It is generally accepted to take the average effect size (e.g., correlation coefficient) if a study reports more than one effect size in meta-analysis. Can I apply the same rule for time-to-event data, where the log hazard ratio and its variance are of interest? Can I take the average of the variances of the effect sizes and use that value as the variance of the averaged effect size?
Dear , In time-to-event meta-analysis, it’s generally not appropriate to take a simple average of the variances from individual studies. Variances in these studies are influenced by factors like sample size, follow-up duration, and censoring rates, which vary widely between studies. Instead, meta-analyses typically use inverse-variance weighting, where studies with more precise estimates (lower variances) are given more weight in the combined analysis. This approach provides a more reliable overall estimate of the effect size because it accounts for differences in each study's reliability (Higgins & Green, 2011).
@ Tulparay Hayirli ,
Simple average of the variances obtained from different studies will not yield the appropriate value of the variable under study.
Calculate the variance of the combined observations of the studies. Formula for calculating variance of the combined observations is available in standard book on descriptive statistics.
Thank you Anitha Roshan Sagarkar for your suggestion. I have entered the effect estimate and its standard error (e.g., log hazard ratio and standard error) directly into RevMan under the "generic inverse variance" as an outcome type. However, I still can't figure out how to solve the "effect size multiplicity" in my sample. Imagine that you are examining the impact of life quality on premature mortality, and some studies measured the "life quality" in multiple ways (e.g. subjective and objective). As I know, we can't treat these multiple measures as independent effect size since they were measured on the same participants. In this case, I can' neither simply take the average of their log hazard ratio and its variance as you said, nor randomly select one measure over the other. How should I enter the data to RevMan since there is a "one effect size per study" rule in RevMan? Thank you again Anitha Roshan Sagarkar for your time.
Thank you Dhritikesh Chakrabarty. I am afraid that it still can't solve the problem of "effect size multiplicity" in my sample, where some studies use more than one measure to measure the same construct, which is based on the same participants.
- Badges
- Science topic
- Similar topics
- Biological Science
- Molecular Biology
- Biomolecules
- Nucleic Acids
- DNA