README.md

EffToxDesign: EffTox Phase I/II Clinical Trial Design

R package implementation of an adaptive Bayesian design for dose-finding in combined phase I/II clinical trials based on trade-offs between the probabilities of treatment efficacy and toxicity, as developed by Thall and Cook (2004, Biometrics 60, 684-693).

Requirements

Use of the package requires that a C++ compiler be installed, such as the one provided by the Rtools software collection available for Windows.

Installation

The package can be installed directly from GitHub. First, install the devtools package in R if not done so already.

install.packages("devtools")

Then, execute the following command.

devtools::install_github("brian-j-smith/EffToxDesign")

Usage

library(EffToxDesign)

Example

Trial Design Parameters

An EffToxDesign R6 class and method functions are provided for EffTrox trial design. Designs are specified with the new method function.

doses <- c(1, 2, 4, 6.6, 10)
design <- EffToxDesign$new(
  # Treatment doses
  doses = doses,

  # Acceptable efficacy and toxicity parameters
  piE = 0.5,
  pEL = 0.1,
  piT = 0.3,
  pTL = 0.1,

  # Points defining the utility function
  pi1E = 0.5,
  pi2T = 0.65,
  pi3E = 0.7,
  pi3T = 0.25,
  
  # Logit model prior hyperparameters
  thetaE_mean = efftox_theta(c(0.2, 0.4, 0.6, 0.8, 0.90), doses, 2),
  thetaE_sd = c(2.5423, 2.4406, 0.2),
  thetaT_mean = efftox_theta(c(0.02, 0.04, 0.06, 0.08, 0.10), doses),
  thetaT_sd = c(3.5487, 3.5018),
  psi_mean = 0,
  psi_sd = 1,
  
  # Starting dose
  starting_dose = doses[2],
  
  # Cohort sizes
  cohort_sizes = rep(3, 13)
)

Once created, characteristics of the design can be obtained, including effective samples sizes of the prior specifications and utility contours.

design$ESS()

      eff       tox 
0.8305102 0.6750093 

design$contour()

Simulation Studies

Trials can be simulated for a design in order to estimate its operating characteristics. With use of the doParallel package, simulations can be run in parallel to shorten runtimes.

library(doParallel)
registerDoParallel(cores = 6)

sim <- design$simulate(n = 10,
                       eff = c(0.20, 0.40, 0.60, 0.80, 0.90),
                       tox = c(0.05, 0.10, 0.15, 0.20, 0.40),
                       seed = 123)
sim$summary()

                 1         2          4       6.6        10  all
Pr(eff)  0.2000000 0.4000000  0.6000000 0.8000000 0.9000000   NA
Pr(tox)  0.0500000 0.1000000  0.1500000 0.2000000 0.4000000   NA
selected 0.0000000 0.1000000  0.4000000 0.2000000 0.3000000  1.0
given    0.1230769 0.1692308  0.4076923 0.1538462 0.1461538  1.0
E(n_eff) 1.3000000 2.6000000  9.5000000 4.9000000 5.1000000 23.4
E(n_tox) 0.1000000 0.8000000  2.9000000 1.3000000 1.5000000  6.6
E(n)     4.8000000 6.6000000 15.9000000 6.0000000 5.7000000 39.0

sim$barplot()
sim$barplot(stat = c("E(n_tox)", "E(n)"))

Dose-Transition Pathways

Dose-transition pathways show all possible outcomes (E = efficacy only, T = toxicity only, B = both, N = neither) and the doses that would result from n = 1 or more subsequent cohorts.

design$dtp(n = 1, seed = 123)

   cohort1 cohort2
1     2BBB       1
2     2BBE       1
3     2BBN       1
4     2BBT       1
5     2BEE       1
6     2BEN       1
7     2BET       1
8     2BNN       1
9     2BNT       1
10    2BTT       1
11    2EEE       2
12    2EEN       3
13    2EET       1
14    2ENN       3
15    2ENT       1
16    2ETT       1
17    2NNN       3
18    2NNT       3
19    2NTT      NA
20    2TTT      NA

Trial Conduct

Efficacy and toxicity events can be added to the design as they are observed, and dose selection performed on the events.

design$add(yE = c(0, 1, 0), yT = c(0, 0, 0), doses = c(2, 2, 2))
selected <- design$select(seed = 123)
selected$dose
selected$plot()
selected$superiority()

Events can be removed from the design altogether with the reset method or selectively with the drop and keep methods (see documenation).

design$reset()