Synthesising data: meta-analysis and network meta-analysis
Different methods are available to combine or synthesise different data sources available from literature on the same subject, for instance, if one needs to determine the effect of drug A in reducing cardiac events compared to the effect of drug B in reducing cardiac events.
When head-to-head trials for drug A versus B are available, direct comparison is possible using a pairwise meta-analysis. Alternatively, a network meta-analysis might be required, where head-to-head trials are not available for drug A versus drug B directly, but where there are studies comparing drug A to drug C and drug B to drug C.
Pairwise meta-analysis
To statistically combine data on effect size from different sources with direct comparison data, one would typically employ a pairwise meta-analysis technique. Two different options are most widely available, namely a fixed-effects meta-analysis and a random-effects meta-analysis [1].
Fixed effects meta-analyses are usually used where the study does not intend to generalise the results beyond studies included in the analysis. Where there isn’t significant heterogeneity between studies (statistical, methodological, or clinical heterogeneity) and where the number of included studies is low (less than 5 studies)[1].
When one aims to generalise the results to studies outside of those included in the analysis, have significant heterogeneity between studies, or at least 5 studies included in the analysis, one would typically employ a random effects meta-analysis [1].
The outcome of either method would be a pooled effect size. This can be a relative effect size (such as relative risk RR or odds ratio OR for dichotomous variables) or an absolute effect size (such as risk difference RD, for dichotomous variables, or weighted mean difference or standardized mean difference, for continuous variables) [1].
The results of meta-analyses are typically reported as forest plots, showing the effect size and 95% confidence intervals for each study included in the analysis, as well as the pooled or summary effect size and 95% confidence intervals.
Network meta-analysis (indirect treatment comparison)
Indirect treatment comparisons or network meta-analyses (NMA) are performed when direct head-to-head trials are not available for drug A compared to drug B, but where there is a common comparator study for both drugs A and B with drug C. Further to this, multiple treatments can be compared at the same time, and by including direct and indirect evidence (mixed NMA) [2]. In addition, NMA can be used to rank different treatments.
Different NMA models are available, including the multivariate model and the hierarchical model. A reference treatment must be selected, to which all studies will be compared (usually placebo, or the treatment that is most commonly compared to).
The results are presented in a league or network table, which shows the summary estimates and 95% confidence intervals for the different comparisons. In addition, the ranking probabilities for each treatment in the comparison can be presented as a table or histogram.
Conclusion
Study effects can be synthesised using different statistical techniques, such as pairwise meta-analysis for direct comparisons, or network meta-analysis for mixed or indirect comparisons.
One needs to consider the different assumptions and limitations of each method, as well as the goal of the data synthesis, to select the relevant method.
References
-
- Tufanaru C, Munn Z, Stephenson M, Aromataris E. Fixed or random effects meta-analysis? Common methodological issues in systematic reviews of effectiveness. International Journal of Evidence-Based Healthcare. 2015. 13(3): 196-207.
- Rouse B, Chaimani A, Li T. Network meta-analysis: an introduction for clinicians. Intern Emerg Med. 2017: 12(1): 103-111.
