simulation.rgs: Streamlined Trial Data Simulations and Analysis Using...
In phe9480/rgs: What the Package Does (One Line, Title Case)

Description Usage Arguments Value Examples

Simulate Randomized two-arm trial data with the following characteristics: (1) randomization time (entry time) is generated according to the specified non-uniform accrual pattern, i.e. the cumulative recruitment at calendar time t is (t/A)^w with weight w and enrollment complete in A months. w = 1 means uniform enrollment, which is usually not realistic due to graduate sites activation process. (2) Survival time follows piece-wise exponential distribution for each arm. (3) N total patients with r:1 randomization ratio (4) Random drop off can be incorporated into the censoring process. (5) Data cutoff dates are determined by specified vector of target events for all analyses. (6) A dataset is generated for each analysis according to the specified number of target events. Multiple analyses can be specified according to the vector of targetEvents, eg, targetEvents = c(100, 200, 300) defines 3 analyses at 100, 200, and 300 events separately. (7) Weighted log-rank test is then performed for each simulated group sequential dataset.

simulation.rgs(
  nSim = 10000,
  N = 672,
  A = 21,
  w = 1.5,
  r = 1,
  lambda0 = log(2)/12,
  lambda1 = log(2)/12 * 0.65,
  cuts = NULL,
  dropOff0 = 0,
  dropOff1 = 0,
  targetEvents = NULL,
  DCO = NULL,
  sf = "LDOF",
  overall.alpha = 0.025,
  type1err = NULL,
  logrank = "N",
  fws.options = NULL,
  H0 = "N",
  Cox = "N",
  Median = "N",
  scanfreq0 = NULL,
  scanfreq1 = NULL,
  seed = 2022
)

`nSim`	Number of trials
`N`	Total number patients in two arms.
`A`	Total accrual period in months
`w`	Weight parameter in cumulative enrollment pattern. The cumulative enrollment at month t is (t / A)^w, eg, at month 6, the enrollment is N*(6/24)^2 = N/16 for 24 months planned accrual period.
`r`	Randomization ratio r:1, where r refers to the experimental arm, eg, r=2 in 2:1 ratio
`lambda0`	Hazard rates for control arm of intervals defined by cuts; for exponential(lambda0) distribution, lambda0 = log(2) / median;
`lambda1`	Hazard rates for experimental arm for intervals; for exponential(lambda1) distribution, lambda1 = log(2) / median; For delayed effect under H1, lambda1 is a vector (below).
`cuts`	Timepoints to form intervals for piecewise exponential distribution. For example, experimental arm has hr = 0.6 after delay, then cuts = 6, and lamda0 = log(2) / m0 or lambda0 = rep(log(2) / m0, 2), lamda1 = c(log(2)/m0, log(2)/m0hr). treatment after 24 mo, so its hazard decreases 20%. Then, lambda0 = c(log(2)/m0, log(2)/m0, log(2)/m00.8), lambda1 = c(log(2)/m0, log(2)/m0hr, log(2)/m0hr), and cuts = c(6, 24), which forms 3 intervals (0, 6), (6, 24), (24, infinity)
`dropOff0`	Drop Off rate per month, eg, 1%, for control arm
`dropOff1`	Drop Off rate per month, eg, 1%, for experimental arm
`targetEvents`	A vector of target events is used to determine DCOs. For example, 397 target events are used to determine IA DCO; and 496 events are used to determine the FA cutoff.
`DCO`	A vector of data cut-off time in months, calculated from first subject in. Default NULL. The date cut-off times will be determined by targetEvents. If provided, then the `targetEvents` will be ignored.
`overall.alpha`	One-sided overall type I error, default 0.025.
`type1err`	Incremental type I error. sum(type1err) = overall.alpha. When targetEvents is NULL, type1err is required.
`logrank`	Indicator whether log-rank test is requested besides the weighted logrank tests. "Y" or "N". Default "Y". If "Y", the traditional log-rank test will be used based on survdiff() function. If "N", the weighted log-rank test with weighting function specified in fws will be used.
`H0`	"Y" or "N" to indicate whether the simulation is for type I error
`Cox`	"Y" or "N" to indicate whether Cox regression model is requested to produce HR estimate
`Median`	"Y" or "N" to indicate whether median is requested to estimate.
`scanfreq0`	A vector of scan frequency for control arm, eg, every 6 weeks for 2 scans, then every 9 weeks thereafter. scanfreq = c(rep(6, 2), rep(9, 1000))*7. Applicable for PFS data.
`scanfreq1`	A vector of scan frequency for experimental arm, eg, every 6 weeks for 2 scans, then every 9 weeks thereafter. scanfreq = c(rep(6, 2), rep(9, 1000))*7. Applicable for PFS data.

An object with a dataframe for each analysis including the following variables:

power

Power for each analysis

overall.power

Overall power of the group sequential design

wlr.simulations

Simulation results for each simulated study data. An array with dimensions: nSim simulations X M testing strategies (fws.options) X K analyses X 5 variables:

z value
p value
analysis
rejection boundary
testing result (1 = positive; 0 = negative)

lr.power

Power for each analysis using log-rank test. Available if logrank ="Y"

lr.overall.power

Overall power of the group sequential design using logrank test

lr.simulations

Simulation results for each simulated study data using logrank test. nSim simulations X K analyses X 5 variables as above

CutOff

A matrix with dimension (nSim, K). Date cutoff date for each analysis in each simulated trial.

lr = function(s){1}
fh01 = function(s){1-s}
fh11 = function(s){s*(1-s)}
sfh01 = function(s){s1 = apply(cbind(s, 0.5), MARGIN=1,FUN=max); return(1-s1)}

#Example (1): Simulate 10 samples from proportional hazards scenario. 
fws1 = list(IA1 = list(lr), FA = list(lr))
fws2 = list(IA1 = list(lr), FA = list(fh01))
fws3 = list(IA1 = list(fh01), FA = list(fh01))
fws4 = list(IA1 = list(lr), FA = list(lr, fh01))
fws5 = list(IA1 = list(lr), FA = list(lr, fh01, fh11))
fws6 = list(IA1 = list(lr, fh01), FA = list(lr, fh01))
fws7 = list(IA1 = list(sfh01), FA = list(lr, fh01))
#' alpha = f.alpha(overall.alpha=0.025, side = 1, sf="LDOF", targetEvents=c(60, 80))
m0 = 11.7
lambda0 = log(2) / m0
h0 = function(t){lambda0}; 
S0 = function(t){exp(-lambda0 * t)}
h1 = function(t){lambda0*0.745}; 
S1 = function(t){exp(-lambda0 *0.65* t)}

F.entry = function(t){(t/21)^1.5*as.numeric(t <= 21) + as.numeric(t > 21)}
G.ltfu = function(t){0}
f.logHR = function(t){log(0.745)}

#7 weighting strategies for exploration 
fws = list(fws1, fws2, fws3, fws4, fws5, fws6, fws7)

#Simulations for exploring 7 weighting strategies
sim = simulation.rgs(nSim=2, N = 100, A = 21, w=1.5, r=1, 
lambda0=log(2)/12, lambda1=log(2)/12*0.65,
cuts=NULL, targetEvents = c(60, 80), 
sf = "LDOF", overall.alpha = 0.025, type1err = NULL,
logrank="Y", fws.options=NULL, Cox="Y", Median="Y")