The Outlaw Torn
★    

Europe,
2015-05-13 14:49
(3242 d 11:00 ago)

Posting: # 14802
Views: 11,018
 

 The power of post hoc power [Power / Sample Size]

Good day to all,

Another return to post hoc power analysis to raise Helmut's blood pressure. Ha!

Is post hoc power analysis also useless for therapeutic equivalence trials (non-inferiority, two arm, active-controlled, parallel group) intended to demonstrate the IOP-lowering efficacy of a generic anti-glaucoma medication versus the originator?

My first thought is that, yeah, it's the same principle as with standard bioequivalence, but since the confidence intervals are measured in mm Hg, I become uncertain as a non-statistician.

Questions are being raised that the sample size may have been inadequate.

Any comment by anyone is appreciated and thank you in advance.

Outlaw Torn


Edit: Category changed. [Helmut]
ElMaestro
★★★

Denmark,
2015-05-13 19:53
(3242 d 05:56 ago)

@ The Outlaw Torn
Posting: # 14810
Views: 9,835
 

 The power of post hoc power

Hi Outlaw Torn,

❝ My first thought is that, yeah, it's the same principle as with standard bioequivalence (...)


You were right from the beginning.

❝ Questions are being raised that the sample size may have been inadequate.


And that could also be correct. Power is based on assumptions. If those assumptions are wrong (like the point estimate, the variability) then that is an issue. So perhaps you'd be revising your assumptions when you do power/Sample sze for the next trial.

Pass or fail!
ElMaestro
d_labes
★★★

Berlin, Germany,
2015-05-13 22:18
(3242 d 03:30 ago)

@ ElMaestro
Posting: # 14811
Views: 9,832
 

 Ad hoc power

Hi Outlaw Torn, Hi Great Maestro,

❝ ... Power is based on assumptions. If those assumptions are wrong (like the point estimate, the variability) then that is an issue. So perhaps you'd be revising your assumptions when you do power/Sample size for the next trial.


And this is not post hoc power but ad hoc power/sample size estimation as always necessary in the design step of a study.

Regards,

Detlew
The Outlaw Torn
★    

Europe,
2015-05-14 13:38
(3241 d 12:10 ago)

@ d_labes
Posting: # 14813
Views: 9,707
 

 Ad hoc power

❝ ❝ ... Power is based on assumptions. If those assumptions are wrong...❝


❝ And this is not post hoc power but ad hoc power/sample size estimation as always necessary in the design step of a study.


Although the results of this trial suggests a greater sample size would be needed in order to repeat the same trial (I mean, shit happens, right?), the initial assumptions can be supported.

Thank you Maestro and d_labes, your points are clear.
The Outlaw Torn
★    

Europe,
2015-06-15 13:20
(3209 d 12:29 ago)

@ The Outlaw Torn
Posting: # 14955
Views: 9,198
 

 Ad hoc power

Good morning, all. I'd like to post a follow up on this topic. Hope you don't mind.

Would it be correct to state the following regarding a non-inferiority trial for an ophthalmic drug:

In a non-inferiority trial, the type-I error rate (alpha value, α), represents the consumer’s risk (the risk of concluding in favor of the non-inferiority hypothesis when, in reality, the product is inferior) while the type-II error rate (beta value, β) represents the producer’s risk (the risk of concluding in favor of the hypothesis of inferiority when, in reality, the product is really non-inferior).

What I'm getting at is this. When conducting a non-inferiority trial, are the roles of the type-I and type-II error rate reversed since the null hypothesis and the alternative hypothesis are reversed (Null Hypothesis H0 states non non-inferiority (i.e. inferiority) while Alternative Hypothesis H1 states non-inferiority between Test and Reference products)?

Thank you for any feedback you may have,
The Outlaw Torn
mittyri
★★  

Russia,
2015-06-16 23:52
(3208 d 01:57 ago)

@ The Outlaw Torn
Posting: # 14962
Views: 9,140
 

 Hypotheses in non-inferiority trials

Hi Outlaw Torn

❝ What I'm getting at is this. When conducting a non-inferiority trial, are the roles of the type-I and type-II error rate reversed since the null hypothesis and the alternative hypothesis are reversed (Null Hypothesis H0 states non non-inferiority (i.e. inferiority) while Alternative Hypothesis H1 states non-inferiority between Test and Reference products)?


