Table 1.

Predictive inverse‐gamma (IG) distributions for between‐study variance τ ² expected in future binary outcome meta‐analyses of log odds ratios. To implement a predictive distribution as an informative prior for τ ² in an approximate Bayesian meta‐analysis by data augmentation, we augment the observed study data using the pseudo data reported.

	Pharmacological versus	Pharmacological versus	Non‐pharmacological
	placebo/ control	pharmacological	(any)
	I G(1.06,0.01)	I G(2.93,0.00003)	I G(0.80,0.007)
All‐cause mortality	Pseudo data:	Pseudo data:	Pseudo data:
	2 studies with	6 studies with	2 studies with
	effects y 0 = 0.100	effects y 0 = 0.003	effects y 0 = 0.084
	I G(1.32,0.08)	I G(1.04,0.04)	I G(0.88,0.05)
Semi‐objective	Pseudo data:	Pseudo data:	Pseudo data:
	3 studies with	2 studies with	2 studies with
	effects y 0 = 0.231	effects y 0 = 0.200	effects y 0 = 0.224
	I G(1.45,0.18)	I G(1.13,0.09)	I G(1.39,0.13)
Subjective	Pseudo data:	Pseudo data:	Pseudo data:
	3 studies with	2 studies with	3 studies with
	effects y 0 = 0.346	effects y 0 = 0.300	effects y 0 = 0.294