Table 3.

 Univariate analyses of associations between favourable results or conclusions and characteristics of meta-analysis

Characteristic	Favourable results		Favourable conclusions
	Odds ratio (95% CI)	P value	Odds ratio (95% CI)	P value
	Financial ties with one drug company	0.65 (0.31 to 1.36)	0.25		4.09 (1.30 to 12.83)	0.016
Better quality*	1.16 (1.07 to 1.27)	<0.001		1.02 (0.93 to 1.12)	0.64
Published in journal supplement	0.40 (0.16 to 1.05)	0.062		2.58 (0.56 to 11.93)	0.23
Searched or included non-English literature	0.74 (0.21 to 2.56)	0.63		0.61 (0.15 to 2.49)	0.49
Described process of data abstraction	2.09 (0.81 to 5.41)	0.13		1.07 (0.36 to 3.20)	0.90
Included non-randomised controlled trials	0.51 (0.23 to 1.11)	0.088		1.09 (0.41 to 2.90)	0.86
Included unpublished studies	2.61 (0.28 to 24.09)	0.40		†	†
Included studies that used only placebo group as control group	1.67 (0.67 to 4.18)	0.28		3.67 (0.81 to 16.71)	0.093
Focused on newer class of drug	0.48 (0.23 to 1.00)	0.050		1.91 (0.73 to 5.01)	0.19
Used surrogate outcomes only	0.48 (0.23 to 1.00)	0.050		0.97 (0.39 to 2.40)	0.95
Used composite outcomes only	0.31 (0.03 to 3.48)	0.34		0.47 (0.04 to 5.40)	0.54
Carried out evaluations of heterogeneity	2.65 (1.25 to 5.59)	0.011		1.00 (0.41 to 2.44)	1.00
Carried out sensitivity analyses	3.61 (1.69 to 7.71)	<0.001		1.12 (0.46 to 2.75)	0.80

Odds ratios >1 indicate that characteristic is associated with more favourable results or conclusions, whereas odds ratios <1 indicates that characteristic is associated with less favourable results or conclusions.

*Quality score was continuous variable from 0-18, with higher scores indicating better quality and lower scores indicating worse quality.

†Model was not possible because observations at one level of the predictor were all same—that is, all meta-analyses that included unpublished studies had favourable conclusions.