keaven/simtrial

Clinical Trial Simulation with simtrial

Note: This skill targets simtrial >= 1.0.2 (main branch at github.com/Merck/simtrial).

API reference

• Full function docs (CRAN): references/llms.txt (source: https://gsDesign.ai)
• Full function docs (local v1.0.2): references/llms_local.txt
• Workflow patterns: references/code_patterns.md

Key functions

Simulation

• sim_pw_surv() - Simulate piecewise exponential survival data (individual patient data)
• sim_fixed_n() - Fixed-sample simulation with analysis pipeline (simpler, less flexible)
• sim_gs_n() - Group sequential simulation (integrates with gsDesign2 designs)
• to_sim_pw_surv() - Convert simple rate format (control rate + HR) to sim_pw_surv format

Data cutting

• cut_data_by_date() - Cut simulated data at a calendar date
• cut_data_by_event() - Cut simulated data at a target event count
• get_cut_date_by_event() - Find calendar date for a target event count
• get_analysis_date() - Get analysis date from multiple criteria (events, calendar time, follow-up)
• create_cut() - Create a cutting function for use in sim_gs_n pipelines

Statistical tests

• wlr() - Weighted logrank test (single weight function)
• maxcombo() - MaxCombo test (multiple FH weight functions, correlation-adjusted p-value)
• rmst() / rmst_two_arm() / rmst_single_arm() - Restricted mean survival time
• milestone() - Milestone analysis (survival difference at fixed time)
• multitest() - Apply multiple tests to one dataset
• create_test() - Create a parameterized test function for use in pipelines

Weight functions

• fh() - Fleming-Harrington weights (rho, gamma)
• mb() - Magirr-Burman weights (delay period for NPH)
• early_zero() - Early zero weight function
• wlr_weight() - General WLR weight specification

Utilities

• counting_process() - At-risk/event tables from survival data (for custom analyses)
• rpwexp() - Random piecewise exponential generation
• rpwexp_enroll() - Random piecewise enrollment times
• fit_pwexp() - Fit piecewise exponential model
• randomize_by_fixed_block() - Block randomization

Output

• summary() - Summarize sim_gs_n results (power, events, timing)
• as_gt() - Convert summary to gt table

Example datasets

• ex1_delayed_effect through ex6_crossing - Pre-built NPH scenarios from Cross-Pharma Working Group

Workflow patterns

For detailed code templates, read references/code_patterns.md.

Topics covered:

• Generating survival data with sim_pw_surv() (PH and NPH)
• Data cutting (calendar, event, flexible criteria)
• Weighted logrank tests (wlr()) and MaxCombo (maxcombo())
• Milestone and RMST tests
• Multiple tests with multitest() and create_test()
• Fixed-sample simulation with sim_fixed_n() (timing_type options)
• Group sequential simulation with sim_gs_n() (multiple tests, boundary updating)
• Integration with gsDesign2 designs (events from design, original_design parameter)
• Weight functions (Fleming-Harrington, Magirr-Burman, early zero)
• Rate format conversion with to_sim_pw_surv()
• Flexible analysis timing with get_analysis_date() / create_cut()
• Stratified simulations
• Standalone wlr() with illness-death model ADTTE data
• Example NPH datasets

Important design considerations

• maxcombo must be used alone in sim_gs_n(): it cannot be combined with other tests in the same test list
• to_sim_pw_surv(): Use this to convert the simpler rate format (control rate + HR) to the treatment-specific format needed by sim_pw_surv()
• sim_gs_n() + original_design: Pass a gsDesign2 design object to get boundaries updated based on actual vs planned information fraction
• ia_alpha_spending: Controls how alpha is spent when observed events differ from planned ("min_planned_actual" is conservative default)
• fa_alpha_spending = "full_alpha": Spends full alpha at final analysis (default); use "info_frac" for event underrunning scenarios
• create_cut() and create_test(): These factory functions are essential for building sim_gs_n() pipelines
• Example datasets: Columns are id, month, evntd, trt — need renaming to tte, event, treatment for use with wlr() / maxcombo()
• Standalone wlr(): Can be used outside sim_gs_n() with any data having tte, event, stratum, treatment columns. Returns positive Z when experimental is better. See references/code_patterns.md for illness-death model integration.