Table 1 Study characteristics and summary risk of bias of the studies included in the five comparisons

Baseline gastrointestinal status* Studies with participants' mean age >65 years Study arms at too high dosage Summary risk of bias Comparison No of studies (participants) Publication dates % of women 1 2 3 4 5 6 Low Medium High H2 receptor antagonist v placebo 15 (2621) 1987-1997 66 4 7 0 2 1 1 0 7 0 13 2 Proton pump inhibitor v placebo 6 (1358) 1996-2002 69 1 3 1 0 1 0 0 1 1 3 2 Misoprostol v placebo 23 (16 945) 1988-2002 68 1 12 1 3 3 3 3¶ 0 1 18 4 COX-2 selective v non-selective 51 (28 178) 1989-2002 65 1 4 0 0 46 0 8** 5 0 34 17 COX-2 specific v non-selective 17 (25 564) 1999-2002 73 0 6 0 0 11 0 0 5 3 13 1

Not stated in three studies. Not stated in two studies. ¶Not stated in two studies. ^**Not stated in four studies.

^* 1=normal gut on endoscopy for all participants; 2=some participants have a normal gut on endoscopy, others have some erosions or haemorrhages, but no frank uclers; 3=all participants have some abnormal symptoms on baseline endoscopy (no ulcers or up to 50% recently healed ulcers); 4=all participants have recently healed ulcers on baseline endoscopy (at least 50% recently healed ulcers); 5=no baseline endoscopy, or no gut status reported; 6=mix, from normal gut on endoscopy to frank ulcers.

Based on assessment of allocation concealment and baseline comparability. If either or both criteria were classed as "inadequate" the summary risk of bias was judged "high," if either or both criteria were "unclear" then summary risk of bias was "moderate," and if both criteria were "adequate" the summary risk of bias was "low."