Version 1.3-6
Released 2016-06-06 via CRAN
Released 2016-05-04 to beta-testers
- Functions for power and sample size of BE decision of highly variable
drugs or drug products via ABEL (average BE with expanding limits)
updated to incorporate the reagulatory settings of Health Canada
(new functions power.scABEL2() and sampleN.scABEL2(), based on simulations
of intra-subject contrasts evaluation).
- The same functions now accept as regulator argument an object of class
'regSet' and allow via this way User definitions of regulatory settings.
- New function reg_const made visible to define objects of 'regSet'.
Class 'regSet' has an S3 print method.
- Default value of argument theta0 in functions power.scABEL(),
power.scABEL2() and power.RSABE() changed to 0.90 to be in agreement
with the setting in the corresponding sample size functions.
- Function scABEL (calculation of widened acceptance limits) no longer
accepts regulator="USER". This case may be handled via an object
with class 'regSet', defined by help of function reg_const().
- Functions CVfromCI()/CI2CV now use pe instead of point as argument due
to more consistency with their dual CI.BE(). For backward compatibility
point may be used also but then a warning is thrown. Argument point will
be removed in future versions.
- Functions for expected power for TOST and non-inferiority updated
to avoid numerical dificulties with integrate() if method="exact".
- Documentation (typos in man pages) rigorously enhanced.
Thanks to Helmut and Ben.
Version 1.3-5
Released 2016-04-12
- Functions for expected power for TOST and non-inferiority updated
(including corresponding sample size functions). Functions now include
an exact method that calculates the expected value of the power with
respect to the (prior) distribution of sigma^2 (inverse gamma distribution).
Former only an approximation according to Julious/Owen was implemented.
Exact implementation contributed by B. Lang.
- The functions for iteratively adjusting alpha for the EMA recommended
ABEL method now have a new argument tol for convergence tolerance of the
internal used uniroot(). Was former hardcoded as tol=1e-5.
Contributed by H. Schuetz.
- plot method of 'pwrA' objects reworked by H. Schuetz to avoid some
ugly overlays.
Version 1.3-4
Released 2016-03-09
- Bug removed in all sample size estimation functions for scaled ABE which
resulted for very low variability in NA for sample size.
Thanks to Shuanghe for detecting this bug.
- All functions for scaled ABE now have an argument imax=100 for the maximum
number of steps for sample size search. Wish of Helmut Schuetz.
Version 1.3-3
Released 2016-01-15 via CRAN
Released 2016-01-04 to alpha testers
- New functions scABEL.ad() and sampleN.scABEL.ad() introduced to
iteratively adjust alpha in order to maintain the consumer risk
and adapt the sample size to compensate for the possible loss in
power for the EMA method of scaled ABE (ABEL).
Contributed by Helmut Schuetz.
- Default theta0 changed to 0.9 in sample size estimation for scaled ABE
(as recommended by the Two Laszlo's).
Version 1.3-02
Released 2015-12-02
- New functions for calculation of power, type 1 errror and sample size
for 2 TOST instances. Contributed by Benjamin Lang.
Version 1.3-01
Released 2015-09-30
Released 2015-09-23 to alpha testers
- New functions power.HVNTID() and sampleN.HVNTID() introduced to calculate
power and sample size for the BE decision via the FDA procedure for
highly variable NTID's (see FDA Dabigatran / rivaroxaban guidances).
- power.HVNTID() and power.NTIDFDA() now return the power (p(BE)) and
the components for the scaled ABE criterion, the conventional ABE test
and the test for the ratio swT/swR <= 2.5 if the argument details=TRUE
(wish of Helmut Schuetz).
- power.RSABE() now returns the power (p(BE)) and the components for the
scaled ABE criterion, for the point estimate criterion and for the
conventional ABE test alone if the argument details=TRUE.
Analogous changes were made in power.scABEL(). See man pages.
