Skip to main content

Table 4 Statistical suggestions for investigating aspects of clinical heterogeneity

From: Investigating clinical heterogeneity in systematic reviews: a methodologic review of guidance in the literature

General Category of Statistical Method Specific Method Suggested Number of Resources1 Citations
Subgroup analyses General 18 60, 2324, 25,46, 48, 50, 75, 92, 94, 93, 27, 97, 100, 98, 115, 105, 19
  Hierarchical testing procedure based on the heterogeneity statistic Q 1 114
  Combining subgroups across studies (i.e., in stratified studies) 1 114
Moderator Analyses    
1. ANOVA2 analogue (e.g., a categorical moderator)   4 48, 94, 95, 114
2. Meta-regression General mention 16 19, 60, 6, 24, 2528, 31,32,43, 50, 75, 94, 95, 100, 98, 93, 1325, 418
  Fixed effects model (general) 4 92, 93, 94, 95
  Bayesian models (general) 4 66, 71, 124, 95
  New maximum likelihood method 2 60, 124
  New weighted least squares model 2 58, 67
  Random effects model (general) 2 67, 114
  Random effects model for IPD3 2 58, 61
  Permutation-based resampling 2 31, 43
  Other nonparametric (e.g., fractional polynomials, splines) 2 69, 85
  Mixed effects model 2 38, 114
  New variance estimators (for covariates) 2 77, 84
  Methods for measurement of residual errors 2 59, 41
  Bayesian model in the presence of missing study-level covariate data 1 110
  Semi-parametric modeling (general) 1 80
  Fixed effects generalized least squares model 1 68
  Hierarchical regression models 3 60, 64, 124
  Random effects model with new variance estimator 1 70
  Logistic regression with binary outcomes 1 25
  Interaction term for meta-regression model 1 95
  Consider nonlinear relationships (e.g., use quadratic or log transformations) 1 48
  Bayesian model for use in meta-analyses of multiple treatment comparisons 1 111
3. Multivariate analyses   1 48
4. Multiple univariate analyses with Bonferroni adjustments   1 48
5. Meta-analysis of interaction estimates   1 61
6. Model to include the repeated observations (time as a variable) using IPD   1 109
7. Z test   1 125
Bayesian Approaches   
1. Hierarchical Bayesian modeling   2 44, 48
2. Random effects models   1 63
Data Specific Approaches    
1. IPD analyses General 5 75, 76, 95, 97, 23
  Regression 1 61, 46
  Adding a treatment-covariate interaction term 1 95
2. Combination of IPD & APD4 Two-step models 2 74, 78
  Multi-level model 2 69, 100
  Meta-analysis of interaction estimates 1 61
Other Approaches    
Models for control event rate / baseline risk General (e.g., control event rate) 10 63, 24, 71, 81, 79, 93, 100, 19, 78, 111
Structural equation modeling (SEM) Integration of SEM with fixed, random and mixed effects meta-analyses 1 42
Mixed treatment comparisons combined with meta-regression   1 72
Combining regression coefficients from separate studies   1 64
  1. 1. The number (N) of resources equals the percentage of resources since we include 101 total resources; 2. ANOVA = analysis of variance; 3. IPD = individual patient data; 4. APD = aggregate patient data.