Hi Amha,

You can conduct your analysis with unconditional logistic regression. Even when you perform matching, depending on how it was done, you don't need to use c-logistic regression. A recent good article about the topic discussing the advantages and disadvantages is below (Pearse, N, BMJ 2016;352:i969).

I am not sure about what you are asking regarding dataset:

You should have only one dataset containing cases (diagnosis/outcome of interest) and controls (patients without the diagnosis/outcome of interest), where same variables are going to be in the same columns. The column for cases and controls will be the same, and in this column, cases will have the value 1 and controls 0 (all controls).

If you want to label, in another column, if each control as, e,g, 1, 2 or 3, it is ok. As I suppose you have 3 controls for all cases, and you don't use any method for choosing them (Nearest neighbor matching, caliper, etc), it is likely you are not doing any sensitivity analysis by choosing the "optimal" control based on some metric. Therefore, a third column, differently labelling each control, is not needed.

Hope it helps,

Otavio

http://www.bmj.com/content/352/bmj.i969

I agree ¡¡¡

Amha Admasie

Sure