Version 1.2-09
Released 2015-08-26
- power.TOST() with method="exact" reworked to give correct values in
case alpha>0.5. OwensQOwen adapted to deal with upper integration
limit R==Inf.
- New power calculation method introduced, direct integration of the
bivariate non-central t-distribution (pmvt() from package mvtnorm),
also an exact calculation method but with somewhat lower precision
and longer run-time. Contributed by Benjamin Lang.
- New function power.TOST.sim() to obtain the power via simulations of
the TOST. Only intended for checking purposes.
Version 1.2-08
Released 2015-07-10
- Maintenance release with NAMESPACE adapted to import the necessary
functions from base R installation (CRAN request)
- power.NTIDFDA() and sampleN.NTIDFDA() now have an design argument to
choose between 2x2x4 (full replicate 4-period) and 2x2x3 (full replicate
3-period design)
- Some minor documentation improvements
Version 1.2-07
Released 2015-06-03
- Functions power.scABEL(), sampleN.scABEL() now allow the power and
sample size calculations for scaled ABEL according to the (inofficial)
regulatory settings of Brasilian ANVISA
- Function pa.scABE() also allows regulator="ANVISA"
- New helper function scABEL for calculation of the (widened) ABE acceptance
limits
- Function power2.TOST() now removed
Version 1.2-06
Released 2015-01-23
- Function power.TOST() now handles balanced as well as unbalanced studies
- Function power2.TOST() is therefore depreciated and will not be contained
in later versions
- Internal change in interface to the (hidden) 'raw' power functions
(sem used in function calls instead of se, n, bk)
- BE_power_sample_size_excerpt.pdf in subdirectory doc changed to reflect
the internal changes
Version 1.2-05
Released 2015-01-07
- Bug in OwensQOwen() removed (debug code left over), THX to Helmut
for discovering this
- Function CVfromCI() and alias CI2CV() now handle unbalanced studies,
contributed by Benjamin Lang
- Some typos in BE_power_sample_size_excerpt.pdf, found by Ben, corrected
Version 1.2-04
Released 2014-12-19
- Bug in OwensT(h,a) for the case of a=-Inf removed. Thanks to
Benjamin Lang for discovering that nasty bug
- Functions power.noninf(), exppower.TOST(), exppower.noninf() adapted to
deal with unbalanced studies
- Function pvalue.TOST() introduced to calculate the two p-values of the
TOST procedure. Contributed by Benjamin Lang
Version 1.2-03
Released 2014-11-13
- sampleN.dp() / power.dp() now cover "incomplete block designs"
- minimum sample size in sampleN.dp() introduced to avoid errors for
small CV <= 0.1 for the crossover design
- partitition of n(total) to (sequence) groups reworked in most power
functions if design is unbalanced
- Liu's 2x2x2 design with 2 repeated measurements in each period added
(see ?known.designs under Notes)
Version 1.2-02
Released 2014-10-06
- functions pa.XYZ() adapted so that they also function under R < 3.1.1
(request of U. Ligges from CRAN) although with minpower>=0.5
Version 1.2-01
Released 2014-09-30
- Some minor bugs removed and spelling in man pages corrected
- print method for class pwrA now calls plot()
- power analysis in case of n=12 for the plan now drops n below 12 subjects
Version 1.2-00
Build 2014-09-19, released to alpha/beta testers only
- Contains now functions for power analysis of a sample size plan
for ABE (pa.ABE()), scaled ABE (pa.sABE()) and scaled ABE for NTID's
(pa.NTIDFDA()) analyzing power for deviations from assumptions for
the sample size estimation
- Contains further experimental functions for power calculations / samplesize
estimation for dose proportionality studies using the Power model
Version 1.1-13
Released 2014-08-12
- Design constant bk(ni) for design="paired" corrected.
THX to Helmut Schuetz who detected that bug.
Version 1.1-12
Released 2014-07-02
- Some minor improvements in documentation.
