Dear Colleagues,
In response to the recent discussion in this forum about sample size
and power in NONMEM, Next Level Solutions has developed tools for
formal "power" assessment in NONMEM using simulation (as described by
Serge Guzy in this discussion). Essentially, one or more (preferably
more) models will be provided to Next Level Solutions by the sponsor,
along with proposed sampling schemes. Additional optional information
includes drop out rates, sampling time errors (reported and
unreported), missing sample probability, trial design (cross over,
parallel), LLQ and dosing time inaccuracies (mean and sd).
Sampling schemes will be generated (with the described random effects)
and NONMEM analysis by bootstrap (with multiple sets of initial
estimates to reduce the problem of local minima). The generated report
will include:
-Probability of model convergence
-Probability of successful covariance step
-Probability of the absolute value of all off-diagonal elements of
the correlation matrix of estimation being < 0.95.
-Mean and SEE for all parameters (by bootstrap, not NONMEM estimates
of SEE)
-Probability of picking up a hierarchical effect (e.g., effect of age
on clearance vs no effect of age on clearance), given the effect size
described by the model(s) and a log likelihood ratio test.
Turnaround time for analysis will depend on the run times for the
models, but typically is a few days to a week - most of which is
computer run time (1000 samples and NONMEM analyses for each sampling
scheme, or twice that if reduced vs full model hypothesis is to be
tested)
Thanks
Mark