Package 'AdaptiveDesignSim'

ADeval

Description

Function to evaluate operating characteristics of trials simulated via ADsim

Usage

ADeval(simdat, zcutoff, finalm = c("z", "ci"), alpha = 0.025)

Arguments

simdat	matrix or data.frame of multiple runs of ADsim
zcutoff	a matrix (or data frame) with the cut-offs on the z-statstics at interim. One row per interim analysis and a column for stop for futilty and efficacy (with NA if not planned)
finalm	Measure used for the final analysis, either z (z-statistics) or ci (confidence interval).
alpha	Alpha level used for the final analysis if finalm=="z", otherwise the ci given in simdat is used.

Value

A list with the evaluation for each repetition (rawdata) and the summarized operating characteristics (opchar). opchar includes the average estimates in each group, effect (point estimate, pe) and (1-alpha) confidence limits (lci, uci), and

• psig: the probability for a significant trial (the power or type I error, depending on the assumptions).

• pstop: the probability for any stop and a stop for futility or efficacy (pstop_fut, pstop_eff).

• the average, minimum and maximum sample size (avn, minn, maxn).

• bias: the average of estimate minus true value.

• sd: the standard deviation of the point estimate.

• mse: the mean squared error, the average of the squared difference between estimate and true value.

• coverage: the proportion of confidence intervals including the true value.

Examples

#Simulate 1000 trials with one interim analysis
simdat <- lapply(1:1000, function(i)
		ADsim(n01 = c(100,100), par01 = c(0.2, 0.4), nia = 1))
simdat <- do.call(rbind, simdat)

# Define stop for futility stoppin (z<0) and efficacy stopping (z>2)
zcutoff<-matrix(c(0,2),1,2)
sr<-ADeval(simdat = simdat, zcutoff = zcutoff)
head(sr$rawdata)
sr$opchar

---

ADsim

Description

Simulates trials with a binary endpoint and one or several interim analysis,

Usage

ADsim(
  n01,
  nia,
  tia = NULL,
  par01,
  simfun = "binom",
  estfun = "rd_wald",
  cilevel = 0.95,
  truefun = NULL,
  direct = c("higher", "lower")
)

Arguments

n01	Total sample size in control and experimental arm, numeric vector of length 2.
nia	Number of interim analysis, not including the final analysis.
tia	Optional vector with the time point(s) of the interim analyses, as fraction of the total (between 0 and 1).
par01	Parameters in control and experimental arm (which is used as argument for simfun).
simfun	Simulation function or one of the default options ("binom", "norm").
estfun	Function to calculate point estimate, CIs and z- statistic or one of the five defaults functions ("rd_wald", "lrr_wald", "rd_score", "lrr_score", "md_t").
cilevel	Level of the two-sided confience interval for the default estfun.
truefun	Function to calculate the true point estimate from par01 (ignored if the default estfun are used).
direct	Direction of the effect, "higher" (default) or "lower" values are better.

Details

Parameters for simulation are defined with par01, the simulation function with simfun. For the default simfun "binom" par01 must be a numeric vector of length 2 with the probabilities in control and experimental group. For the default simfun "norm", par01 must be a 2x2 data frame with rows mean and sd and first column for control and second for experimental. User-specified simfun, must take par01 as argument.

For estimation, five defaults can be specified via estfun - standard Wald-type approximation for the risk difference (wald_rd) and risk ratio (wald_rr), the score-based method for the risk difference suggested by Newcombe (score_rd), the score-based method for the risk ratio according to Koopman (score_rr), and Student's t-test for the mean difference (md_t). The effect is allways given as experimental (second element of par01) vs control (first element of par01).

User-specified estfun must take the output from simfun and cilevel as argument, and produce a named numeric vector with point estimate (pe), confidence limits (lci, uci) and z-statistic (z). If a user-specifid estfun is used, a function to calculate the true effect must be given via truefun.

The direction if the effect in direct only affects the sign of the z-statistic. So that positive values of the z-statistic are always assocaiated with a benefit of the experimental arm.

Value

Numeric vector with true point estimate (true_pe), and for each interim stage (prefix iax_) and the final stage (prefix fa_), the sample size (suffix n0 and n1), point estimate (suffix pe), confidence interval (suffix lci and uci) and z-statistic (z) And depending on the setting further group-specific paremeters (e.g. the simulated number of successes and observed probabilities for both arms).

Examples

set.seed(1)
ADsim(n01 = c(100,100), par01 = c(0.2,0.4), nia = 1)

#Earlier interim analysis
set.seed(1)
ADsim(n01 = c(100,100), par01 = c(0.2,0.4), nia = 1, tia = 0.4)

#Difference effect estimation
set.seed(1)
ADsim(n01 = c(100,100), par01 = c(0.2,0.4), nia = 1, estfun = "lrr_wald")

#Assume higher would be better
set.seed(1)
ADsim(n01 = c(100,100), par01 = c(0.2,0.4), nia = 2, direct="lower")

#User defined function that reproduces the default setting
efun<-function(out,cilevel) {
	xs<-unlist(lapply(out,sum))
	ns<-unlist(lapply(out,length))
	pe<-diff(xs/ns)
	se<-sqrt(sum(xs/ns*(1-xs/ns)/ns))
	lci<-pe - qnorm(1-(1-cilevel)/2)*se
	uci<-pe + qnorm(1-(1-cilevel)/2)*se
	z<-(-pe)/se
	res<-c(pe,lci,uci,z)
	names(res)<-c("pe","lci","uci","z")
	return(res)
}

tfun <- function(x) diff(x)

set.seed(1)
ADsim(n01 = c(100,100), par01 = c(0.4,0.2), nia = 1, estfun = efun, truefun = tfun)

#Continuous outcome 
par01<-data.frame(matrix(c(5,1,5.5,1),2,2))
rownames(par01)<-c("mean","sd")
ADsim(n01 = c(100,100), par01 = par01, nia = 1, estfun = "md_t", simfun="norm")

---

AdaptiveDesignSim shiny app

Description

The shiny app to AdaptiveDesignSim provides an intuitive way to

1. check the results of individual simulations (via forest plots) and

2. of operating characteristics over a range of conditions.

Usage

launch_ADSim_app()

Examples

# launch the app
## Not run: 
launch_ADSim_app()

## End(Not run)

---