Proc logistic help with a covariate

aaflower — Fri, 16 Aug 2019 18:47:07 GMT

Hello. I am looking for help with a certain covariate for a logistic model. This is a vaccine study where the endpoint is an event (0 or 1). There are 3 doses administered. We would like to somehow incorporate the effect of when the event occurs: between dose 1 and 2 (dose=1), between dose 2 and 3(dose=2), or after dose 3(dose=3). Due to the design of the study we do not feel that using dose=1,2,3 as a continuous variable is appropriate.

Could it be treated as an ordinal variable? Could be decomposed into indicator variables like Dose=dose1, dose2, dose3. What is the best method to use in the model such that SAS applies the effect appropriately?

Currently we have model event = treatment dose1 dose2 dose3 (and the interactions)

where dose1, dose2, dose3 are categorical y/n=1/0, but the model fails due to low counts for dose1. We tried combining dose1 and dose2 and still get this warning: WARNING: There is possibly a quasi-complete separation of data points. The maximum likelihood estimate may not exist.

WARNING: The LOGISTIC procedure continues in spite of the above warning. Results shown are based on the last maximum likelihood iteration. Validity of the model fit is questionable.

However, when I use the ordinal option, it runs fine. But want to make sure it's valid.

Thank you for your time.

Re: Proc logistic help with a covariate

PaigeMiller — Fri, 16 Aug 2019 19:29:28 GMT

First, when you talk about logistic regression, ordinal always refers to the Y variable, in this case EVENT. So, when you talk about dose being an ordinal variable (and also DOSE is an X variable), this doesn't make sense to me, X variables cannot be ordinal in the sense used in Logistic regression.

Also, I usually think of dose as an amount (on a continuous scale) of something administered, rather than a 0/1 variable. So I'm not really clear what these doses are.

Based on my (possibly limited) understanding of what you are trying to do, this seems more like a survival model than a logistic regression, where the subject can "survive" — not have the event — until time (dose) 1 or time (dose) 2 or time (dose) 3. In this case, you would want to run PROC PHREG, but as I have no experience with survival models, I'll have to drop out here.

topic Re: Proc logistic help with a covariate in Statistical Procedures

Proc logistic help with a covariate

Re: Proc logistic help with a covariate