- Function power.TOST() now throws a warning if used with imbalanced
designs (n not an even multiple of no. of sequences)
- Internal change of start value of sample size search in sampleN.TOST()
to avoid failed searches if variability is high and theta0 is near 1
Version 1.1-11
Released 2014-04-30
- Utility function added which calculates 1-2*alpha confidence interval(s)
given point est., CV and n using log-tansformed evaluation.
- Utility function added which calculates 1-2*alpha Fieller confidence
interval(s) given point est., CV (, CVb) and n for the ratio of
un-transformed means.
Version 1.1-10
Released 2014-01-31
- The package is now pre-compiled to 'byte code' via the compiler
package for speed reasons. F.i. OwensQOwen() gains a sixfold boost.
- Some bulk code changes to make the power calculations for extrem
cases more bullet proof without computation time burden.
- Bug removed where noncentrality parameters in the power calculations
became NaN (not a number).
- OwensQ() now tries to return a value via nct-approximation if its value
is due to numeric difficulties falsely equal zero.
This approximation is up to 6 decimals correct as far as tested.
OwensQ() issues a warning if the nct-approximation is used.
- Some improvements in documentation (man pages and *.pdf) to
reflect the code changes
Version 1.1-09
Build 2014-01-03 Not released to the public.
- OwensQ() now uses OwensQOwen() in case of high delta and/or high upper
integration limit. Thus extrem cases can be handled properly where
the former implementation via integrate() was prone to fail.
Thanks to Jiri Hofmann and Helmut Schuetz for pointing me to
such extrem cases.
Version 1.1-08
Released 2013-12-27
- Dates typo in the history (NEWS) corrected from PowerTOST 1.1-00 on.
THX to Julien Grassot.
- CVfromCI() now accepts either both CLs or one CL and the point
estimator, programmed by Helmut Schuetz.
- power.RatioF() and sampleN.RatioF() now have an argument setseed=TRUE
which avoids the dependence if the power from the state of the
random number generator (due to the calculation method of pmvt()
from package mvtnorm). Thanks to Benjamin Lang.
Version 1.1-07
Released 2013-09-02
- design 2x2x3 (TRT|RTR) implemented in power.scABEL(), sampleN.scABEL()
and in power.RSABE(), sampleN.RSABE().
- power.scABEL() now gives a warning if CVwT not equal CVwR in the design
2x3x3, partial replicate.
- Simulation details for the full replicate design slightly changed
to obtain better numeric agreement of power to subject data sims.
Version 1.1-06
Released 2013-06-21
- Fat bug corrected in sampleN.NTIDFDA() which lead to false sample size
for cases where the test of equal variabilities of Test vs. Reference
comes into effect.
Version 1.1-05
Released 2013-06-17
- The functions sampleN.scABEL(), sampleN.RSABE() and sampleN.NTIDFDA()
return now a data.frame with the input and the sample size result.
The "Sample size" column contains the total sample size.
The "nlast" column contains the last n value handled. May be useful for
restarting.
- Start values for sample size search in functions sampleN.scABEL(),
sampleN.RSABE() reworked.
Failed sample size searches are now more seldom observed.
- Flaw in sample size search removed which does not indicate a failed
search (n=NA) if started with a too high n (Thanks to Helmut Schütz).
Version 1.1-04
Released 2013-05-15 to alpha testers only
- Functions power.NTIDFDA() and sampleN.NTIDFDA() introduced to calculate
power and sample size for the BE decision via the method of the FDA for
narrow therapeutic index drugs (NTID's). Power and sample size
are based on simulations.
Version 1.1-03
Released 2013-05-03
- Flaw in implementing the sims of 2x3x3 design with different intra-subject
variabilities in functions power.RSABE() and sampleN.RSABE() as well as
in functions power.scABEL() and sampleN.scABEL() disposed.
- Methods and implementation details of the simulations for scaled ABE
documented in a PDF file in the /doc subdirectory of the package.
- Simulation method for the EMA scaled ABEL changed to conform better
with power values from simulations via subject data.
