I am working with NIS data which are basically a 20% sample of all in-hospital admissions in the US every year. I have small group of cases that carry diagnosis "X" in their record and overall the only consist of 0.1% of the total n.

I would like to compare that group with cases "without diagnosis X". As you realize, with a total n= 6 million, the "diagnosis X" group is way too small for comparing associated characteristics against the "without diagnosis X" group (chi-square and t-tests, even regression)

Is there a way to control for this "overpowering" of the control group?

Thank you.

I use SPSS 23

  • $\begingroup$ Are all 6 million non-cases relevant to the study problem? To put it another way, is X an "equal opportunity" disease, or are there certain risk factors for X that you can use to exclude non relevant observations? What kind of biases would you introduce in the sample if you do. For example, if X only affects older adults (50 years+), you can probably exclude small children. $\endgroup$ – Marquis de Carabas Apr 13 '16 at 1:25
  • $\begingroup$ that's a good thought but X theoretically can affect all ages. I already excluded <18year old to limit the study to adults. Is there another way that doesn't have to do with excluding by specific variables? $\endgroup$ – Paris Char Apr 13 '16 at 1:36
  • $\begingroup$ are there risk factors for X that you want to test between the cases and controls? My concern is that your study would be overpowered and you will find statistically significant differences that may not be clinically important. Am I right in reading that you have 60,000 cases of X? Is your outcome binary? $\endgroup$ – Marquis de Carabas Apr 13 '16 at 2:09
  • $\begingroup$ What exactly is it that you want to compare between the group with X and the broader population? $\endgroup$ – EdM Apr 13 '16 at 6:54
  • $\begingroup$ Binary outcome (in hospital death) with categorical and continuous covariates. Chi square and t-tests at first, then multivariable regression if possible $\endgroup$ – Paris Char Apr 14 '16 at 12:37

Your Answer

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

Browse other questions tagged or ask your own question.