Hello,
I am meta-analysing studies looking at different levels of physical activity. I want to compare moderate against low levels of physical activity and for each study need the log hazard ratio and standard error for this comparison.
Unfortunately different studies use different levels of physical activity as the reference value (either high or low) see below.
Study 1
Low physical activity Hazard ratio 1.00 reference
Moderate: 0.68, 95% CI 0.55 to 0.83
High physical activity: 0.71, 95% CI 0.58 to 0.87
Study 2
High physical activity: Hazard ratio 1.00 reference
Moderate 1.22, 1.07 to 1.40
Low: 1.54, 1.34 to 1.77
To include the studies in the meta-analysis I need to calculate the standard error of the logHR for the comparison of moderate against low physical activity for study 2. But I only have confidence intervals for the comparison of moderate against high, and low against high. What I don't have is the covariance of the low and moderate physical activity categories. In some cases I know the number of life years that contributed to each study group.
Does anyone have a suggestion on how I can approximate the standard error with the data I have?
many thanks,
James
|