- Warning section in the help file of power.scABEL() introduced to
appoint the fact that simulations via subject data and via the
methods implemented in power.scABEL() gave empirical power values
that are only approximately in agreement.
Version 1.1-02
Released 2013-02-28
- Functions power.RSABE() and sampleN.RSABE() introduced to calculate
power and sample size for the BE decision via linearized scaled
ABE criterion as favored method of the FDA. Power and sample size
are based on simulations.
- Default of nsims changed to 1E5 (suggested by Helmut Schuetz).
- Argument setseed introduced in the scaled ABE functions to avoid
different outcomes depending on the state of the (pseudo) random
number generator. If setseed=TRUE a set.seed(123456) is issued before
each call of the simulation functions.
- Documentation improved.
Version 1.1-00
Released 2013-02-08
- Functions power.scABEL() and sampleN.scABEL() introduced to calculate
power and sample size for the BE decision via scaled (widened)
BE acceptance limits based on simulations.
Thanks to Helmut Schuetz for power testing these functions.
http://forum.bebac.at/mix_entry.php?id=9997
- False News file included. Sorry for that.
Version 1.0-01
not released, integrated in 1.1-00
- Upper one-sided CL of the CV and therefore argument alpha2 removed from
expsampleN.noninf() and expsampleN.TOST() because it lead to some
confusion in users thinking this had to do with the algorithm via
expected power.
- CI2CV() as alias to CVfromCI() introduced because my self always typed
this name if aimed to calculate the CV from a given CI.
- Function CVCL() now returns a 2 element vector also if an one sided
interval is requested.
Version 1.0-00
Released 2012-10-26
- Bugfix in power.noninf() to get the correct power if theta0 is below
margin (if margin <1) or theta0 is above margin (if margin >1).
power.noninf() calculated up to now the power of an inferiority test.
Thanks to Helmut Schuetz.
- New helper functions CV2mse() and mse2CV.
- New function CVCL() to calculate a confidence interval of a CV.
Version 0.9-11
Released 2012-08-07
- Bug fix in power.TOST(), power2.TOST() and power.noninf() to use the
right df depending on argument robust. robust=FALSE had used the
robust df's. Thanks to Ben.
Version 0.9-10
Released 2012-07-20
- Bug-fix in internal function .Q.integrand()
- Bug fix in nmin - must be a multiple of steps to assure balance
in sequence groups.
Again thanks to Helmut Schuetz for detecting both.
Version 0.9-9
Released 2012-07-18
- Workaround introduced to handle numeric problems in integrate()
if CV<5.3E-6. Thanks to Helmut Schuetz.
- Minimum sample size adapted to design used (f.i. n=2 if design="paired")
Version 0.9-8
Released 2012-04-05 (Easter egg)
- Functions added for 'expected' power and sample size calculations based
on it for the non-inferiority test for sake of completeness.
- Functions CVpooled(), exppower.TOST() and expsampleN.TOST() now
also implemented for logscale=FALSE, i.e. contain that argument in
their calls.
- Function OwensQOwen() made public. This is an implementation of the
algorithm given by Owen itself in the original paper (Biometrica 1965)
via repeated integration by parts.
This function is only for comparative purposes.
- Function OwensT() made public. It is needed internal in OwensQOwen()
but may be useful for other purposes.
Version 0.9-6/7
Released 2012-03-26
- PowerTOST 0.9-7 is a small bug fix PowerTOST required from CRAN.
Contains the old PowerTOST number 0.9-6 in the package man page.
- Functions added for power and sample size calculations based on
non-inferiority t-test. This is not a TOST procedure but eventually
useful if the question of 'non-superiority' within a BE study must be
evaluated.
Hint: Evaluation of Fluctuation in the EMA MR NfG (1999) between
modified release formulation and immediate release product.
