0
$\begingroup$

I'm writing a proposal for a restrospective cohort study using registry data. Am I right to think of the registry data as the whole population?

If it is the population, then am I right in thinking there is no need to calculate required sample size (as is required for other cohort studies) or confidence intervals?

If it is a sample, what would be the study population, and how would I go about calculating sample size?

Additionally, with this in mind, how do I treat missing data? If there are 1000 heart transplant patients in a registry but only 500 of them have complete data (i.e. all the variables have values), do I treat these 500 patients as a sample? Furthermore, if it is a sample, it isn't random. Therefore is it worth calculating the confidence intervals from this actual sample size?

Improve this question
$\endgroup$
1
$\begingroup$

Whether 'registry' is a population or a sample depends partly on your goals, but it is probably a convenience sample from a larger sample of (in your case) everyone who has had a heart attack (or whatever qualifies people for your registry).

Missing data is a different problem; if half of the people have missing data, then you want to deal with that. One way is multiple imputation. But if the data is not missing at random, you will have a problem (esp because there is so much missing data). You might want to remove some variables from your analysis, if there are some variables that are not essential and that have lots of missing people.

Is it still worth calculating CI from non-random samples? Technically, not really, but in a practical sense - people do it all the time and it can be useful. Only a very small portion of all interesting variables can be studied in random samples.

Improve this answer
$\endgroup$
3
  • $\begingroup$ I think I will treat the registry data as a sample. However, what would be an acceptable thing to write in the proposal regarding the non-randomness of the sample? $\endgroup$ – oisyutat Nov 25 '12 at 13:33
  • 1
    $\begingroup$ You just write that this is a convenience sample. If you like, you can add something like "but we have no reason to believe that it is substantially different from XXXX" but you should only write this if you can defend it. (XXXX would be some larger population). If you have any information on XXXX that you can compare to the registry, you can note that and then do the comparison. This is evidence that the sample is not biased, but it is not conclusive evidence (they could differ on some variable that you don't have). $\endgroup$ – Peter Flom Nov 25 '12 at 13:52
  • $\begingroup$ In any case, you should give descriptive information on the registry so that people trying to apply your results to other cases have some basis for comparison. $\endgroup$ – Peter Flom Nov 25 '12 at 13:52

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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