0521 ☆ China, 20181109 07:02 Posting: # 19550 Views: 1,481 

Dear Helmut Schütz, When using Phoenix WinNonlin Crossover object for Tmax testing, how can I judge whether it is equivalent by the result? I read the "The Use Of NonParametric Methods In The Statistical Analysis" document (PMID: 4556704), and I think he is doing the following steps to judge: 1. Determine if the P value of the sequence is greater than 0.05, 2. If the P value of the sequence is less than 0.05, the test of "Treatment" cannot be performed. 3. If the P value of the sequence is greater than 0.05, it is judged whether the P value of the treatment is greater than 0.05. 4. If the P value of the Treatment is greater than the equivalent, otherwise it is not equivalent. 5. If the P value of the sequence and the P value of the Period are both greater than 0.05, a sign test can be used instead of the rank sum test used in the above steps. Assuming α = 0.05, all of the above P values are the results in the "Effects" table. Is the above judgment process correct? If not, what is the correct judgment process? PARAMETER Test T_stat L_stat p_Value Best, 0521 
mittyri ★★ Russia, 20181111 00:21 @ 0521 Posting: # 19561 Views: 1,277 

Dear 0521, » Dear Helmut Schütz, Not interested in other opinions? » I read the "The Use Of NonParametric Methods In The Statistical Analysis" document（PMID: 4556704） Good start! » 1. Determine if the P value of the sequence is greater than 0.05, It depends on the significance level you've chosen (p). WNL cannot do that for you. no sequence effect (or drug residual effect) in WNL terminology.» 2. If the P value of the sequence is less than 0.05, the test of "Treatment" cannot be performed. No treatment effect given no sequence effect in WNL terminology.I would follow the text of the article you referred: Finally, it is appropriate to note here that if the residual effects cannot be deleted from the model, then the test procedures described in section 3.2 and 3.3 are no longer valid Corresponding sequence test (unequal carryover) in ANOVA couldn't be a reason for analysis interruption. Likely unequal carryover doesn’t exist in a properly designed study (sufficiently long washout), hence false positives could happen. » 3. If the P value of the sequence is greater than 0.05, it is judged whether the P value of the treatment is greater than 0.05. » 4. If the P value of the Treatment is greater than the equivalent, otherwise it is not equivalent. Significant != not Equivalent and vice versa! That's why we are using confidence intervals, not pvalues in our BE decisions. » 5. If the P value of the sequence and the P value of the Period are both greater than 0.05, a sign test can be used instead of the rank sum test used in the above steps. You omit 3.3 Testing the equality of period effects when residual effects are absent ([no period effect given no sequence effect] and [no treatment and no sequence effect] in WNL terminology)Yes, it could, but WNL does the trick with 3.4: Since the bivariate Wilcoxon test requires only an ordinal scale for ranking across subjects, it should be used instead of the sign test for general situations in which residual and/or period effects are unequal and within subject linear functions are invalid. Using the pvalues from the Effects sheet is the same as using pvalues from ANOVA: good exploratory analysis without any credible conclusions regarding bioequivalence. It is better to compare CIs to some reference limits as Helmut noted here. — Kind regards, Mittyri 
0521 ☆ China, 20181111 10:56 (edited by mittyri on 20181111 23:05) @ mittyri Posting: # 19562 Views: 1,283 

» Not interested in other opinions? Thank you very much for your reply, I welcome anyone to reply to this post. » » I read the "The Use Of NonParametric Methods In The Statistical Analysis" document（PMID: 4556704） » Good start! This paper is a reference to the Crossover object in the WinNonlin User Guide. So I think I should follow the paper to interpret the results of the WinNonlin Crossover object. » » 1. Determine if the P value of the sequence is greater than 0.05, » It depends on the significance level you've chosen (p). WNL cannot do that for you. I assume that the significance level is 0.05. » no sequence effect (or drug residual effect) in WNL terminology.Does the first row of the effects sheet not indicate a test for sequence effects? » » 2. If the P value of the sequence is less than 0.05, the test of "Treatment" cannot be performed. » No treatment effect given no sequence effect in WNL terminology.Does the second row of the effects sheet not indicate a test for treatment effects? » I would follow the text of the article you referred: » Finally, it is appropriate to note here that if the residual effects cannot be deleted from the model, then the test procedures described in section 3.2 and 3.3 are no longer valid » Corresponding sequence test (unequal carryover) in ANOVA couldn't be a reason for analysis interruption. Likely unequal carryover doesn’t exist in a properly designed study (sufficiently long washout), hence false positives could happen. Although I also think that the residual effect can not be used as a reason for the interruption test, However, for this paper, 3.2 and 3.3 depend on 3.1, and only after the 3.1 test is passed, the tests of 3.2 and 3.3 can be performed. » » 4. If the P value of the Treatment is greater than the equivalent, otherwise it is not equivalent. » Significant != not Equivalent and vice versa! That's why we are using confidence intervals, not pvalues in our BE decisions. I agree. Do you want to tell me this: 1. In the effect sheet, the P value judgment of the Treatment is independent of the sequence and the period. 2. All P values in the effect sheet are only statistically significant. The P value has no substantial clinical significance and should not be overly concerned. We should pay more attention to the results in the Confidence Intervals sheet. 3. For the effect sheet, we should pay attention to whether the median of Treatment_Diff_( T  R ) has clinical significance. Other questions: 1. For the nonparametric test of Tmax, is the only way to calculate the P value using "Mann–Whitney U test"? 2. For the nonparametric test of Tmax, can we use the "Wilcoxon signed rank test" to calculate the P value? 3. For the effect sheet, is the "Conclusion_Diff_(T  R)" confidence interval "0" necessary? Edit: Full quote removed. Please delete everything from the text of the original poster which is not necessary in understanding your answer; see also this post #5! [Mittyri] 
mittyri ★★ Russia, 20181111 23:21 @ 0521 Posting: # 19563 Views: 1,248 

