Now Playing: Meta-Analysis for Two Proportions Comparisons (4:05)
A meta-analysis seeks to systematically review all pertinent evidence, provide quantitative summaries, integrate results across studies, and provide an overall interpretation of these studies.
We will consider a dataset that contains summary data for 34 randomized clinical trials that were conducted to study the effects of cholesterol-lowering treatments. Nine of the studies used diet as the treatment, 23 used a cholesterol-lowering drug, and 2 used surgery. The mortality of patients in each treatment arm and each control arm were recorded.
In the procedure, the columns are specified according to the event type. An optional Label Variable is included, and the Treatment column identifies the treatment type.
Next, we consider the method used to combine treatment effects.
The Fixed Effects method assumes homogeneity of study results, where there is one effect size underlying all the studies. This assumption may not be realistic when combining studies with different patient pools, protocols, follow-up strategies, doses, durations, etc.
The Random Effects model method is used here, since it is assumed that each study is performed independently and has its own effect size.
When the procedure is run, we can go to the 3rd and 4th reports to see the results of the significance tests. The combined effect of cholesterol-lowering drug studies gives a highly significant odds ratio, perhaps in part to there being a high number of studies in this group. The diet group also shows significance for the nondirectional test. The overall-combined treatment is also highly significant, probably driven by the drug studies.
The effect-equality tests for both diet and drug treatments are significant. These further assert that the Random Effects Model was the proper model choice for this analysis.
The Forest Plot summarizes the odds-ratio results of all 34 studies. The size of the plot symbol is proportional to the sample size of the study. The points on the plot are sorted by group and by the odds ratio. The lines represent the confidence intervals about the odds ratios.
The radial (or Galbraith) plot shows the z-statistic for each study on the vertical axis and a measure of weight on the horizontal axis. Studies that have the largest weight are closest to the Y axis. Studies within the limits are interpreted as homogeneous. Studies outside the limits may be outliers.
The final plot displays the treatment risk on the vertical axis versus the control risk on the horizontal axis. Homogenous studies will be arranged along the diagonal line. This plot is especially useful in determining if the relationship between the treatment group and the control group is the same for all values of the control group risk.