Fast and efficient simulation tools for Bayesian Optimal Interval (BOIN) designs in Phase I clinical trials.

simFastBOIN provides comprehensive functions for simulating Phase I dose-finding trials using the BOIN design methodology. The package enables researchers to evaluate operating characteristics and design performance across different dose-toxicity scenarios, essential for protocol development and regulatory submissions.

Key Features:

• High Performance: Vectorized implementation and optimized algorithms
• BOIN Compatible: Results match BOIN package within <0.5% across all scenarios
• User-Friendly: Simplified API with automatic decision table generation
• Flexible: Multiple safety stopping rules and customizable design parameters
• Multi-Scenario Support: Evaluate multiple dose-toxicity scenarios simultaneously
• Conservative MTD Selection: Optional boundMTD constraint for enhanced safety
• Publication-Ready Output: Professional table formatting with HTML, LaTeX, and Markdown options

You can install the development version from GitHub:

# Install if not already installed
# install.packages("devtools")
devtools::install_github("gosukehommaEX/simFastBOIN")

library(simFastBOIN)

# Define design parameters
target <- 0.30  # Target DLT rate (30%)
p_true <- seq(0.05, 0.45, by = 0.05)  # True toxicity probabilities

# Run simulation (runs silently by default)
result <- sim_boin(
  n_trials = 10000,
  target = target,
  p_true = p_true,
  n_cohort = 20,
  cohort_size = 3,
  seed = 123
)

# Display results
print(result$summary)

# To see progress messages, use verbose = TRUE
result_verbose <- sim_boin(
  n_trials = 10000,
  target = target,
  p_true = p_true,
  n_cohort = 20,
  cohort_size = 3,
  verbose = TRUE,  # Show progress
  seed = 123
)

# Display with percentages instead of absolute numbers
print(result$summary, percent = TRUE)

# Display in Markdown pipe format for R Markdown documents
print(result$summary, kable = TRUE, kable_format = "pipe")

# LaTeX format for publication
print(result$summary, kable = TRUE, kable_format = "latex", scenario_name = "My Scenario")

# HTML format for web display
print(result$summary, kable = TRUE, kable_format = "html")

# Define multiple dose-toxicity scenarios
scenarios <- list(
  list(name = "Scenario 1: Linear Increase",
       p_true = c(0.05, 0.10, 0.15, 0.20, 0.25, 0.30, 0.35, 0.40, 0.45)),
  list(name = "Scenario 2: Steep Early",
       p_true = c(0.01, 0.025, 0.05, 0.075, 0.10, 0.125, 0.15, 0.20, 0.30)),
  list(name = "Scenario 3: Plateau",
       p_true = c(0.15, 0.20, 0.25, 0.30, 0.45, 0.50, 0.50, 0.50, 0.50))
)

# Run simulations across all scenarios
result <- sim_boin_multi(
  scenarios = scenarios,
  target = 0.30,
  n_trials = 10000,
  n_cohort = 48,
  cohort_size = 3,
  seed = 123
)

# View aggregated results
print(result)

# View as HTML table
print(result, kable = TRUE, kable_format = "html")

# Access scenario-specific results
result$results_by_scenario[["Scenario 1: Linear Increase"]]

• sim_boin(): Run BOIN trial simulations with automatic decision table generation

  • Automatically generates BOIN boundaries and decision tables
  • Optional safety stopping rules (extrasafe, boundMTD)
  • Flexible stopping criteria (n_earlystop_rule)
  • Returns detailed trial-level results and summary statistics

• sim_boin_multi(): Run simulations across multiple scenarios

  • Orchestrates sim_boin() for each scenario automatically
  • Aggregates results into a unified comparison table
  • Ideal for protocol development evaluating multiple dose-toxicity relationships

• get_boin_boundary(): Calculate BOIN interval boundaries (lambda_e, lambda_d)
• get_boin_decision(): Generate decision table for dose escalation/de-escalation
• get_boin_stopping_boundaries(): Generate safety stopping boundaries

• isotonic_regression(): Apply isotonic regression to estimate dose-toxicity curve
• select_mtd(): Select MTD with optional boundMTD constraint

• print.boin_summary(): Print formatted summary tables with flexible options
• print.boin_multi_summary(): Print multi-scenario summary tables

simFastBOIN uses vectorized operations and efficient algorithms for superior performance:

Benchmark (5 doses, 20 cohorts, cohort_size=3, all safety features enabled):

Results validated against the BOIN package using comprehensive test scenarios. All operating characteristics (MTD selection rates, sample sizes, toxicity counts) matched within <0.5%.

simFastBOIN produces results identical to the original BOIN package while offering significantly improved performance. This makes it ideal as a drop-in replacement for large-scale simulations where speed is critical.

Comprehensive validation was conducted comparing simFastBOIN with BOIN::get.oc() across multiple configurations:

Test Design:

• Test Scenarios: 5 dose-toxicity scenarios (5 doses each)

  • scenario1: c(0.05, 0.10, 0.20, 0.30, 0.45) - MTD at dose 4
  • scenario2: c(0.30, 0.40, 0.50, 0.60, 0.70) - All doses ≥ target
  • scenario3: c(0.05, 0.10, 0.15, 0.20, 0.25) - All doses < target
  • scenario4: c(0.15, 0.30, 0.45, 0.60, 0.75) - MTD at dose 2
  • scenario5: c(0.01, 0.03, 0.08, 0.15, 0.30) - MTD at dose 5

• Parameter Combinations: 8 configurations per scenario

  • extrasafe = TRUE/FALSE
  • boundMTD = TRUE/FALSE
  • titration = TRUE/FALSE

• Total Configurations: 40 (5 scenarios × 8 parameter combinations)

• Trials per Configuration: 10,000

Validation Results:

Conclusion: All 40 configurations showed perfect metric match (all_match = TRUE) with simFastBOIN and BOIN package results. Maximum differences across all metrics were well within acceptable tolerance levels, confirming full compatibility.

