Introduction | Part 1: Hierarchy of Evidence | Part 2: Case Reports | Part 3: Cross-Sectional Studies | Part 4: Case Control Studies | Part 5: Cohort Studies | Part 6: Randomized Controlled Studies | Part 7: Meta-Analysis | Part 8: Systematic Reviews
The Random Controlled Trial discussed in the previous section, is the most robust of the three primary research methods used in the medical literature to evaluate study and control groups (the prospective random controlled trial, the prospective cohort study and the retrospective case-control study). To analyze the effects of an intervention, any, or all of these studies may be used. The results of a case-control study may help justify the inherently most costly yet better designed random control study. When multiple studies, using similar or varying research methods, are employed to analyze a similar intervention, there is often divergence and inconsistencies in the results.
Meta-analysis is the statistical combination or mathematical summation of the results of two or more principal studies. This can be a powerful tool to not only resolve conflict between studies, but to combine the power of multiple studies into a single effort. It may improve the reliance on outcome of small studies due to a small sample size. While a meta-analysis may be published alone, they are often included as a foundational methodology in a systematic review. They are indexed in MEDLINE as a publication type in MeSH using the term meta-analysis. For this reason, I’ve placed meta-analysis precariously between RCT’s and Systematic Reviews (and in italics) in the evidence hierarchy diagram because of their multi-purpose usage as either a distinct publication type or a tool used within a systematic review.
Prior to analysis, study methodologies should be similar. Meta-analysis requires a thorough review of the literature and study criteria must be clearly specified.
The meta-analysis typically combines studies into a forest plot (also, blobbogram). In the example to the right ("Impact of Streptokinase as Function of Delay") [Figure 2 pending], each effect from each study is grouped into a single plot. The horizontal lines represent the 95% CI from each study. The odds ratio results are often weighted to account for the varying size of each study. These results are plotted as point estimates (the shaded points within each horizontal line).
Figure 2: See the example forest plot to the right from The role of concurrent chemoradiotherapy in the treatment of locoregionally advanced nasopharyngeal carcinoma among endemic population: a meta-analysis of the phase iii randomized trials; Li Zhang et al; BMC Cancer. 2010; 10: 558.; doi: 10.1186/1471-2407-10-558; PMCID: PMC2970609
Where the odds ratio/relative risk equals 1.0 is the 'line of no effect'. If the point estimates lie on the line of no effect then, then the difference in efficacy between the treatments is negligible. If the line of no effect lies to the left, then outcome of the study favors treatment. If it lies to the right, it favors the placebo group (in this example) or another intervention depending on the study.
Heterogeneity exists if the horizontal lines do not overlap each other. This is important when analyzing the relative compatibility between studies and whether there is significant variation in study designs or methods to disallow a valid comparison between them. Homogeneity occurs when the 95% CI's overlap. Also, outliers, study results with outcomes that fall considerably far to the right or to the left of the line of no effect should be critiqued similarly.
|From 2000-2010, meta-analyses were reported most often in Journal of Dairy Science (17), Theriogenology (5), Journal of Animal Science (4), Preventative Veterinary Medicine (4), JAVMA (3), Veterinary Research (2), Journal of Veterinary Medicine - B - infectious diseases and veterinary public health (1), BMC Genomics (1), Australian Veterinary Journal (1), Canadian Journal of Veterinary Research (1).|
|Top concept representation (MeSH Terms), excluding meta-analysis, were cattle, diet, milk, lactation, animal nutritional physiological phenomena, mastitis (bovine), dogs, dairying, animal feed and pregnancy respectively.|
- Methods for Meta-Analysis in Medical Research, Sutton, AJ, et al, , 2000, John Wiley & Sons, ISBN: 0-471-49066-0
- Introduction to Meta-Analysis (Statistics in Practice), Michael Borenstein et al (2009);
- Tools for performing Meta Analysis:
This series has been loosely organized from a set of lectures given by the author within graduate courses in Biomedical Informatics beginning in 2002. Content is being edited to improve organization, depth, correct inaccuracies as well as updates with new information during Winter/Spring 2012. Feedback is greatly appreciated. © 2011, Stuart Turner.