I am doing an analysis to identify independent predictors of a positive drug test result for patients who self-report being on medication in a cohort study (i.e., I am assessing recent medication adherence). The drug test was done at two follow up times, but some patients missed one of them. There are 53 patients with 2 follow ups, and 30 patients with only one. I want to use all 83 people with any follow up data, and it would be nice to use both follow up timepoints for patients who have both, if it is methodologically possible and "worth" whatever added complexity. The purpose would be to strengthen statistical power as well as not throw out data unnecessarily. Timepoint is not part of the research question as a predictor.
Is there a statistical model that would allow me to do this? To resummarize, I want to estimate the independent associations between a set of predictors and a binary outcome taken at either one or two timepoints, using all available data points.
I saw a similar analysis that stated that they used logistic regression with robust variance for repeated measures, but that study had more follow up periods. If you think trying to use the repeated measures dimension of the data in this analysis is more trouble than it's worth (e.g., it would make results hard to explain to a non-research population, or it would be likely to produce other statistical power or convergence problems), I appreciate hearing those thoughts as well. I am using SAS and also have SPSS as a second choice, just fyi. Thank you for considering my question