/gs-design

Group Sequential Design Workflow

Overview

Design and simulate group sequential trials with interim analyses for early stopping due to efficacy or futility.

Workflow Phases

Phase 1: Design Specification

subagent_type: gs-design-specialist prompt: | Specify group sequential design: 1. Number of analyses (interim + final) 2. Information fractions (timing of analyses) 3. Alpha spending function (OBF, Pocock, HSD) 4. Futility stopping rules (binding vs non-binding) 5. Integration with gsDesign2 (if applicable) 6. Stopping boundaries derivation

Phase 2: Simulation Setup

subagent_type: tte-specialist prompt: | Configure GS simulation with simtrial: 1. Define enrollment rates 2. Define failure rates (under H0 and H1) 3. Create cutting functions with create_cut() 4. Configure sim_gs_n() parameters 5. Set up test functions (create_test()) 6. Link to original_design for bound updates

Phase 3: Boundary Calculation

subagent_type: gs-design-specialist prompt: | Calculate and validate boundaries: 1. Compute efficacy boundaries using spending function 2. Compute futility boundaries (if applicable) 3. Verify cumulative alpha spending ≤ alpha 4. Check information fraction assumptions 5. Document boundary derivation

Phase 4: Operating Characteristics

subagent_type: power-optimizer prompt: | Evaluate GS design performance: 1. Simulate under H1 (treatment effect present): - Overall power - Stopping probabilities at each analysis - Expected sample size/events 2. Simulate under H0 (no effect): - Type I error at each analysis - Overall FWER - Expected sample size under null 3. Compare to fixed design: - Expected sample size savings - Power trade-off

Success Criteria

• Boundaries derived and documented
• Type I error controlled at alpha
• Power meets target under alternative
• Operating characteristics fully characterized
• Design integrates with gsDesign2

Final Deliverables

1. Design specification with boundaries
2. Boundary table (analysis, efficacy Z, futility Z)
3. Operating characteristics summary
4. sim_gs_n() code for replication
5. Comparison to fixed design

Configuration Options

• n_analyses: Number of analyses (default: 3)
• info_frac: Information fractions (default: evenly spaced)
• alpha_spending: "OBF", "Pocock", or list(sf=, param=)
• futility: TRUE/FALSE (default: TRUE)
• futility_binding: TRUE/FALSE (default: FALSE)
• n_sim_h1: Simulations under H1 (default: 10000)
• n_sim_h0: Simulations under H0 (default: 100000)