SKBD: Shared Keyboard Designs for Phase I Dose-Finding Trials

SKBD implements the Shared Keyboard Design (SKBD) for model-assisted phase I dose-finding trials.

The package provides practical tools for:

• generating SKBD decision tables,
• evaluating trial operating characteristics through simulation,
• extending SKBD to time-to-event settings with delayed toxicity outcomes, and
• assessing adaptive dose insertion under a shared-borrowing framework.

SKBD is designed for method development, simulation-based validation, and practical trial planning in early-phase oncology studies.

Installation

You can install the stable version of SKBD from CRAN with:

install.packages("SKBD")

Or install the development version from GitHub with:

# install.packages("pak")
pak::pak("Jiangyan-Zhao/SKBD")

Then load the package:

library(SKBD)

Main functions

The current public interface includes five core functions:

• get_boundary_SKBD() generates pre-tabulated decision boundaries for the shared keyboard design.
• get_OC_SKBD() simulates operating characteristics for the standard SKBD.
• get_OC_TITE_SKBD() simulates operating characteristics for time-to-event SKBD.
• get_OC_Insert_SKBD() simulates operating characteristics when adaptive dose insertion is enabled.
• PUA() generates monotone dose-toxicity scenarios using a pseudo-uniform algorithm.

Shiny app

A Shiny app is included for interactive use of SKBD decision tables and simulation outputs.

What you can do in the app

• Configure trial design settings, including dose levels, target toxicity probability, borrowing strength, sample size, and overdose-control parameters.
• Input current trial observations and generate SKBD decision tables interactively.
• Run multi-scenario operating-characteristic simulations using manual input or uploaded CSV templates.

Launch the Shiny app

After installing SKBD, load the package and launch the app by running:

library(SKBD)

run_SKBD_shiny()

By default, run_SKBD_shiny() follows the standard launching behavior of shiny::runApp(). In RStudio, the app is typically opened in the RStudio Viewer or Shiny window.

To open the app in the system default browser, use:

run_SKBD_shiny(launch.browser = TRUE)

To start the app without automatically opening a browser, use:

run_SKBD_shiny(launch.browser = FALSE)

The app provides two main modules:

• Trial Setting: interactively generates dose-escalation and de-escalation boundary tables.
• Simulation: runs batch simulations under user-specified scenarios and summarizes the operating characteristics.

Decision tables for SKBD

A typical starting point is to generate a decision table at the current dose level, conditional on the observed data across all dose levels.

y <- c(0, 1, 2, 2, 0)
n <- c(3, 6, 9, 3, 0)

out_boundary <- get_boundary_SKBD(
  target_prob = 0.30,
  d = 3,
  y = y,
  n = n,
  table_type = "continue"
)

out_boundary$boundary_tab
#>                                                         
#> Number of patients treated  3  6  9 12 15 18 21 24 27 30
#> Escalate if # of DLT <=    NA NA NA  2  2  3  4  4  5  6
#> de-escalate if # of DLT >= NA NA NA  4  5  6  7  8  9 10
#> Eliminate if # of DLT >=   NA NA NA NA NA 10 11 13 14 15

The output is a keyboard-style table with escalation, de-escalation, and elimination boundaries based on the SKBD pseudo-posterior at the current dose.

Simulating operating characteristics under SKBD

The standard SKBD can be evaluated under a prespecified toxicity scenario as follows:

out_skbd <- get_OC_SKBD(
  target_prob = 0.30,
  tox_prob = c(0.05, 0.12, 0.30, 0.45, 0.60),
  n_cohort = 10,
  cohort_size = 3,
  n_trial = 1000
)

out_skbd$PCS
#> [1] 61.8
out_skbd$PCA
#> [1] 37.49
out_skbd$ROD60
#> [1] 1.8

The returned object includes accuracy and safety summaries, such as the percentage of correct selection (PCS), percentage of correct allocation (PCA), and overdose risk.

Time-to-event SKBD

Delayed toxicity outcomes can be handled using time-to-event SKBD:

out_tite <- get_OC_TITE_SKBD(
  target_prob = 0.20,
  tox_prob = c(0.05, 0.12, 0.20, 0.35, 0.50),
  n_cohort = 10,
  cohort_size = 3,
  tau = 3,
  accrual = 2,
  dist_DLT = "weibull",
  dist_enter = "exp",
  n_trial = 1000
)

out_tite$PCS
#> [1] 44.3
out_tite$duration_mean
#> [1] 23.56665

This extension accounts for pending toxicity outcomes by incorporating weighted follow-up information within the DLT assessment window.

Adaptive dose insertion

SKBD also supports simulations for insertion-enabled shared keyboard designs:

out_insert <- get_OC_Insert_SKBD(
  target_prob = 0.30,
  tox_prob = c(0.14, 0.45, 0.63, 0.74, 0.80),
  dose_set = c(5, 15, 25, 35, 45),
  n_trial = 1000
)

out_insert$insertion
#> $sel_pct
#> [1] 80.7
#> 
#> $pts_pct
#> [1] 49.38
#> 
#> $dose_mean
#> [1] 10.64145
#> 
#> $dose_sd
#> [1] 3.651881
#> 
#> $trial_pct
#> [1] 95.3
#> 
#> $cohort_mean
#> [1] 3.373277
#> 
#> $n_median
#> [1] 3

The insertion summary reports how often new doses are inserted, where inserted doses tend to be located, how often they are selected as the final MTD, and how patients are allocated to inserted doses.

Random monotone scenarios

For random-scenario simulation studies, PUA() can be used to generate monotone dose-toxicity curves with a well-defined target dose.

scen <- PUA(
  dose_set = 1:5,
  target_prob = 0.30,
  n_scenarios = 5
)

scen
#>            [,1]       [,2]      [,3]      [,4]      [,5]
#> [1,] 0.01944263 0.02871687 0.1536818 0.1940659 0.2761874
#> [2,] 0.04828428 0.25536821 0.3181795 0.5105885 0.5796248
#> [3,] 0.19502259 0.27180464 0.5551730 0.6691403 0.7765835
#> [4,] 0.17182210 0.33929376 0.4044292 0.6745865 0.8366659
#> [5,] 0.09726789 0.10416956 0.1283349 0.1508885 0.2799666

Development notes

SKBD is under active development. The current version focuses on decision-table utilities, simulation engines, time-to-event extensions, and adaptive dose insertion for the SKBD framework. Additional examples, validation materials, and extended tutorials will be added as the package evolves.