assuming that half of true users misclassified as non-users are
taking aspirin at medium-high doses and applying a prevalence
of aspirin use 3.5 times higher in cases than in controls (the
one we found for such doses in our study—Table 2) (see
Appendix 3 in supplementary material for details). In the first
scenario, the impact would be rather small, with an increase in
the crude odds ratio of only 12 % (from 1.79 to 2.01), which is
consistent with the results reported by Ulcickas-Yood et
al. [18]. In the second scenario, the corrected crude
odds ratio would increase to 3.21 (79 % increase), but still
be lower than the one found in hospital-based studies
performed in Spain. The selection of controls in hospitalbased
studies are obtained through a pseudo-sampling, and
this could also be a potential source of bias that may
contribute to the differences.
Notwithstanding these methodological issues, other reasons
may contribute to the lower RR found in our study. Firstly is the