Table 2.

Data analysis summary.

Outcome	Dataset	N	Effect size		One study removed	Heterogeneity
			Std diff in means ± Std error	p-value	p-value	I-squared	Q statistic p-value
Body weight	extended	75	0.52 ± 0.11	1.10-6	1.10-6	75.1	<0.001
	main	49	0.81 ± 0.13	<1.10-8	<1.10-8	67.1	<0.001
Body fat	extended	63	1.26 ± 0.13	<1.10-8	<1.10-8	69.9	<0.001
	main	47	1.38 ± 0.16	<1.10-8	<1.10-8	70.8	<0.001
Leptin	extended	43	0.93 ± 0.12	<1.10-8	<1.10-8	61.8	<0.001
	main	26	1.02 ± 0.10*	<1.10-8	<1.10-8	32.6	0.06
Glucose	extended	50	0.49 ± 0.11	2.10-5	2.10-5	61.6	<0.001
	main	33	0.64 ± 0.16	9.10-5	9.10-5	68.8	<0.001
Insulin	extended	52	0.99 ± 0.15	<1.10-8	<1.10-8	78.5	<0.001
	main	33	1.37 ± 0.22	<1.10-8	<1.10-8	81.0	< 0.001
HDL-c	extended	21	−0.28 ± 0.19	0.1	0.1	59.8	<0.001
	main	14	−0.21 ± 0.29	0.5	0.5	70.6	<0.001
Triglycerides	extended	44	0.54 ± 0.14	0.0001	0.0001	70.1	<0.001
	main	26	0.83 ± 0.17	1.10-6	1.10-6	60.0	<0.001
SBP	extended	14	1.26 ± 0.16*	<1.10-8	0.0003	0.0	0.5
	main	12	1.52 ± 0.18*	<1.10-8	<1.10-8	0.0	1.0

For each outcome, effect size stands for Cohen’s standardized difference in means (D), which was the difference of means between groups (experimental vs. control) divided by the common within-group SD. We used a random-effect model if heterogeneity was observed, while the fixed-effect model (*) was applied in the absence of heterogeneity. We performed sensitivity analyses by omitting one study at a time and calculating the pooled effect size for the remainder of the studies. Heterogeneity was evaluated with the Q statistic and I-squared statistic.