survival analyses show the mortality estimates to differ with time and age. Different results depend on statistical models. This illustrates how mortality as a construct not only represents hard and unequivocal evidence given by the samples studied, but also includes factors related to the statistical model used. Such methodological factors need to be incorporated in the scientific discussion of mortality studies generally.

Key words: Continuous and discrete time survival analysis, interaction models, matched data, mortality.