Logistic regression: correction for multiple comparisons

[Razkolnikova]

Hi all,

I have a question regarding correction for multiple comparisons in logistic regression using SPSS.

I have a huge data set of n=762. I am testing whether regional infarct volume in ischemic stroke (a continuous variable) predicts cognitive impairment (yes or no --> binary outcome). I have 90 regional volumes in total and I need to run 90 separate analyses. In each one, I will be setting the following parameters:

- Outcome variable is cognitive impairment;
- Predictors are sex (correction factor), total lesion volume (correction factor), and regional infarct volume (entering one at a time) = so, 3 predictors for each of the 90 models that I will be running.

Do I need to and how can I apply correction for multiple comparisons using SPSS in this case? I don't see how the program can do that so far. Would setting the confidence intervals to 99% be of any help?

Thanks in advance, all help is very much appreciated!!

 

[obh]

So you run 90 separated regressions? and you ask about the overall model significance? (F)
Why?

 

[Karabiner]

I have a huge data set of n=762.

This is insufficient information in logistic regression. It is important how frequent the 2 outcomes are.
For example, if you have n=7 without cognitive impairment and n=755 with cognitive impairment,
then the total n isn't so impressive (since via logistic regression you would try to predict n=7 cases
within the smaller group).

Do I need to and how can I apply correction for multiple comparisons using SPSS in this case?

I have rarely seen a studies where such a huge number of analyes on a limited dataset appeared
to be useful. Is it really necessary to use 90 regions? Is it possible to cluster/aggregate regions somehow?

Maybe you could give a short description of the study background and the study objectives?

With kind regards

Karabiner