Most published meta-regression analyses based on aggregate data suffer from methodological pitfalls: a meta-epidemiological study

Table 3 Association between any inappropriate meta-regression and review characteristics

Any potential meta-regression pitfall	Yes (n = 57)	No (n = 24)	Odds Ratio (95% CI)
Published in 2012	36 (63%)	16 (67%)	0.87 (0.33 to 2.34)
Journal characteristics
Core clinical journals	12 (21%)	7 (29%)	0.64 (0.22 to 1.85)
General medical journals	11 (19%)	4 (17%)	1.13 (0.34 to 3.77)
Impact factor higher than median	29 (51%)	11 (46%)	1.22 (0.47 to 3.11)
Author characteristics
Affiliated with industry	5 (9%)	0 (0%)	5.13 (0.27 to 96.57)
Affiliated with biostatistics or epidemiology department	23 (40%)	12 (50%)	0.68 (0.27 to 1.75)
Ten or more of studies	53 (93%)	20 (83%)	2.61 (0.64 to 10.61)
Drug intervention	28 (49%)	10 (42%)	1.33 (0.52 to 3.44)
Binary outcome variable	25 (44%)	14 (58%)	0.57 (0.22 to 1.47)

Odds ratios are for the comparison of meta-regression analyses with the characteristic as compared to meta-regression analyses without the characteristic. An odds ratio of 2.61 for ‘Ten or more studies’ indicates, for example, that the odds of any potential meta-regression pitfall is 2.61 times higher in meta-regression analyses that include 10 or more studies as compared with meta-regression analyses that include a lower number of studies

ISSN: 1471-2288