Heterogeneity in systematic Review Meta-analysis:
Variability among studies which included in the systematic review termed as heterogeneity.
Why to read heterogeneity ?
Studies included in the meta-analysis vary methodologically, clinically, statistically across individual studies. Presence of heterogeneity affects the conclusion of the study and the review. Therefore, our findings may be misleading.
How does heterogeneity impact meta-analyses?
If heterogeneity is present among included studies our summary estimates could be overestimated or underestimated.
How is it present in the included studies?
Heterogeneity may be present in the study due to clinical variation, methodological variation and statistical variation among the individual studies.
1) Clinical heterogeneity- Variability in the studies on participants, intervention, outcomes.
2) Methodological heterogeneity – Variability in the studies on study design, outcome measurement tools and risk of bias. If RCT studies (use of blinding and concealment of allocation sequence)
3) Statistical heterogeneity – variability on methodological diversity and outcome evaluation (studies do not estimate the same quantity) among the individual studies.
Heterogeneity must be considered in the meta-analysis. For that we have two models in the meta-analysis.
1) Random effect model- (heterogenous) · It will average effect estimate · True effect estimates are not equal across the studies. In this case, the random effect model will assume effects follow the normal distribution and allow heterogeneity to interpret the findings. · But it is always preferable to explore heterogeneity. · If heterogeneity is present across studies, the confidence interval of summary effect estimates is wider than the fixed effect model.
2) Fixed effect model- (homogenous) · It will provide typical effect estimate · True effect estimates are equal across the studies · There is no presence of statistical heterogeneity
Note: When we found homogenous among the individual studies both effect model weightage and summary estimates will be equal
Identifying the heterogeneity:
1) Poor overlap of confidence intervals in individual studies which indicate clinical heterogeneity.
2) Low p value which indicates evidence of presence of heterogeneity. (p value 0.10)
3) I2 value helps to identify the heterogeneity.
But we should not confirm the presence of heterogeneity by using only the I2 value when we have few individual studies in our review. · 0% to 40%: might not be important; · 30% to 60%: may represent moderate heterogeneity; · 50% to 90%: may represent substantial heterogeneity; · 75% to 100%: considerable heterogeneity.
Strategy for addressing heterogeneity:
If heterogeneity presents across studies, use a random effect model to interpret the findings because of the random effect model considering summary effect estimates from mean of individual studies estimates. · It is only possible when we have reason to understand the presence of heterogeneity.
There are other way addressing heterogeneity:
1) Check the data entered correctly
2) Do not do the meta-analysis
3) Explore the heterogeneity (subgroup analysis and meta-regression)
4) Ignore the heterogeneity by choosing fixed effect model
5) Perform random effect meta-analysis- It is intended primarily for heterogeneity (But this is not substitution for investigation of heterogeneity)
6) Exclude the outlying studies
Investigating heterogeneity:
1) Subgroup analysis
2) Meta-regression
Subgroup analysis:
Analysis will provide estimates on subset of the population or participants, subset of studies to investigate heterogeneous results. (ex: male and female, different geographical location) also it answers specific questions about particular patient groups, types of intervention or types of study. Significance will be determined by non-overlapping different subgroups’ confidence intervals from overall confidence intervals. If confidence intervals widely overlap each other, there is no significance in the subgroups.
Meta-regression:
Meta-regression is similar to simple regression in which a dependent variable is predicted according to one or more independent variables. But in meta-regression outcome variable is effect estimate and the explanatory or independent variable is characteristics of studies which influence the overall estimates. 1
Reference:
1. Chapter 10: Analysing data and undertaking meta-analyses | Cochrane Training.
https://training.cochrane.org/handbook/current/chapter-10#_Ref180060294