I have some observational data where treatment lines and times are being tracked and am interested to see how length of specific treatments are associated with a biomarker. The dataset contains 2 records per patient. 1 record is a prior line of therapy and the 2nd record is the line of therapy of interest.
When modeling length of treatment = biomarker + treatment (prior or current) + biomarker*treatment. Does it make sense to also include other baseline covariates such as age and gender? I have been talking myself in a little circle here since all baseline covariates are controlled between treatments (and biomarker status) since the exact same patients are being used across treatments.