Talk:Crossover study

I revised and greatly expanded this article. I motivate many of the changes below: —Preceding unsigned comment added by Kiefer.Wolfowitz (talk • contribs) 00:38, 23 May 2009 (UTC)

Introduction:
NEW: It would be useful to provide an example of a cross-over designs, e.g. from Jones & Kenward (page 142): AABB BBAA ABBA BAAB Kiefer.Wolfowitz (talk) 00:53, 23 May 2009 (UTC)


 * I agree - could someone add a simple example preferably comprehensible by non-statisticians? Ben Finn (talk) 22:47, 14 March 2019 (UTC)

OLD:

The introduction should mention longitudinal studies, of which cross-over designs are an important subclass (within trials).

The wording is unclear, with its use of logical quantifiers---sorry for the vagueness! Let me explain: Most cross-over designs have some balance properties, so that every patient receives (ITT) the same number of treatments, for example; I believe that most such designs are balanced in that each patient is to receive all of the treatments (and each treatment the same number of time). The writing is unclear.

There are a number of "minimization" claims that should be reworded or justified. (Maybe there is a trade-off between two (mentioned) minimand objective-functions, for example?)
 * Variability between patients.
 * Carry-over

Each patient serves as his own control (or as his or her own control)---not "their" own control. I don't know Wikipedia's policy about androgynous pronouns, so I didn't correct the error.

The draw-back of not being able to study "long term" effects occurs with any controlled human-subject trial (versus epidemiological study); it's not a disadvantage of cross-over trials per se.

Kiefer.Wolfowitz (talk) 21:28, 22 May 2009 (UTC)

Limitations and Disadvantages
It would be useful to warn people about the disadvantages of 2-period 2-treatment designs: risk of carry-over biasing estimates, inefficiency, etc. Jones and Kenward recommend 3 or 4 period designs when feasible. (2-period 2-treatment designs are easy to teach and easy to test, but they are not recommended generally!)

Kiefer.Wolfowitz (talk) 00:56, 23 May 2009 (UTC) It seems that the "order issue" and the "carryover issue" are the same: the first treatment may affect the second treatment, and therefore it's needed a wash-out period to cancel the effect of the first treatment. Shouldn't they combine into one disadvantage?

Tom