Use of a multinomial additive hazards model to assess the disability burden using cross-sectional data

Abstract

The global phenomenon of population ageing results in a larger number of elderly individuals living with disability. Few methods have been proposed to address the impact of chronic diseases on disability using cross-sectional data. The attribution method based on the additive hazards model for binary outcomes is an attractive option, as it enables the partition of the disability prevalence into additive contributions of chronic diseases. The link function used in the model imposes a constraint on the parameter space, which limits the use of the method, as it is not available in standard software. In this study, we propose an extension of the binomial model to a multinomial response and the use of linear constrained optimization to estimate the disability rates using the R function "constrOptim". For illustration, we assess the contribution of chronic diseases to the disability prevalence using the data from the Belgian Health Interview Surveys of 2001, 2004, and, 2008. The models proposed can be used by health professionals to provide information to assist policy-makers on the development of strategies to tackle the disability burden. Further research should focus on the extension of the multinomial model to ordinal multi-category outcomes.