Skip to main content
. 2011 Nov 7;343:d6829. doi: 10.1136/bmj.d6829

Table 3.

 Summary design difference* estimates of the 31 meta-analyses that examined biomarkers for cardiovascular risk and included data from at least one observational study (OS) and one randomised controlled trial (RCT)

Comparison No of meta‑analyses Design difference (95% CI) I2 (95 % CI) P value (Q)
Random effects Fixed effects
All data 31 0.24 (0.07 to 0.40) 0.33 (0.23 to 0.43) 39 (6 to 60) 0.015
Design contrast:
 RCT v prospective OS 31 0.23 (0.07 to 0.40) 0.33 (0.23 to 0.43) 37 (2 to 59) 0.023
 RCT v retrospective OS 9 0.34 (−0.02 to 0.70) 0.34 (−0.02 to 0.70) 0 (0 to 65) 0.457
Type of meta-analysis:
 Of individual participant data 7 0.42 (0.14 to 0.70) 0.51 (0.37 to 0.65) 50 (0 to 79) 0.064
 Of published literature 24 0.15 (−0.01 to 0.31) 0.16 (0.02 to 0.30) 10 (0 to 44) 0.320
Type of RCT:
 Effective intervention† 21 0.14 (−0.08 to 0.37) 0.34 (0.22 to 0.46) 52 (20 to 71) 0.003
 Other 23 0.31 (0.07 to 0.54) 0.27 (0.12 to 0.42) 42 (4 to 64) 0.02
Significant biomarker effect:
 Yes 27 0.22 (0.05 to 0.39) 0.33 (0.23 to 0.43) 45 (14 to 65) 0.006
 No 4 1.19 (−0.48 to 2.86) 1.19 (−0.48 to 2.86) 0 (0 to 85) 0.901
Biomarker recommended for clinical use:
 Yes 16 0.22 (−0.01 to 0.45) 0.26 (0.12 to 0.39) 55 (20 to 74) 0.005
 No 15 0.43 (0.28 to 0.58) 0.43 (0.28 to 0.58) 0 (0 to 54) 0.502

*The difference in biomarker effects between datasets from observational studies and from randomised controlled trials as a proportion of the summary effect of each meta-analysis.

†Refers to trials showing significant difference between treatment arms for cardiovascular disease end points and populations of both arms were considered in analysis of the biomarker.