Table 2 |
||||
|
Demonstration of Simpson's paradox in the pooling of data from three observational studies, showing the value of meta-analysis |
||||
|
Mortality |
||||
|
|
||||
|
Treatment |
Placebo |
Risk ratioa |
Significance |
|
|
|
||||
|
Study 1 |
97/492 = 19.7% |
202/802 = 24.8% |
0.794 |
0.03 |
|
Study 2 |
590/795 = 74.2% |
410/510 = 80.4% |
0.924 |
0.01 |
|
Study 3 |
300/610 = 49.2% |
286/490 = 58.4% |
0.843 |
0.002 |
|
Pooled data |
987/1,897 = 52.0% |
898/1,812 = 49.6% |
1.050 |
0.13 |
|
|
||||
|
Meta-analysis of the above data (using [40]), fixed-effects model: Q statistic = 3.398, P = 0.18; combined risk ratio = 0.898, P = 0.0001. aTreatment/placebo. |
||||
|
Reade et al. Critical Care 2008 12:220 doi:10.1186/cc6941 |
||||