To the editor,

We have read with interest the article “Clinically relevant drug-drug interactions between elderly people with dementia” [1] published by Sönnerstam E et al. The study reveals important results regarding the prevalence and characteristics of pharmacological interactions and their associated factors in elderly patients with the diagnosis of dementia or cognitive impairment.

Of note, 43.2% of the population studied had at least one clinically relevant pharmacological interaction. This is important when carrying out the statistical analysis because the odds ratio as a measure of association tends to overestimate the magnitude of the effect in case of high prevalences [2, 3] such as the one found. In these cases, it is preferable to use the prevalence ratio (PR) as a measure of association which can be calculated in most statistical packages and can even be obtained from multivariate analysis with statistical routines such as Poisson regression with robust variance [4]. As example, the OR found for the number of medications upon admission was 1.31, which would correspond to a PR of 1.16, showing clearly the overestimation of the effect magnitude.

The use of the prevalence ratio would provide an estimate with greater interpretability from the clinical point of view, so it should be used routinely to evaluate the association between dependent and independent variables in cross-sectional studies in which outcomes are frequent as in this study.