Hi, » I assume that the significance level is 0.05. Note that this is your assumption, not WNL » Does the first row of the effects sheet not indicate a test for sequence effects? It indicates. » Does the second row of the effects sheet not indicate a test for treatment effects? It indicates. » Although I also think that the residual effect can not be used as a reason for the interruption test, » However, for this paper, 3.2 and 3.3 depend on 3.1, and only after the 3.1 test is passed, the tests of 3.2 and 3.3 can be performed. WNL is software, not allmighty wisdom. Only YOU can decide whenever it is possible to delete residual effects from the model and take into account that rows. » 1. In the effect sheet, the P value judgment of the Treatment is independent of the sequence and the period. The treatment is dependent on Sequence only, not period, as authors of paper mentioned (see above). It is, IF you cannot neglect unequal carryover, you cannot test Treatment (3.2), but can try bivariate Wilcoxon statistic (3.4). » 2. All P values in the effect sheet are only statistically significant. The P value has no substantial clinical significance and should not be overly concerned. We should pay more attention to the results in the Confidence Intervals sheet. It depends on the Hypothesis you are stating in the Protocol. » 3. For the effect sheet, we should pay attention to whether the median of Treatment_Diff_( T  R ) has clinical significance. and related to the hypothesis you are testing. » 1. For the nonparametric test of Tmax, is the only way to calculate the P value using "Mann–Whitney U test"? Obiously not. You can use any test aplicable for ordinal data with scientific justification. » 2. For the nonparametric test of Tmax, can we use the "Wilcoxon signed rank test" to calculate the P value? Sure, you can. Just keep in mind that you are neglecting period/sequence effects. It tests the withinsubject treatment differences against zero and is therefore a nonparametric analogon of the paired Student ttest (see the link to the discussion above in the previous post). » 3. For the effect sheet, is the "Conclusion_Diff_(T  R)" confidence interval "0" necessary? Cannot get the question — Kind regards, Mittyri 
0521 ☆ China, 20181112 07:15 @ mittyri Posting: # 19564 Views: 1,199 

Dear Mittyri, Your reply is very helpful. I hope that I can fully understand what you mean, so I hope that I can repeat your point of view with my own words: 1. The meaning of the value of P should be stated in the hypothesis of my Protocol and set a significant level α, which I assume here α = 0.05. 2. For the Tmax test, I can use any test method for ordinal data as long as there is scientific evidence. 3. If I use the method of the paper to test Tmax, only the "3.2 treatment test" is meaningful when the P value of the "3.1 sequence test" sequence effect is >0.05. such as, Scene 1: P value of sequence effect = 0.6, P value of treatment effect = 0.6 Scene 2: P value of sequence effect = 0.6, P value of treatment effect = 0.01 Scene 3: P value of sequence effect = 0.01, P value of treatment effect = 0.6 Scene 4: P value of sequence effect = 0.01, P value of treatment effect = 0.01 Among the above four scenarios, only the treatment effect of scene 1 is not significantly different. Only the therapeutic effects of Scene 2 are significantly different, Scene 3 and Scene 4 cannot determine whether the treatment effect is significantly different. 4. If I use the paper method to test Tmax, and if the P value of the "3.1 Sequence Test" sequence effect is <0.05, then I should use the "3.4 Treatment & Residual (simultaneous) test" instead of "3.2 treatment test". 5. I must assume that there is no significant difference between the sequence effect and the pairing effect before the "Wilcoxon signed rank test" method can be used. 6. Does it mean that the method in the paper as a whole is more extensive than the data applied by the "Wilcoxon signed rank test". In other words, the method in the paper has fewer assumptions than the "Wilcoxon signed rank test". Kind regards, 0521 
mittyri ★★ Russia, 20181116 22:52 @ 0521 Posting: # 19641 Views: 1,139 

Dear 0521, » 3. If I use the method of the paper to test Tmax, only the "3.2 treatment test" is meaningful when the P value of the "3.1 sequence test" sequence effect is >0.05 Citing Hauschke et al.^{1)}: When assessing treatment effects, the fundamental assumption is the absence of carryover effects, which would again imply a reformulation of the test problem to ensure applicability of a direct test approach. For the sake of simplicity, it has been proposed to use a more liberal significance level for the above test, for example p = 0.10; see also Jones and Kenward (2003) for a further discussion of this indirect approach. A direct approach for testing the absence of a relevant difference in carryover effects has been proposed by Wellek (2003). However, as discussed later in Chapter 4, carryover effects can be excluded in bioequivalence studies based on medical grounds since healthy volunteers are recruited and an adequate washout has to be chosen. » 6. Does it mean that the method in the paper as a whole is more extensive than the data applied by the "Wilcoxon signed rank test". In other words, the method in the paper has fewer assumptions than the "Wilcoxon signed rank test". Well, I would say to neglect period effects after Hauschke^{2)} proposal is out of date. ^{1} Hauschke D, Steinijans VW and Pigeot I. Bioequivalence studies in drug development methods and applications. John Wiley & Sons, New York (2007). ^{2} D. HAUSCHKE, V. W. STEINIJANS and E. DILETTI A distributionfree procedure for the statistical analysis of bioequivalence studies International Journal of Clinical Pharmacology, Therapy and Toxicology, Vol. 28 No. 2 1990 (7278) / Vol.30, Suppl. No. 1 1992 (pp,S3743) — Kind regards, Mittyri 