Version 0.9-4
Released 2012-03-05
- Little bug in sampleN.TOST removed which causes extra doubled output
of n and power if n=4. Thanks to Ben:
http://forum.bebac.at/mix_entry.php?id=8206
Version 0.9-3
Released 2012-02-13
- Bug in power.TOST removed which prevented calculation of Power
according to method ^= "exact".
- Sample size tables for replicate design 2x2x3 in data sub-directory added.
- Sample size tables for replicate design 2x4x4 in data sub-directory added.
- Scripts in the /test sub-directory made available.
Version 0.9-2
released 2011-12-24
- Function power2.TOST() added to allow power calculations for
studies with unbalanced (sequence) groups.
- Argument exact replaced by method in power.TOST(), sampleN.TOST().
See help for these functions.
- Attention! The sample size for the parallel group design is now
the TOTAL sample size (to be consistent across all functions).
- Sample size tables added for the 2x2 crossover and for the parallel
group design to alleviate validation/qualification of the package.
See data(package="PowerTOST").
- Scripts added in the \tests sub-directory that create the sample size
tables from the data section.
- Updated BE_power_sample_size_excerpt.pdf in the doc sub-directory.
Version 0.9-0
released 2011-12-15
- Paired means design introduced
- 'robust' argument added to nearly all functions.
With robust=TRUE the degrees of freedom for the so-called robust
evaluation (df2 in known.designs() output) will be used.
This may be helpful if planning is done for higher order designs
evaluated via mixed model or via intra-subject contrasts
(aka Senn's basic estimator).
- Due to the necessary NAMESPACE from R14.0 on the internal functions
(names starting with .) are no longer exported.
Version 0.8-7
released 2011-10-20
- Problem with slash in \name field of manual resolved
(requested by B. Ripley)
Version 0.8-6
released 2011-05-18
- Bug removed which gave incorrect exact power values in case of
alpha>=0.5 (very unusual setting). Thanks again to Craig Zupke.
Cross checked results of power at equivalence margins against
SAS Proc Power.
Version 0.8-5
released 2011-05-16
- Code in Owen's Q adapted to account for large delta or large b
leading to integrand function almost zero over the whole range
which then gave an error in integrate(). Thanks to Craig Zupke.
Version 0.8-4
released 2011-03-11
- Number of maximal steps in sample size search in the sample size
functions, formerly hard coded as 50, made accessible to users via
argument imax. but needs only adapted in rare extremal cases.
- Start value for sample size search improved around theta0=1 (logscale)
or theta0=0 (untransformed).
Version 0.8-3
released 2011-01-18
- Bug fix in known.designs().
Version 0.8-2
released 2011-01-10
- error in df for 3x3 and 4x4 crossover removed
- function for pooling CV's from different studies made public
see ?CVpooled
Version 0.8-1
released 2010-11-25
- helper function CVfromCI() added to estimate the CV from a confidence
interval. Useful if no CV but CI was given in literature.
Version 0.7-3
released 2010-10-09
- bugs in documentation removed
- input argument diff removed from sampleN.TOST, expsampleN.TOST,
power.TOST and exppower.TOST
Version 0.7-2
released 2010-08-27
- little bug causing warnings in case of 2x2 alias 2x2x2 removed
Version 0.7-1
released 2010-08-12
- functions added for the power and sample size for the ratio of
two means with normally distributed data on the original scale
(based on Fieller's confidence (fiducial) interval).
AFAIK until now only implemented in the commercial nQuery.
- Note that the parameter diff (Null ratio / Null diff.) is now
named theta0 since it was annoying for users to call it diff in case
of ratios (logscale=TRUE). The parameter diff is still supported
but will be ceased in the next release. Therefore a warning is issued
if diff is used.
Version 0.6-2
released 2010-07-21
- some internal code consolidation
- internal hidden functions .CV2se() and .se2CV() made public
- minor enhancements in help pages
- more examples
- short documentation of used statistical apparatus
BE_power_sample_size_excerpt.pdf
for classical power / sample size in directory /doc added
Version 0.5-1
- first release to CRAN on 2010-05-07