I'm currently doing a study on already existing medical data in oncology. The aim of the study is to compare the survival between two groups of patients : the groupe A continued its treatment until major toxicity or progression of the disease, and the groupe B had an elective discontinuation after a variable duration.

The issues is that :

  1. The majority of patients in group B (60%), stopped their treatment after 24 months, which introduce a major immortal time bias.
  2. Patients in group B discontinued their treatment at a variable duration, between 12 and 36 months.

I was wondering if, to asses those biases, I could analyse the treatment status (treated vs off treatment) as a time-dependant covariates, in order to compare overall survival between the two groups. Each patient would contributes person-time to the "continuation" state until their discontinuation date, then switches to the "discontinuation" state.

I would also do a supplementary survival analysis using a landmark at 24 months, but the time-dependant covariates model seems interesting.

I hope my question is clear, and I thank you very much for your help.