Tip for data extraction for meta-analysis – 6
April 22, 2019
How do you calculate a standard error of a beta coefficient?
I have shown how to rescale hazard ratios (HRs) to a common change in a predictor variable, and also beta coefficients which are HRs (natural) log-transformed. We need to enter beta coefficients and their standard errors (SEs) into meta-analysis software. In this post I’ll show how to calculate these SEs.
Beta coefficients may be reported with a SE or a t-value. HRs are usually reported with a confidence interval.
A bit of maths (see below if you’re interested) shows us:
and rescaling from a change of x units to y units in the predictor variable
and if the HR hasn’t already been rescaled to y units
Let me show you some examples. Returning to the studies mentioned in my previous post. I had already scaled them to a 5mmHg increase in systolic blood pressure variability:
study by Mancia et al has a beta coefficient of -0.092 and its
study by Hansen et al has a beta coefficient of 0.01 and its
study by Pringle et al has a beta coefficient of 0.16 with its
So the data are ready for meta-analysis, but all these data cannot be pooled together. The data of Mancia et al may be pooled with the data of Hansen et al, or the data of Pringle at al, but not the data from both studies, because the patient populations of these two studies overlap – double counting patients can inflate the effect estimates. I’ll return to the issue of overlapping patient populations in a later post.
Here’s a tip…
Standard errors of beta coefficients can be calculated from t values and confidence intervals
In my next post I will show how to pool data from studies reporting categorical risk data with different numbers of categories and different thresholds.
Where did the equations come from?
(You can skip this if you are only interested in carrying out the calculations)
This can be rearranged to
A 95% confidence interval is
upperCI = mean + 1.96 x SE
lowerCI = mean – 1.96 x SE
This equation applies to any symmetric 95% confidence interval.
As HRx has a confidence interval (lowerCIx to upperCIx)
As βx has a confidence interval (ln(lowerCIx) to ln(upperCIx))
Rescaling beta coefficients from a change of x units to y units in the predictor variable,
both follow on from the explanation I gave in my last post for
βy = y x beta
The same applies for
which follows on from the explanation given for
Dr Kathy Taylor teaches data extraction in Meta-analysis. This is a short course that is also available as part of our MSc in Evidence-Based Health Care, MSc in EBHC Medical Statistics, and MSc in EBHC Systematic Reviews.
Follow updates on this blog and related news on Twitter @dataextips