Significant period effect [Design Issues]
❝ Question. What can cause a period effect? Aside from carryover due to poor washout selection, and other unexplanable events (which I can't think of)?
Lots of speculation out there and regrettably very little actual substance.
Carry-over in terms of pharmacological effect (e.g. some effect on the liver or intestine starting after dosing in period 1, still present in period 2). Anxiety, "this drug made me sick in period 1, hell, I am going to get sick in period 2, too" and John Doe will have diarhoea way before he event reports to the clinic but won't say anything out of fear of reduced compensation. And so forth. Don't waste your time medlining it.
❝ If someone decides to analyze study samples by period instead of analyzing the samples after completion of all study periods, could that cause a significant period effect?
Well, in theory, yes? But then again. This is highly speculative and there would be a need to find the root cause (like stock instability of the IS or something like that).
❝ Is it possible to determine the interaction between period and sequence effect?
Yes. It is a treatment effect.
E.g. Period 1 in (or x) Sequence TR is effectively just treatment T (in that period). Note that some serious DF elimination may come into play depending on how you specify the model. For example, in the model matrix the columns for "period 1 in Seq TR" and "period 2 in Seq RT" will add up to exactly the column for the Treatment T effect. If you have the interaction before the treatment in your model spec, then the latter will get a type I SS and MS of zero.
Pass or fail!
- Significant period effect jag009 2016-12-20 04:48 [Design Issues]