Version 0.5.2 (easter egg)
2019-04-21
- Maintenance release with mainly bug fixes.
- Bug fix in calculation of median unbiased estimate within final.tsd.in().
- Typos and minor wording and formatting corrections (man pages and x.tds.in()).
- Bug fix in power.tsd.GS (regarding variable hw).
- Bug in interim.tsd.in() w.r.t to power > target triggering
futulity fixed (THX to mittyri).
Version 0.5-1 (belated easter egg)
2018-04-03
- All functions xyz.2stage.ab() now have an alias xyz.tsd.ab(), e.g. instead
of power.2stage.fC() one may use power.tsd.fC().
The versions with .2stage. in their names will be removed in later versions.
- The futility rule for a CI in power.2stage.GS() now based on 90% CI
to be in line with the other functions using such a futility rule.
- Default fCrit in power.2stage.GS() changed to "CI".
- Added interim.2stage.in() and final.2stage.in() to perform interim and
final analysis of 2 stage designs based on the Standard Combination or
Maximum Combination Test (also includes a print class).
- Update defaults of power.2stage.in() in order to be consistent with
Paper from Maurer et al (2018).
- Aliases power.tsd.xyz introduced (not public yet).
- Experimental function power.tsd.2m() to deal with 2 PK metrics
in the TSD decision scheme Potvin "B". Not public yet.
Version 0.4-6
2017-10-27
- Updates of various man pages.
- New function sampleN2.TOST() to estimate n2 in 2x2 and parallel designs
based on df=n-3 for stage 2.
- df for sample size re-estimation changed to n-3 for Potvin like TSD schemes
(pooled evaluation with additionally stage term)
- Same for power.2Stage.ssr().
- Function power.2stage.in() to calculate the operational characteristics of
TSDs evaluated via p-value combination using the inverse normal approach.
Contributed by Benjamin Lang.
- GitHub URL and bug reports URL added.
Version 0.4-5
2017-01-18
- DOI for references in man pages added.
- Bug in power.2stage.p() and power.2stage.pAF() removed which causes
negative residual variances in case of test="anova" and imprecise power
values in all other cases for certain scenarios of n1, CV.
- Large sample approximation to the sample size adapted to deal with
unsymmetric BE acceptance ranges, one limit = Inf allowed
- Maintenance release to refelect changes in power calculation of
package PowerTOST
- Some small cosmetic changes in code w.r.t min.n2
Version 0.4-4
2016-01-21 (not released via CRAN)
- Method "B0" (re)introduced in power.2stage() and power.2stage.fC() which
employs the decision scheme of the so-called MSDBE (Modified Sequential
Design for BE) of Zheng et al.
- Method "B" in power.2stage.p() / power.2stage.pAF() also adapted to
decision scheme E of Xu et al.
- Figures with TSD decision schemes added in /doc subdirectory.
Version 0.4-3
2015-11-24
- power2stage.fC() now has a new argument max.n which constraints the total
sample size
- Method "B" in power2stage.fC() and power.2stage() is redefined to deal
with unequal alpha[1] and alpha[2] in the same manner as method "E" in
Xu et al.This gives small differences compared to previous calculations
with unequal alphas, f.i. Haybittle-Peto alphas
- The CI futility criterion is now based on the 90% CI according to Xu et al.
Was formerly the 1-2*alpha[1] CI
- Default futility criterion changed to "CI" with a lower bound = 0.925
- Defunct print argument removed from all power.2stage.xyz() functions
Version 0.4-2
2015-07-13
- Bug removed in power.2stage.ssr() which prevented correct sample-size
re-estimation if power was calculated via pmethod="ls"
- power.2stage.ssr() now has an argument usePE to use the point estimate from
the interim analysis in sample-size re-estimation (makes only sense if
blind=FALSE)
Version 0.4-1
2015-06-19
- Major rewrite of the power.2stage.xyz() functions wich now return an S3 object
of class "pwrtsd". Output is now done more R-like via the S3 print method for
class "pwrtsd". Therefore the print argument is defunc and will be removed
in the next version
The detail argument has now default=FALSE. Thus be patient if you simulate
for alpha with 1 Mio sims
- The power.2stage.xyz() functions (with exception of power.2stage.GS()) now
return a component ntable, a 'table' object for summarizing the discrete
distribution of ntotal. The nhist component was removed
- The default of fClower in case of fCrit="PE" was changed to 0.8 in function
power.2stage.fC()
- The nsims argument, if missing, is now set to 1E6 if you simulate for alpha
(i.e. with theta0 at border or outside acceptance range theta1 ... theta2)
and to 1E5 otherwise
Version 0.3-2
2015-06-10 released to alpha tester(s) only
- The power.2stage.yy() functions (with exception of power.2stage.GS() now
return a component nhist with class "histogram" which can be used with
plot() to visualize the distribution of Ntotal (suggested by H. Schuetz)
- Slight improvements and typo fixes in man pages
- Handling of non-integer df's in power.2stage.p() in case of Welch's test
changed (now no truncation)
- power.2stage.fC() now has also power calculation method "shifted" implemented
Version 0.3-1
2015-01-24 (dedicated to my brothers Stefan's 50 birthday)
- Major code revision to reflect the internal changes in upcoming
PowerTOST V1.2-06 'raw' power functions
- Sample-size estimation routine (vectorized!) reworked which gives
a considerable run-time boost
- Internal change of start value of sample size estimation to Zhang's
formula (may change the extremal value of n and its mean if usePE=TRUE)
- Function power.2stage.p() accepts now unbalanced stage 1 and uses the
correct power of Welch's test in the power monitoring steps and in
sample size estimation
- Former function power.2stage.p() now available as power.2stage.pAF()
which does the calculations exactly as described in A.Fuglsang's paper
Version 0.2-2
2014-12-08
- CRAN version of V0.2-1 with two nasty bugs removed.