• Results validated across all scenario types (favorable, unfavorable, balanced)
• Testing included all major design options:
  • Safety stopping rules (extrasafe TRUE/FALSE)
  • MTD selection constraints (boundMTD TRUE/FALSE)
  • Titration phase options (titration TRUE/FALSE)
• Same random number generation (when seed is fixed)
• Full compatibility with BOIN methodology

The boundMTD option provides an additional safety constraint during MTD selection:

# Without boundMTD: may select doses near toxicity boundary
result1 <- sim_boin(
  n_trials = 1000,
  target = 0.30,
  p_true = c(0.10, 0.20, 0.28, 0.36, 0.50, 0.65),
  n_cohort = 20,
  cohort_size = 3,
  boundMTD = FALSE,
  seed = 123
)

# With boundMTD: more conservative selection
result2 <- sim_boin(
  n_trials = 1000,
  target = 0.30,
  p_true = c(0.10, 0.20, 0.28, 0.36, 0.50, 0.65),
  n_cohort = 20,
  cohort_size = 3,
  boundMTD = TRUE,
  seed = 123
)

Selected MTD must have isotonic-estimated toxicity rate below the de-escalation boundary (lambda_d), preventing selection of doses too close to overly toxic doses.

Two stopping rules are available:

• Stop when sample size at current dose reaches n_earlystop
• Faster trials, fewer patients

• Stop when sample size reaches n_earlystop AND next decision is "Stay"
• Ensures algorithm convergence before stopping
• More thorough dose evaluation

# Simple rule: Stop immediately at n_earlystop
result_simple <- sim_boin(
  n_trials = 1000,
  target = 0.30,
  p_true = c(0.05, 0.10, 0.20, 0.30, 0.45, 0.60),
  n_cohort = 20,
  cohort_size = 3,
  n_earlystop_rule = "simple",
  seed = 123
)

# With stay: Wait for convergence
result_with_stay <- sim_boin(
  n_trials = 1000,
  target = 0.30,
  p_true = c(0.05, 0.10, 0.20, 0.30, 0.45, 0.60),
  n_cohort = 20,
  cohort_size = 3,
  n_earlystop_rule = "with_stay",
  seed = 123
)

Additional safety rule to stop the entire trial if the lowest dose is overly toxic:

result <- sim_boin(
  n_trials = 10000,
  target = 0.30,
  p_true = seq(0.05, 0.45, by = 0.05),
  n_cohort = 48,
  cohort_size = 3,
  extrasafe = TRUE,   # Enable safety stopping
  offset = 0.05,      # Cutoff = cutoff_eli - offset
  seed = 123
)

Multiple output format options for different use cases:

# Plain text (default)
print(result$summary)

# With percentages
print(result$summary, percent = TRUE)

# Markdown pipe format (for RMarkdown)
print(result$summary, kable = TRUE, kable_format = "pipe")

# LaTeX format (for publications)
print(result$summary, kable = TRUE, kable_format = "latex")

# HTML format (for web display)
print(result$summary, kable = TRUE, kable_format = "html")

# Simple format (minimal formatting)
print(result$summary, kable = TRUE, kable_format = "simple")

# Combine options
print(result$summary, scenario_name = "My Analysis", percent = TRUE, kable = TRUE)

Recommended settings for BOIN standard compliance:

result <- sim_boin(
  n_trials = 10000,
  target = 0.30,
  p_true = c(0.10, 0.25, 0.40, 0.55, 0.70),
  n_cohort = 10,
  cohort_size = 3,
  boundMTD = TRUE,              # Conservative MTD selection
  n_earlystop_rule = "with_stay",  # Stop when converged
  extrasafe = TRUE,             # Safety monitoring
  seed = 123
)

print(result$summary, scenario_name = "BOIN Standard")

library(simFastBOIN)

# Step 1: Define scenario
target <- 0.30
p_true <- c(0.05, 0.10, 0.15, 0.20, 0.25, 0.30, 0.35, 0.40, 0.45)

# Step 2: Run simulation
result <- sim_boin(
  n_trials = 10000,
  target = target,
  p_true = p_true,
  n_cohort = 48,
  cohort_size = 3,
  boundMTD = TRUE,
  n_earlystop_rule = "with_stay",
  extrasafe = TRUE,
  return_details = TRUE,  # Enable detailed trial-level results
  seed = 123
)

# Step 3: Display results
print(result$summary, scenario_name = "Primary Analysis")

# Step 4: Export for RMarkdown
print(result$summary, kable = TRUE, kable_format = "pipe")

# Step 5: Check stopping reasons
table(sapply(result$detailed_results, function(x) x$reason))

The package tracks why each trial terminated:

• "trial_completed": Normal completion
• "n_earlystop_reached": Reached sample size limit (simple rule)
• "n_earlystop_with_stay": Converged at sample size limit (with_stay rule)
• "lowest_dose_too_toxic": Safety stopping at lowest dose (extrasafe)
• "lowest_dose_eliminated": Lowest dose eliminated during trial
• "no_dose_below_lambda_d": No dose satisfies boundMTD constraint
• "max_cohorts_reached": Maximum number of cohorts completed

Liu, S. and Yuan, Y. (2015). Bayesian Optimal Interval Designs for Phase I Clinical Trials. Journal of the Royal Statistical Society: Series C, 64, 507–523.

See NEWS.md for detailed version history and changelog.

MIT © Gosuke Homma