From firstname.lastname@example.org Tue Jun 17 19:56:59 1997
Subject: Sample size
I am beginning preliminary work to use NONMEM4 to estimate the population PK of two cytotoxics in a cooperative grooup Phase III trial. I would like to use just two samples from each patient, similar to what Dr. Bruno et al reported in their Phase I and II work with docetaxel. Can the group point me in a direction to learn:
1. How should the sampling times be defined and distributed based upon prior known PK parameters for the drug? Should random times be selected (on one extreme) or points suggested by an algortihm such as ADAPT's Sample?
2. How does one define the necessary sample size for a limited sampling study? The clinical study will retrospectively attempt to test the association between PK and response or survival duration. Response would be tested with logistic regression, along with other non-PK parameters if they contribute. I am unsure about the optimal process for the more continuous variable of survival or response duration.
I appreciate any guidance that the group can provide. Thanks
Paul Hutson, Pharm.D.
Associate Professor (CHS)
UW School of Pharmacy
From ETTEE@cder.fda.gov Wed Jun 18 14:29:43 1997
Subject: Re: Sample size
It would be useful to use your mean PK parameters from Phase I study(ies) with the appropriate model to obtain informative times (using the SAMPLE module in ADAPT II)and create sampling blocks (windows) around the informative times. Patients should then be randomly allocated to these times. The idea here is to be able to describe the population profile. This informative profile (block) randomized design protects against model misspecification.
The issue of obtaining 2 samples/subject depends on the number of
parameters you want to estimate. If, for instance, you are dealing with a one compartment model with intravenous input(s) you will be able to obtain reliable (accurate and precise) estimates for both fixed and random effects parameters with the above sampling design.
It would be best to carry out a simulation study to determine the
appropriate sample size for your study.
There are a few references which I think would be helpful:
(1) Jones CD, Sun H, Ette EI. Designing cross-sectional pharmacokinetic studies: implications for pediatric and animal studies. Clin Res Regul Affairs 1996; 13 (3&4): 133-165.
(2) Ette EI, Sun H, Ludden TM. Design of population pharmacokinetic studies. Proc Am Stat Assoc (Biopharmaceutics Section) 1994; pp 487 - 492.
(3) Ette EI, Sun H. Sample size and population pharmacokinetic parameter estimation. Clin Pharmacol Ther 1995; 57 (2): 188.
From rene.BRUNO@rhone-poulenc.com Thu Jun 26 08:14:10 1997
Subject: RE : Sample size
Sampling times : I basically agree with the proposal of Ene to compute optimal times and then randomly allocate those times to patients. It is exactly what we did for docetaxel to assess population PK eraly during its development and it worked. This approach could well be done again for new studies. However, we
have now a "validated" population PK model drawn from 547 patients (see JPB 24(2), 153-172, 1996). We recently used this information to reassess optimal sampling times and demonstrated that it was possible to get unbiased and precise estimates of docetaxel CL from 2 points (5 min prior to the end of the 1 hour infusion and 5 hours after) (Clin. Cancer Res. in press). I believe that this sampling strategy could be used for further studies with minimal disturbance of the clinical trial. I can of course provide you with the priors to be used for Bayesian estimation using NONMEM.
Sample size : In our studies using data from 640 patients we were able to demonstrate that docetaxel exposure at first cycle was a predictor of first hematologic toxicity (incidence of Grade 4 neutropenia and febrile neutropenia, logistic regression models). Such a huge sampling size was probably not required for Grade 4 neutropenia which was quite frequent; however it was definitely required for febrile neutropenia which was, fortunately, much less frequent. In this model CL/AUC were better predictors than time over any threshold level. First cycle exposure was also a predictor (to a lesser extent however) of the time to onset of fluid retention (Cox model).
Regarding efficacy endpoints we did not found any predictive power of exposure for response rate and time to first response in breast and NSCLC. However, the sampling size was less (< than 200 patients) and I believe that a larger sample size would be required because systemic exposure is much less directly related
to tumor shrinkage.
The idea of Ene to perform a simulation study is a good one to assess sampling size for the PK study; however, I don't see how it would work to assess sample size for PK/PD and particularly efficacy since I believe it would require the use on at least a preliminary PK/PD model. But may be I am wrong I have no experience with simulation.
For the analysis of time to event variables, you need to use a model for survival analysis (e.g. COX model or any other parametric model such as exponential Weibull ...).
Drug Metabolism and Pharmacokinetics