Statins and outcomes in patients with pneumonia

The work by Majumdar et al is a major step forwards in dealing with confounding of uncertain direction and magnitude in studying pleiotropic effects of statins.¹ However, we believe that despite the limited statistical power of the study, the investigators should have planned subgroup analyses for institutionalised and non-institutionalised patients separately as, for example, in influenza vaccine studies. When examining table 1 it becomes clear that 20% of the non-statin users and only 8% of the statin users were nursing home residents (P<0.001).

Nursing home residents have other risk profiles than community dwelling people, and uptake of medication is determined by many factors other than their absolute risk of outcome. Importantly, in-hospital treatment may also be different for the separate groups. Including such a subgroup of study subjects may therefore distort the association under study in such a way that confounding cannot be effectively controlled for or, even worse, more unobserved bias through external interventions may be introduced. Apart from statistical control for observed confounders, it is of importance to note that increasing the numbers of covariates in the model will decrease precision of the adjusted estimate of association. We are, however, currently not aware of any formal power calculation for non-randomised studies including the number of covariates.