Table 4.

Univariate random effects meta-regression (methods of moments) and subgroup analyses for GENDER (female)

			Subgroup analyses				Results from meta-regression
			Results		Heterogeneity
Variable	Subgroup	Number of studies	OR	95% CI	P-value	I 2	Coefficient	SD	P-value
Outcome measurement	Mixed	2	1.224	0.895–1.674	0.800	0%	Reference
	Formal	9 (13 different samples)	1.061	0.907–1.243	0.000	89%	−0.1424	0.2379	0.550
Proportion of volunteers (%) in baseline study sample	Continuous	9a (13 different samples)	1.099	0.917–1.317	0.000	89%	0.0004	0.0003	0.177
	0–100%	8	1.038	0.805–1.268	0.000	93%	Reference
	0%	2 (3 different samples)	0.918	0.808–1.043	0.836	0%	−0.0689	0.2266	0.761
	100%	3 (4 different samples)	1.306	1.000–1.705	0.296	19%	0.2926	0.2156	0.175
Mean age at baseline	Continuous	10b (13 different samples)	1.109	0.920–1.337	0.000	86%	−0.0000	0.0006	0.952
	≤ 55 years	6 (8 different samples)	1.136	0.939–1.374	0.000	85%	Reference
	>  55 years	6 (7 different samples)	1.023	0.765–1.367	0.000	90%	−0.1296	0.1695	0.445
Continent	USA	6	1.279	1.120–1.460	0.063	52%	Reference
	Europe	5 (9 different samples)	0.906	0.770–1.067	0.000	77%	−0.3531	0.1135	0.002
Year of baseline measurement	Continuous	11c (15 different samples)	1.078	0.931–1.249	0.000	88%	0.0008	0.0126	0.951
	<  2006	9 (11 different samples)	1.083	0.924–1.270	0.000	91%	Reference
	≥ 2006	3 (4 different samples)	1.084	0.775–1.516	0.474	0%	−0.0301	0.2334	0.897
Risk of bias items
Study participation	Unclear/high risk of bias	9 (11 different samples)	1.025	0.871–1.205	0.000	89%	Reference
	Low risk of bias	3 (4 different samples)	1.288	1.094–1.515	0.383	2%	0.2436	0.1809	0.178
Study confounding	Unclear/high risk of bias	1	1.174	0.748–1.842	N.A.	N.A.	Reference
	Low risk of bias	10 (14 different samples)	1.073	0.922–1.250	0.000	89%	−0.0898	0.3302	0.786

^aThe studies of Ajrouch et al. (2014) and Voorpostel & Coffé (2014) are not included in this analysis, because the proportion of volunteers (%) in the baseline study sample is not reported

^bThe study of Hank & Erlinghagen (2010) is not included in this analysis, because the mean age at baseline is not reported

^cThe study of Broese van Groenou & Van Tilburg (2012) includes two different samples in the analyses. For one of the samples, the year of baseline measurement is 1992, for the other sample, the year of baseline measurement is 2002. No separate results for the two samples are provided. In this specific analysis, we took 1992 as the year of baseline measurement, although this actually only is the case for the first sample