Version 0.2-1
2014-12-06 (Dedicated to my daughter Antje's 40 birthday)
released to alpha-testers only
- Depreciated function power.2stage.Bf() removed
- Functions power.2stage() and power.2stage.fC() now have an argument
min.n2 to restrict the sample size for stage 2 to a lower limit
- Calculation of % studies in stage 2 re-defined to studies having total n>n1
(was in the past studies which had to be evaluated with alpha2, affects
only 'unsymmetric' alpha settings)
- Function power.2stage.ssr() for (blinded) sample size re-estimation added
(2-stage design without BE decision at interim)
- Some code streamlining
- Some enhancements and unification of output of power functions
Version 0.1-5
2014-10-09
- power.2stage.Bf() reworked into the new function power.2stage.fC() to
include Potvin method "C" and to include an additional futility criterion
for the CI also. power.2stage.Bf() is depreciated and will not be contained
in one of the next versions. Use power.2stage.fC() instead.
Version 0.1-4
2014-07-24
- Bug in power.2stage.GS() removed which prevented the convergence of power
values with increasing number of sims
Version 0.1-3
2014-07-02
- Power calculation method "shifted" added in power.2stage() to the aim
of comparison to Potvin et al.
- Internal change of start value of sample size search to avoid
failed searches if variability is high and theta0 is near 1.
Version 0.1-2
not released to the public
- Special version to alpha testers
Version 0.1-1
2014-05-08 released via CRAN
2014-04-24 release to alpha testers
- Contains now the function power.2stage.p() for calculation of power in
BE studies with sequential (2-stage) designs with 2 parallel groups.
- Some internal code streamlinig.
Version 0.0-8
2014-04-11
- Bug in power.2stage.KM() removed if using method="B" (Karalis TSD-2).
Thanks again and again to Helmut Schuetz!
- Man page of power.2stage.KM() changed to the correct definition of
the Karalis/Macheras TSD or Karalis TSD-1, TSD-2 respectively.
Version 0.0-7
2014-02-13
- Maintenance release to reflect the changes in PowerTOST V1.1-10.
- Contains now the function power.2stage.GS() for non-adaptive group
sequential (2-stage) BE studies.
Version 0.0-6
2014-01-02
- Examples adapted to complain with CRAN policy "Examples should run for
no more than a few seconds each". Few seconds means below 5 sec as I learned.
Attention! Number of sims are too low to get meaningful results.
Minimum number of sims should be 1E5 for 'power', 1E6 sims for 'alpha'.
Version 0.0-5
2013-12-27
- First attempt to release via CRAN
Version 0.0-4
2013-09-12 not released, internal version only
- Function power2.2stage() renamed to power.2stage.KM()
- Arguments in power.2stage.KM() omitted which are not necessary for
the TSDs described in the Karalis & Macheras and in Karalis papers
(see references).
- Default method in power.2stage.KM() shifted to "C".
- Function power.2stage.Bf() implemented which evaluates a 2-stage design
derived from Potvin method 'B' with a futility criterion for the PE
of stage 1.
Version 0.0-3
2013-09-03 released to alpha testers
- function power2.2stage() implemented which uses PE and MSE of stage 1
also for the power calculation steps if usePE=TRUE
Version 0.0-2
2013-07-10 released to alpha testers
- Default for npct argument changed to c(0.05, 0.5, 0.95) in accordance
with the Potvin papers.
- Cases with n(total)>Nmax now counted at stage 1 since I feel this is more
logical.
- More checks of input added (THX to Helmut Schuetz).
- Bug in output removed claiming that sample size is estimated with PE
& mse from stage 1 even if argument usePE=FALSE (THX to Helmut).
- Internally stage vector introduced and output based on that.
version 0.0-1
2013-07-09
- First release to alpha testers