Lecture 3 - Meta-analysis Flashcards
Clinical heterogeneity
Differences in population, context and intervention administration
Statistical heterogeneity
Degree to which study estimates are more different from each other than we would expect on the basis of chance alone
I-squared value
Percentage of variability between studies which goes beyond difference expected due to chance
I2: 0-40%
Might not be important
I2: 30-60%
May represent moderate heterogeneity
I2: 50-90%
May represent substantial heterogeneity
I2: 75-100%
Considerable heterogeneity
Cochran’s Q
Tests for statistically significant heterogeneity
Fixed effects models
common intervention effect, with variation only due to sampling
Random effects models
‘distribution’ of intervention effects, with important between-study variance (tau-squared)
Fixed effects analysis with lots of heterogeneity
Misleading in its precision
Random effects analysis with no heterogeneity
May be too conservative
Fixed effects: common effect is
Specific intervention Specific population Variation only from sampling error No (or very little) between-study variance (tau-squared) I2 should be low
Random effects: On average, the effect is
Class of interventions
Broad population
variation due to uncaptured heterogeneity
Significant and meaningful between-study variance (tau-squared)
I2 can be high but effect sizes must be independent
Framework for understanding and interpreting meta-analyses
Examination of included studies
Extraction of data
Expectation, exploration and exegesis of heterogeneity
Explanation of pooled effect sizes
