Is it prudent to use both logistic regression and parametric survival models on same dataset?

[Wumbe Emma]

I am doing a study which has two objectives:

1. effects of variables on timing to vaccination
2. factors associated with delays to vaccination

I have conducted a parametric survival model to assess the effect of those variables to timing of vaccination.(after the cox proportional assumption was violated)
I now want to define my delay variable and then run a logistic regression to determine the factors associated with delay

Is this a prudent way to go about my analysis?

[Maarten Buis]

I am having a bit of trouble interpreting your wording. I will assume that timing to vaccination means time till vaccination after becomming eligable, e.g. person A got her vaccination after 6 days and person B got his vaccination 16 weeks after becomming eligable. I will assume that with delays to vaccination you mean that you have some cut-off point after which a person is considered delayed. Say we set our cut-off at 4 weeks, then our hypothetical person A was not delayed and person B was delayed.

One thing I would look at is how do those two analyses fit into the main storyline of your paper. I would have two worries:

1. does this make substantive sense
2. even if it does make sense: am I trying to do too much in one paper

As to worry 1: The delay to vaccination is in essence just a discrete form of the timing of vaccination. In many cases there is no added value, and your second analysis should not be done. It may make sense if passing a certain threshold has very real and sudden consequences. Such sudden changes often happen when laws are involved. For example, in many countries/states your options are severly limited if you wait more than x weeks to discuss your options with a medical professional after the start of an unwanted pregnency (although, the number of countries/states where x=0 is increasing, but that is a topic for another forum). That would be a case to really focus on the distinction between less than x weeks versus more than x weeks.

As to worry 2, and continuing the example: We could also study the timing as the timing does influence the options available in more gradual ways. But now our paper discusses a lot of different topics at once. Do you really want to do that? (Hint: if I ask my students that question, then the answer is always no)

So to answer your question: You would need to justify that your cut-off makes sense for the story you want to tell. You also need to justify that your timing model adds something useful to your story but does not broaden your story so much that it is too much for a single paper.

[Wumbe Emma]

Thanks for the response and I appreciate the insights shared.