Type of analysis | Main model(s) used | Comments |
Narrative or qualitative description of evidence | Narrative systematic review | - Used if study data are not amenable to meta-analysis (due to differences in populations, interventions, comparisons, outcomes, or study designs).
|
Pooling of multiple studies to provide an overall effect estimate | Meta-analysis using random effects model | - Incorporates between-study heterogeneity.
- Appropriate model for most meta-analyses due to inherent differences and variability of included studies.
- When studies are heterogeneous, confidence intervals will be wider with random effects model compared with fixed effect model.
|
Meta-analysis using fixed effect model | - Assumes a single truth across populations and homogeneity among studies.
- Rarely appropriate for meta-analysis due to inherent differences and variability of included studies.
- A specific form of fixed effect model (Peto's odds ratio) may be appropriate for meta-analysis of rare events.
|
Subgroup analyses | Random effects model (or rarely, Peto's odds ratio) | - Estimates treatment effects for each subgroup.
- May provide a possible explanation for heterogeneity.
- Subgroups should be based on study level factors (not aggregate patient features) to avoid ecological fallacy.
- May not be possible if reported data are sparse or missing.
- Arbitrary selection of subgroups may result in spurious findings.
- Should be analyzed by conducting statistical testing for interaction (determining a p value for the between-group difference).
|
Regression across studies | Meta-regression | - Tests interaction between various factors and treatment effect.
- Allows for multivariable analysis.
- May provide a possible explanation for heterogeneity.
- Factors analyzed should be based on study level factors (not aggregate patient features) to avoid ecological fallacy.
- May not be possible if reported data are sparse or missing.
- Arbitrary selection of multiple factors for analysis may result in spurious findings.
|
Meta-analysis of multiple interventions simultaneously | Network meta-analysis | - Estimates relative effects of multiple interventions compared with each other.
- Allows for indirect comparisons of interventions that have not been directly compared in individual studies.
- May be able to rank effectiveness of various interventions.
- Requires a sufficient number of studies performed in similar patient populations and settings with consistent results across studies to allow analysis and meaningful interpretation.
|
Pooled analysis of data across individuals from multiple studies | Individual patient data meta-analysis | - Allows most complete analysis of data and evaluation of heterogeneity.
- Requires collaboration across research groups and willingness to share data, which may result in incomplete inclusion of study data.
- Can be costly and resource intensive.
|