2
$\begingroup$

I am currently performing a meta-analysis where I am pooling area under the curve values.

For some papers I only have access to the figure. For example:

enter image description here

Using WebPlotDigitizer I can extract the mean and standard error/standard deviation at each time point. I can then calculate the mean area under the curve. However, I am unsure as to the correct approach in order to calculate a corresponding variance term.

Is it possible to use the values depicted by the error bars at each time point to calculate an estimation of variance for the area under the curve?

Improve this question
$\endgroup$
6
  • $\begingroup$ Interesting question... if you know patients at each time point you can surely generate variances... Otherwise, you need to factor in some assumptions (eg 100% follow-up completion... and then run some sensitivity analysis, such as 10% loss to follow-up at each time point) $\endgroup$
    – Giuseppe Biondi-Zoccai
    Commented Jun 19, 2021 at 8:33
  • 1
    $\begingroup$ All studies that I am pooling are crossover, so in almost all cases I know the number of participants at each time point (it is typically the same within each study). How would you suggest to calculate variance? The only idea I have at the moment is to calculate the AUC using either the upper or lower bound of the error bar. $\endgroup$
    – Jamie Frampton
    Commented Jun 21, 2021 at 7:34
  • $\begingroup$ Look at the legend... if it states it is standard error, then you can back calculate the variance... if it states it is a standard deviation, it's even easier... You may exploit the normal distribution assumptions, with SE=SD / sqrt ( N ) $\endgroup$
    – Giuseppe Biondi-Zoccai
    Commented Jun 21, 2021 at 7:54
  • 1
    $\begingroup$ I think you may be misunderstanding my question. In the above example, I will use the mean at each of the 9 time points to calculate the mean AUC for each condition (in the above example, this would be one AUC for the fasted condition and one AUC for the fed). However, I also need a corresponding SD or SEM for each AUC that I have calculated. Is it possible to use the error bars to calculate this corresponding SD/SEM. For example, calculate AUC using the top or bottom error bar at each of the 9 time points, and then use this AUC to calculate the SD/SEM for the mean AUC previously calculated. $\endgroup$
    – Jamie Frampton
    Commented Jun 21, 2021 at 10:12
  • 1
    $\begingroup$ Ok - thank you for your help with this problem $\endgroup$
    – Jamie Frampton
    Commented Jun 22, 2021 at 13:24

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.