Let me cite Emmanuel Lesaffre:
"Demonstrating non-inferiority necessitates rejecting the null hypothesis (H0: ∆ > ∆NI) in favor of the alternative hypothesis (HA: ∆ < ∆NI) with an adapted classical statistical test (as for an equivalence test). A significant result (p < 0.025) means, in this case, that the experi­mental treatment is not (much) worse, i.e., non-inferior, to the control treatment, whereby the conclusion “non-inferior” depends on the chosen value for ∆NI."
I highly recommend you to read his great article (for me it's the best one). I think the relationships between equivalence, non-inferiority and superiority will be more clear.

Kind regards,
Mittyri
The Outlaw Torn
★    

Europe,
2015-06-17 13:35
(3207 d 12:14 ago)

@ mittyri
Posting: # 14963
Views: 9,114
 

 Hypotheses in non-inferiority trials

I can't seem to be able to respond to Mittyri (the system logs me out when I press the respond button), so I'll respond to Mittyri here via responding to myself.

Thank you, Mittyri. I've printed it out and browsed it. Will grab another coffee and re-read it with my full attention.

In my research, I keep running into statements like this: Type II error has heightened importance in non-inferiority trials and must be managed. If sample size is inadequate, then a non-inferiority trial may lead to a false claim of a drug being non-inferior to a comparator when in fact it is worse.

These types of statements seems to undermine the argument on post hoc power—they seem to claim that post hoc power is meaningful. Am I misinterpreting these kind of statements? Are they valid?

Thank you,
Outlaw Torn


Edit: Strange – I moved your post in the database. [Helmut]
d_labes
★★★

Berlin, Germany,
2015-06-18 12:17
(3206 d 13:32 ago)

@ The Outlaw Torn
Posting: # 14966
Views: 9,012
 

 Hypotheses in non-inferiority trials

Dear Outlaw

❝ ... Am I misinterpreting these kind of statements? Are they valid?


As far as I understand that statement: It must be from or leftover from the ancient times where non-inferiority trials or equivalence trials where tried to evaluate via the "normal" statistical tests (for non-equality) but setting alpha and beta errors reversed. I.e. high alpha, low beta, f.i. alpha=0.2, beta=0.05.

Regards,

Detlew
The Outlaw Torn
★    

Europe,
2015-06-22 12:23
(3202 d 13:26 ago)

@ d_labes
Posting: # 14968
Views: 8,831
 

 Hypotheses in non-inferiority trials

I still can't respond to the last comment in a thread without getting logged out. Is this a common glitch?

d_labes, sorry for the delay responding, I was in the Nederlands. I'm still trying to wrap my head around these types of comments, but it seems the issue related to these comments is resolved. I hope in our favor. Thank you.


Edit: I moved your post in the database. No, this is a very uncommon glitch – I’ve never seen it before… :confused: [Helmut]
arjunroy
☆    

India,
2015-06-29 10:24
(3195 d 15:25 ago)

@ The Outlaw Torn
Posting: # 14994
Views: 8,630
 

 Hypotheses in non-inferiority trials

Time is changing so do p-value calculations of NI, because everybody likes p-value.
Some of the publication (refer below: NEJM Effect of Sitagliptin on Cardiovascular Outcomes in Type 2 Diabetes, 2015).

Sitagliptin was noninferior to placebo for the primary composite
cardiovascular outcome (hazard ratio, 0.98; 95% CI, 0.88 to 1.09; P<0.001).
Though hazard ration includes 1 however p-value for the test of non-inferiority was noted significant.
The P value is for the noninferiority of sitagliptin, as compared with placebo, which was calculated by determining whether the upper boundary of the two-sided 95% confidence interval of the hazard ratio exceeded 1.30.
Hope this helps.


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! [Helmut]
UA Flag
Activity
 Admin contact
22,957 posts in 4,819 threads, 1,636 registered users;
88 visitors (0 registered, 88 guests [including 12 identified bots]).
Forum time: 00:49 CET (Europe/Vienna)

Nothing shows a lack of mathematical education more
than an overly precise calculation.    Carl Friedrich Gauß

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5