Explanation of sensitivity analysis with observation time truncated at the time of the event.
Observed IRRs of drug monotherapy for the main analysis in SCCS and sensitivity analysis. A key assumption of the SCCS is
that the exposure distribution within the observation period and the observation period itself must be independent of prior
event times. This could have been violated for some subjects, because use of an nsNSAID without gastroprotection is
relatively contraindicated after UGIB. By truncating follow-up at the time of UGIB, we observed that the IRRs changed in
magnitude for some drugs. We used a change of 10% of the initial estimate as an arbitrary cutoff to quantify the magnitude of
the change. The estimates were higher for monotherapy of corticosteroids, SSRIs, GPAs, aldosterone antagonists, antiplatelets,
and nitrates and lower for nsNSAIDs and low-dose aspirin in the analysis with truncation of follow-up time at the
event. The estimates did not change by more than 10% of the initial estimate only for COX-2 inhibitor monotherapy. These
analyses indicate that the exposure of, for instance, nsNSAIDs did not change significantly after the event but show the relative
contraindication of nsNSAIDs after UGIB, because the IRR in the initial analysis was higher than that in sensitivity analyses
(Supplementary Figure 1). However, these analyses show that confounding by contraindication is unlikely