The RR associated with non-aspirin NSAIDs has been
published to range from 2.8 to 7.4 [13, 14], the variability
depending on the study design and, particularly, on the source
of information (lower estimates in computer-based studies than
in hospital-based studies). In our study the RR with non-aspirin
NSAIDs was lower than hospital-based studies performed previously
in Spain (ranging from 5.3 to 7.4 [15–17]). This result
can be partly explained by the non-recorded over-the-counter
use of aspirin (at medium-high doses) that would lead to the
consideration of subjects who are actually taking this drug as
non-users. To evaluate the impact of such a misclassification
error, we performed a sensitivity analysis simulating two extreme
scenarios: a) a non-differential misclassification error as
large as 50 % among cases and controls (that is, 50 % of true
aspirin users are erroneously considered as non-users); and b) a
differential misclassification error as large as 50 % among
controls and 62 % among cases, the latter figure resulting from