Bioequivalence and Bioavailability Forum

Hi Simon,

❝ The Phoenix 1.4 release is now officially available on the Support page;

---

System Requirements

Phoenix 1.4 is supported for both 32-bit and 64-bit versions of Windows 7 SP1 and Windows 8.1. Phoenix 1.4 is also supported for 64-bit versions of Windows 2008 Server R2 and Windows Server 2012 R2.

Minimum Hardware Requirements:
Processor: Intel Core 2 Duo 2.0 Ghz, or a CPU of equivalent processing power
RAM: 4 gigabytes of RAM (8 gigabytes or more is recommended)
Hard drive space: 300 megabytes free

---

Wow, 516MB zipped.

I would appreciate if you could post (here or there) the method of keeping v1.3 while installing v1.4. I think this is useful for people who don’t want to immediately change their running system (never, never!)…

---

Testing: Hey, the inverse t and χ² – saves one step in the RSABE-template! Nice ‘ANOVA’-tables (similar to SAS’ GLM output). After sooo many years I don’t have to endure the ugly looks of Westlake’s symmetric CI any more. Power – for our extreme and imbalanced data­set H PHX1.4 spits out…

Power_TOST = 0.966301  (significance level = 0.0500)

…fitting nicely to my ‘Gold Standard’…

library(PowerTOST)
power2.TOST(CV=0.992664, n=c(288, 429), theta0=0.934232)
[1] 0.9663017

If you generally evaluate stu­dies with EMA’s fixed ef­fects model you can change the default set­up (☑ the box). Although not stated, the set­up works for higher-order cross-overs as well. As a side-effect you don’t get the intra-subject CV in the output any more and have to setup a custom-transformation from the residual variance. Why?

Still:

• Linear trapezoidal the default. IMHO, lin-up/log-down is a better choice.
  
• Power_TOST for the actual CV/PE. Makes Southeast Asian regulators happy and hope­fully I will not have to read “power=0.999” in Indian reports of failed studies any more. I asked Phar­sight for a field in the UI which would allow to enter a desired value (e.g., 0.95 for Pot­vin’s / Montague’s TSDs and 0.9 for Fuglsang’s). Maybe in the next release?
  
• No transparent background of the plotting area in exported PNGs and EMFs possible. Always #FFFFFF.
  
• Only merging/joining two worksheets. In FDA’s full replicate RSABE we need three opera­tions instead of one in SAS.

In the User’s Guide p544 (p560 of the PDF) I read for replicate designs:

4. Select the Variance Structure’s Random 1 tab.

• Formulation is automatically selected in the Random Effects field. Do not change this setting.
  
• Subject is automatically selected in the Variance Blocking Variables (Subject) field. Do not change this setting.
  
• Banded No-Diagonal Factor Analytic(f) is automatically selected in the Type menu. Do not change this setting.
  
• 2 is automatically entered in the Number of factors (f) = field. Do not change this setting.
  

A user can expect that about 50% of datasets analyzed will produce a non-positive definite G matrix. This does not imply that the model-fitting is invalid, but only that a user must be careful not to over-interpret the variance estimates. The confidence interval on the formulation difference will still have the expected statistical properties.

Well, for some partial replicate design datasets one gets no convergence at all. Don’t leave users out in the rain! “Do not change this setting” is not a particular good advice because factor 1 (instead of the 2 according to FDA’s guidances) would do the job.