Model-based estimation of the excess fraction (attributable fraction): day care and middle ear infection

The methods of adjustment for estimation of the excess fraction (EF), or attributable fraction, based on conventional and dynamic (or regressive or autoregressive) logistic regression modelling in a cohort study for a disease with recurrent episodes are considered. Throughout the paper, the computation of estimates with corresponding confidence intervals is illustrated in a cohort study associating the incidence of acute middle ear infection (acute otitis media) with type of day care in northern Finland. The incidences of at least one episode and at least three episodes as well as the total number of episodes during the first two years of life are considered. In our example, the estimates obtained from the dynamic model appear to have smaller standard errors since the dynamic model effectively utilizes the time-dependency of the covariates.