I am in the process of performing analysis for effectiveness of mono, dual and triple drug treatments. I am using time-varying Cox models, where my exposure is time-varying (1, 2 or 3 drugs). I am, however, struggling to understand the theoretical foundation of such an approach. In my analysis, patients can use 1 drug for, say, 100 days, then add another and use 2 drugs for another 100 days and finally go back to using 1 drug for the last 100 days. How does my model account for the fact that first 100 days may be drastically different from the last 100 days. Does the time variable that I create and use in the model (which has values of 100, 200 and 300 respectively for 1 drug, 2 drugs and 1 drug) account for this? Thank you.

Not my area of expertise, exactly, but since nobody is responding, I’ll give it a shot -

Putting these variables naively in a model does not seem like it will result in valuable information. You need to think carefully about what is going on in order to model this situation correctly. For example, what causes people to switch from one drug to two? Is their condition worsening? Do you have a measure for that? What is the expected amount of time between treatment initiation and a potential effect? It seems to me that the questions that matter here aren’t statistical, exactly. More that you need a solid understanding of the underlying processes to build a model.

Part three of Hernan & Robins’ excellent free textbook describes many of the considerations required for this kind of modelling.