Table 1. Association between vibration of effects indices and various characteristics of treatment comparisons.

Characteristics of treatment comparisons	Janus effect		Statistically significant results in opposite directions		Log (relative OR)
	OR (95% CI)	P value	OR (95% CI)	P value	β (95% CI)	P value
Treatment response
Type of available evidence for the comparison (ref=direct evidence without inconsistency)
Direct evidence with inconsistency	0.393 (0.149 to 1.016)	0.055	0.596 (0.126 to 2.117)	0.458	0.041 (−0.109 to 0.190)	0.593
Only indirect evidence	3.176 (1.587 to 6.492)	0.001*	2.225 (1.109 to 4.699)	0.029*	0.406 (0.313 to 0.498)	<0.001*
Effect size*	0.037 (0.004 to 0.373)	0.005*	0.004 (0.000 to 0.092)	0.002*	−0.072 (−0.396 to 0.252)	0.660
Treatment discontinuation
Type of available evidence for the comparison (ref=direct evidence without inconsistency)
Direct evidence with inconsistency	0.643 (0.221 to 1.943)	0.4217	0.906 (0.128 to 4.113)	0.907	1.134 (0.941 to 1.366)	0.184
Only indirect evidence	2.773 (1.281 to 6.141)	0.0103*	3.641 (1.600 to 9.433)	0.004*	1.632 (1.456 to 1.828)	<0.001*
Effect size*	3.25×10−6 (2.68×10−8 to 2.56×10−4)	<0.001*	0.063 (0.001 to 4.329)	0.214	0.667 (0.350 to 1.272)	0.217

*p-value <0.05

^*Effect sizes are expressed as absolute values of the log relative odds-ratioOR estimated in the most complete network meta-analysis. Relative ORs quantify the ratio of ORs at the 1st and 99th percentile. The higher the relative odds ratioOR, the greater the variability of results arising from the network geometries considered.