Non-Proportional Hazards Sample Size Simulation Engine

Role: Principal Biostatistician and Clinical Trial Methodologist

Context

• Target Endpoint: {{endpoint_name}}
• Hypothesized Treatment Effect Dynamic: {{hazard_ratio_profile}}
• Accrual Pattern: {{recruitment_timeline_months}}
• Censoring Assumptions: {{dropout_rate_annual}}
• Power Specification: {{target_statistical_power}}
• Interim Monitoring: {{interim_analysis_schedule}}

Task

Synthesize these trial parameters into an end-to-end, reproducible R script that conducts Monte Carlo trial simulations to determine the required sample size, target event count, and empirical power under non-proportional hazard assumptions using weighted log-rank and standard log-rank estimators.

Method

1. Initialize global simulation parameters, random seeds, iteration count (minimum 2,500 runs), and parallel processing backend (future.apply or parallel).
2. Program a piecewise exponential data-generating function reflecting the {{hazard_ratio_profile}} and non-uniform patient entry matching {{recruitment_timeline_months}}.
3. Integrate right-censoring mechanisms based on both administrative study cutoff dates and uninformative dropout from {{dropout_rate_annual}}.
4. Embed an alpha-spending boundary calculation using the specified {{interim_analysis_schedule}} to track cumulative Type I error.
5. Execute test statistics across each simulated cohort, computing standard log-rank, MaxCombo, and restricted mean survival time (RMST) tests.
6. Track event counts at each interim milestone to evaluate early stopping criteria for efficacy and futility.
7. Aggregate iteration metrics to solve for sample size achieving {{target_statistical_power}} across varying cohort sizes.
8. Generate summary data structures and output publication-grade simulation diagnostic plots using ggplot2.

Constraints

• Code MUST be fully self-contained R script with explicit library loading and zero unresolved object references.
• Script MUST include robust error handling for non-convergent iterations or zero-event subgroups.
• MUST NOT use deprecated R functions or undocumented analytical approximations.
• All generated datasets within the simulation loop MUST be dynamically memory-managed to avoid memory overflow.
• Variable definitions and parameter blocks must be placed at the top of the script for parameter tuning.

Output format

Provide the response in three distinct sequential sections:

1. Execution Architecture: A concise technical summary (150-250 words) detailing the simulation framework and convergence logic.
2. Executable R Script: Fully commented, vectorized R code structured into Config, Data Generation, Simulation Loop, Analysis/Testing, and Summary Reporting.
3. Simulation Diagnostics Table: Formatted markdown table defining the required sample sizes, expected event milestones, and comparative power across tested statistics.

Self-review

• Verify that the data generation function accurately reflects {{hazard_ratio_profile}} over the specified timeline.
• Check that interim alpha-spending calculations in the script match {{interim_analysis_schedule}}.
• Ensure all packages utilized (survival, simsurv, ggplot2) are loaded with package verification checks.