How to analyze cross-over effect in SAS?

[sy2426]

Hi,

I want to see if there is a carry-over effect in my study and I don't know how do I do so in SAS. Basically, I have three timepoints (baseline, time1, time2). All subject received diet A from baseline to time1, and then they all received diet B from time1 to time2. The study outcome is energy expenditure. There was no washout period in between. I wonder how may I analyze the data to investigate if energy expenditure from time1 to time2 is affected/carried over by diet A? Attached is my sample data. TEE_change  is the outcome. Period=ru is from baseline to time1, Period=TD is from time1 to time2. I want to see if tee_change during period=TD is affected by diet=VLC. Thank you so much in advance! 

Thanks,

[sbxkoenk]

Hello,

Paper PO16
Estimate Carryover Effect in Clinical Trial Crossover Designs
David Shen, WCI, Inc.
Zaizai Lu, AstraZeneca Pharmaceuticals
https://www.lexjansen.com/pharmasug/2006/Posters/PO16.pdf

Paper 221
CROSS CROSSOVER STUDIES OFF YOUR LIST
Pippa M. Simpson, University of Arkansas for Medical Sciences, Little Rock, Arkansas
Robert M. Hamer, UMDNJ Robert Wood Johnson Medical School, Piscataway, NJ
Shelly Lensing, University of Arkansas for Medical Sciences, Little Rock, Arkansas
https://support.sas.com/resources/papers/proceedings/proceedings/sugi24/Posters/p221-24.pdf

Good luck,

Koen

[sy2426]

Hi Koen,

  Thank you so much! The article "Estimate Carryover Effect in Clinical Trial Crossover Designs" is super helpful. However,  in my case, there is no "sequence" variable, because everyone have one diet during period A, and other diet during period B. So I have no "sequence" variable to add to the t-test.

Thanks,

Shui

[ballardw]

---

@sy2426 wrote:

Hi Koen,

  Thank you so much! The article "Estimate Carryover Effect in Clinical Trial Crossover Designs" is super helpful. However,  in my case, there is no "sequence" variable, because everyone have one diet during period A, and other diet during period B. So I have no "sequence" variable to add to the t-test.

Thanks,

Shui

---

Period A is sequence 1, Period B is sequence 2 most likely if B was done after A. Or if all of your subjects have the same diet for period A and the same different at Period B you may be looking for a Paired T-test

[Ksharp]

It is repeated measure test . I think should try Mixed model.

or calling @StatDave  @SteveDenham @lvm 

[SteveDenham · Accepted solution]

I really don't see a good way to estimate any carry-over effect.  A repeated measures analysis or a paired t test (simplest case of repeated measures) can let you know if there is a difference between treatments, but without either a "blank" period following Treatment A or a true AB/BA crossover, carryover is not estimable.

SteveDenham

[sy2426]

Thank you SteveDenham, that makes sense. I don't see how mixed effect model and paired t-test can help to estimate carry-over effect neither. 

[sy2426]

Thanks Ksharp!

[Ksharp]

If it was frequency data , PROC FREQ 's Trend Analysis is very like this, but no paired .
Or could you run Covariance Analysis if you have 'before' and 'after' data .

maybe @Rick_SAS know something .

[sy2426]

Thanks Ksharp. I contacted SAS support and got the same answer- carryover effect cannot be estimated for my data structure.

[Ksharp]

Yeah. Glad you get answer. Today I go through PROC TTEST documentation.

And find this interesting thing, but still not paired .