Statistical analysis for a randomized controlled trial where dose is possibly changed during every measurement? Suppose you're conducting an RCT to assess the efficacy of a medicine. You recruit, say, two groups and you plan to measure a primary outcome at baseline and at two and four weeks follow-up. You also have a secondary outcome which measures the presence of adverse events. Based on whether adverse events were reported in the preceding time interval, the administered dose (exposure) might be changed at each follow-up measurement.
I want to know more about how to appropriately analyze the results of such a trial. It is my understanding that this is a so-called adaptive trial and I am guessing that a linear mixed model should be used, but this is all I know. What are best practices in a case like this? What is the go-to reference for learning more about statistical analyses in contexts like these?
 A: One recent "go-to" reference on this topic is the Causal Inference book by Hernán and Robins. They call this a "dynamic treatment strategy." Part III of the book is specifically about complex longitudinal studies like this, with Chapter 19 introducing time-varying treatments. The authors say near the start of that chapter:

This chapter describes some key terminology and concepts for causal inference with time-varying treatments. Though we have done our best to simplify those concepts (if you don’t believe us, check out the causal inference literature), this is still one of the most technical chapters in the book. Unfortunately, further simplification would result in too much loss of rigor. But if you made it this far, you are qualified to understand this chapter.

So you will need to absorb the principles of causal inference in fixed-treatment studies (Parts I and II of the book) before you can design a proper analysis of this type of study. It's not a simple matter of the type of statistical model to use. You have to specify the causal diagram representing the situation at hand and choose the analysis strategy accordingly. Hard work, but necessary if you want to draw casual inferences about the "efficacy of a medicine" in this context.
