Thirty-six case-control studies (19,648 cases and 105,373 controls) and 8 cohort studies (approximately 400,000 exposed people and 1 million non-exposed people) were included.
The overall quality score was not associated with marked differences in ORs. However, low scores for the selection of controls, definitions of controls, comparability of cases and controls, and ascertainment of exposure tended to inflate the estimated risk of complications.
The use of NSAIDs significantly increased the risk of serious gastrointestinal complications compared with non-use. The unadjusted pooled OR was 4.06 (95% CI: 3.47, 4.75). Significant heterogeneity was detected (P<0.00001). After excluding studies using aspirin, the pooled OR (25 studies) was 3.81 (95% CI: 3.17, 4.58), still with significant heterogeneity (P<0.00001).
The use of NSAIDs in cohort studies significantly increased the risk of serious gastrointestinal complications compared with non-use. The unadjusted pooled OR was 2.29 (95% CI: 1.50, 3.51). Significant heterogeneity was detected (P<0.00001). None of the cohort studies used aspirin.
Studies showed that ibuprofen was associated with a significantly lower risk than non-aspirin NSAIDs.
The unadjusted ORs for individual NSAIDs ranged from 1.81 (95% CI: 1.3, 2.4) with ibuprofen (16 studies) to 7.46 (95% CI: 5.1, 10.9) for piroxicam (15 studies). The adjusted ORs were slightly higher but the rank order was the same (ibuprofen, aspirin, naproxen, diclofenac, indomethacin, ketoprofen and piroxicam).
Comparisons of the individual drugs with ibuprofen showed that aspirin was not significantly different, but the other NSAIDs were associated with an increased risk of complications. The RR was 1.63 (95% CI: 0.64, 4.2) for aspirin, 1.78 (95% CI: 1.6, 2.2) for naproxen, 1.73 (95% CI: 1.4, 2.1) for diclofenac. 1.88 (95% CI: 1.6, 2.3) for indomethacin, 2.45 (95% CI: 1.8, 3.3) for ketoprofen, and 3.21 (95% CI: 2.6, 4.0) for piroxicam.
There appeared to be a rank order of NSAIDs at low doses but the ORs (except for piroxicam) tended to converge at high doses.