1 reply to this topic

[LLLi]

Dear All,

 

I am reading a protocol and the study design is a crossover study. The protocol states that "AUCs and C_max will be log transformed and analysed using a mixed model analysis of variance with fixed effect terms for treatment arm, treatment period, and treatment sequence, and subject within sequence as a random effect. _T½ will be reciprocated prior to analysis using the same mixed model approach as for AUCs and C_max. Least square means of the contrasts will be back transformed to the original scale and presented. 

 

For the AUC and Cmax, the statistical analysis is just what we normally do. But for the T1/2, i am totally lost. As my understanding, the reciprocal of T1/2 is similar to kel since T1/2 is ln2/kel. So we will observe the kel difference?

 

If we have to do in the way specified in the protocol, do we need to log transform the reciprocal of T1/2 first?

 

If no, we will get the lsmean difference and the ratio of lsmean. How to back transform the results to the original scale? For the ratio, we just do another reciprocation? How about the lsmean difference?

 

If log transformation is needed before the mixed model, we will get the lsmean difference and the ratio of geoslmean. Still the same question. How to back transform?

 

Thanks!

 

Best,

LLLi

[Angus McLean]

Dear All,

 

I am reading a protocol and the study design is a crossover study. The protocol states that "AUCs and C_max will be log transformed and analysed using a mixed model analysis of variance with fixed effect terms for treatment arm, treatment period, and treatment sequence, and subject within sequence as a random effect. _T½ will be reciprocated prior to analysis using the same mixed model approach as for AUCs and C_max. Least square means of the contrasts will be back transformed to the original scale and presented. 

 

For the AUC and Cmax, the statistical analysis is just what we normally do. But for the T1/2, i am totally lost. As my understanding, the reciprocal of T1/2 is similar to kel since T1/2 is ln2/kel. So we will observe the kel difference?

 

If we have to do in the way specified in the protocol, do we need to log transform the reciprocal of T1/2 first?

 

If no, we will get the lsmean difference and the ratio of lsmean. How to back transform the results to the original scale? For the ratio, we just do another reciprocation? How about the lsmean difference?

 

If log transformation is needed before the mixed model, we will get the lsmean difference and the ratio of geoslmean. Still the same question. How to back transform?

 

Thanks!

 

Best,

LLLi

This statistical approach you are referring to is for statistical analysis of BE for Cmax and AUC which are Ln transformed to make the data conform to the norm distribution.  Do not understand why the clinical protocol  has asked for the same approach to be applied to the half life, which is a time.   Suggest a noparametric appraoch should be preferred eg. Wilcoxan Matched Pairs signed Rank Test just like has been used for Tmax .

 

Hope this helps.

 

Angus