There was linked-in discussion about performing Tmax (a discrete variable) statistical comparison for a cross-over study data in Phoenix WinNonlin. There was response from a user: he indicated that go to NCA> Toolbox>Xover.
By following this routine you can perform a nonparametric analysis of Tmax for the parameters for the cross-over study. I have found an example in the EXAMPLES. I repeated the example (for 2 way comparison) and it worked fine. It provides the confidence intervals for the comparison of two formulations. I am wondering if it provides the intrasubject and intersubject variance information analogous to what one gets for bioequivalence of Cmax or AUC.
I think this Tmax comparison is limited to a 2 way cross-over study and will not work for a higher order comparison…yes?
Hi Angus,
I answered you in the BEBA-Forum.
Correct. There is a method for higher-order cross-overs, but it is not implemented in PHX.
Duchatau et. al
Adjusting pairwise nonparametric equivalence hypothesis tests and confidence intervals for period effects in 3x3 crossover trials
J Biopharm Stat 12(2), 149–60 (2002)
I think Helmut has already given yo uthe info you need but Page 100 of Phoenix WinNonlin 6.4 User’s Guide.pdf gives you more back ground of what is being done in Phoenix,
Simon
Thanks Helmut and Thanks Simon: I will look at the manual carefully.
Hi Angus,
Crossover Design follows this reference for the confidence intervals:
Conover (1980). Practical Nonparametric Statistics 2nd ed. John Wiley & Sons, New York. (page 111 in the 2nd ed. in the section called “Confidence Interval for a Quantile”)
and this reference for the tests of “Effects”:
Koch (1972). The use of nonparametric methods in the statistical analysis of the two-period
change-over design. Biometrics 577-83.
Crossover Design in PHX WNL is not the same as the Hodges-Lehmann Estimate. In the documentation in the WNL Users Guide, note that the formulas on page 101 for N<=20 are not actually used since these values can just be obtained from a Mann-Whitney table.
Regards,
Linda Hughes
Linda: Thank you for your response. I will look at this material in NLM. When I raise this topic immediately the thought is the confidence intervals for a comparison of Tmax values for two different formulations say A, B in a cross-over study. I understand well that the variance parameter does not apply for nonparametric data.
If we can put the confidence intervals to one side for the moment.
Let me construct a very simple situation: say have done a Latin Square Design for a cross-over study in 18 subjects. So we have 18 values for Tmax for one treatment (A) and 18 values for Tmax for a second treatment ( B). {We are assuming no carry-over effects or sequence or period effects}.
I want to make a very simple descriptive assessment of whether the set of Tmax values observed for the one treatment group appears to be more variable in the treatment group compared to the other.
I am thinking in terms of making a frequency plot of the 18 values for TRTA. I am also thinking of making the same frequency plot for 18 values for Treatment B. The two plots will me to see the range and the frequency of values for the two treatments. The median value for B and A will be marked on each chart.
I will then construct a Table corresponding to the frequency plots.
Additionally I will make a percentile Table for each of the two sets.
By looking at this data I will prove a descriptive commentary of the Tmax variability for the two treatments.
Linda; do you have any thoughts on above suggestion,
Angus
Hi Angus,
No, I don’t really have any comments on the approach. It seems that whoever you are submitting this to would have the final say.
Regards,
Linda