The following were included:
15 RCTs involving 19,364 patients taking NSAIDs and 21,053 control receiving a placebo or NSAID plus a protective agent;
3 cohort studies with 1,131,174 patients, of whom 215,076 were exposed to NSAIDs;
6 case-control studies involving 2,957 cases with a diagnosis of gastroduodenal ulcer, or bleeding or perforation;
20 uncontrolled series involving 7,406 patients with a diagnosis of gastroduodenal ulcer, or bleeding or perforation; and
4,447 case reports to drug monitoring centres on patients with upper gastrointestinal complications.
Event rate scatter (RCTs and cohort studies): a consistently increased risk was noted with NSAIDs for all three levels of harm (ulcer, bleed or perforation, and death). The L'Abbe plot was presented.
Endoscopically diagnosed ulcer (12 RCTs of 4,111 patients with arthritis): the average absolute risk for an endoscopically diagnosed ulcer with NSAID was 21%. The trials did not report any patients with a more serious level of harm.
Clinically diagnosed (symptomatic) ulcer (3 RCTs and 1 cohort study): the incidence in the RCTs was significantly higher than in the cohort study; the average incidence was 1.48% (range: 1.35 - 1.53) for the RCTs versus 0.39% for the cohort study.
Ulcer bleeding or perforation (4 RCTs and 2 cohort studies): the incidence in the RCTs was significantly higher than in the cohort studies; the average incidence was 0.48% (range: 0.34 - 0.86) for the RCTs versus 0.22% (range: 0.06 - 0.33) for the cohorts.
Death attributable to ulcer bleeding or perforation (2 RCTs and 1 cohort study): the pooled incidence was 0.008% (range: 0.007 - 0.023).
The pooled ulcer bleed/perforation ratio was 3.3 (9 studies involving 1,769 patients with the 3 RCTs (231 patients with ulcer and 84 bleeds/perforations) to 6.0 in the cohort study (131 patients with ulcer and 22 bleeds/perforations).
The pooled bleed/perforation death ratio was 8.3 (50 studies involving 11,040 patients with bleed or perforation). This ranged from 6.3 (23 case reports with 3,669 bleeds/perforations and 585 deaths) to 35.5 (2 RCTs with 71 bleeds/perforations and 2 deaths). <RESULTS OF THE REVIEW> The NNT with NSAID for one patient to die was 1,220 patients (range: 909 - 2,500) when using the first method with data from RCTs for bleed or perforation. The NNT was 1,960 patients (range not reported) when using the second method, based on outcome of symptomatic ulcer.