Table 3. Fixed- and random-effects model results.

Includes the non-pharmaceutical and pharmaceutical study sets and the three corresponding sensitivity analysis sets for each.

	Non-pharmaceutical study set
	Original (k = 50)	Outlier removal (k = 49)	Trim-and-fill (k = 59)	With unreported (k = 51)
Fixed-effects model
${\overline{\theta }}_w$ (95% CI)	0.340 (0.318 to 0.363)	0.351 (0.329 to 0.374)	*	0.331 (0.308 to 0.353)
Q (p-val)	314.930 (<0.0001)	192.145 (<0.0001)	*	314.620 (<0.0001)
Random-effects model
μ (95% CI)	0.402 (0.307 to 0.496)	0.423 (0.356 to 0.490)	0.320 (0.223 to 0.418)	0.381 (0.289 to 0.473)
τ2 (95% CI)	0.093 (0.060 to 0.166)	0.038 (0.023 to 0.085)	0.119 (0.081 to 0.204)	0.090 (0.059 to 0.162)
I2 (95% CI)	92.74% (89.15 to 95.80%)	84.03% (76.47 to 92.21%)	93.49% (90.70 to 96.11%)	92.63% (89.11 to 95.76%)
	Pharmaceutical study set
	Original (k = 47)	Outlier removal (k = 47)	Trim-and-fill (k = 51)	With unreported (k = 51)
Fixed-effects model
${\overline{\theta }}_w$ (95% CI)	0.104 (0.103 to 0.105)	**	*	0.104 (0.103 to 0.105)
Q (p-val)	1835.407 (<0.0001)	**	*	1840.086 (<0.0001)
Random-effects model
μ (95% CI)	-0.114 (-0.212 to -0.016)	**	-0.071 (-0.175 to 0.034)	-0.102 (-0.191 to -0.013)
τ2 (95% CI)	0.090 (0.058 to 0.177)	**	0.113 (0.076 to 0.223)	0.080 (0.053 to 0.157)
I2 (95% CI)	99.78% (99.66 to 99.89%)	**	99.81% (99.72 to 99.90%)	99.73% (99.59 to 99.86%)

${\overline{\theta }}_w$: Average true effect of the set of studies included in the analysis.

Q-statistic: Measure of heterogeneity. Calculated as the weighted sum of squared differences between individual study effects and the pooled effect across studies, with the weights being those used in the pooling method chi-square statistic with k-1 degrees of freedom μ: Average true population effect size.

τ²: Total amount of heterogeneity among the true effects.

I²: Percent of the total variability in effect size estimates due to heterogeneity among the true effects.

* Trim-and fill analysis results for fixed-effects model not relevant due to non-homogeneous effect size distribution.

** Results not shown because study set unchanged from original.