6

I'm planning an analysis that will require pooled logistic regression to analyze a longitudinal dataset with time-dependent explanatory variables (see here - D'Agostino, et al. 1990). However, the standard analysis uses regularly spaced intervals. My study has followup points at 0, 2, 6, 12, 24 months. What is a good way to model the effects of time given this structure of unevenly spaced intervals? Would a restricted cubic splines for the effects of time address the irregularity issue?

Thanks!

stats134711
  • 721
  • 1
  • 7
  • 17

1 Answers1

4

You can model the time trend in the log odds of the probability that $Y=1$ using a spline function (and using other methods). I do not believe that the time points need to be equally spaced, but rather that they be discrete and have roughly the same schedule for every subject.

Frank Harrell
  • 91,879
  • 6
  • 178
  • 397