2
$\begingroup$

I am aware that sample size calculations for cluster RCTs must adjust for clustered data. I would like to understand better if such adjustment might be necessary for other study designs. For instance, I am working on a nested case-control study to identify clinical and socio-demographic factors associated with patient retention. We will use medical record data collected from health facilities. All patients from all sites meeting the inclusion criteria will be included in the study (no sampling per se). The study is not trying to make any inference about any population parameter.

Should the sample size calculation adjust for clustering? After all, it is likely that the data abstracted from medical records will be clustered. At the same time, should the analysis adjust for clustering?

How about if the study were a cluster RCT and all patients from all sites meeting the inclusion criteria will be included in the study--would the sample size calculation need to adjust for clustering?

$\endgroup$

1 Answer 1

0
$\begingroup$

I can't speak to sample size calculations for case-control studies. I can answer the second question.

How about if the study were a cluster RCT and all patients from all sites meeting the inclusion criteria will be included in the study--would the sample size calculation need to adjust for clustering?

Yes, if treatment was assigned at the site level such that all patients at a site were either assigned to the control arm or the treatment arm you would need to adjust for clustering. The mean for the treatment arm and the mean for the control arm would constitute the mean from a one-stage cluster sample.

For a derivation showing why clustered standard errors is required in this case see my answer here.

$\endgroup$

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.