Sample Size Determination Workflow

Overview

Find the minimum sample size (or event count) required to achieve target power across specified treatment effect scenarios.

Workflow Phases

Phase 1: Requirements

<Task> subagent_type: simulation-architect prompt: | Gather sample size requirements: 1. Target power level (e.g., 90%) 2. Type I error rate (alpha) 3. Primary endpoint type 4. Treatment effect assumptions (range of plausible effects) 5. Any constraints (max N, budget, timeline) 6. Robustness requirements (power across scenarios) </Task>

Phase 2: Search Strategy

<Task> subagent_type: power-optimizer prompt: | Implement sample size search: 1. Define sample size grid or binary search range 2. For each candidate N: - Run power simulations - Record power at each scenario 3. Find minimum N achieving target power 4. Calculate optimal interval (range of acceptable N) 5. Determine joint optimal region across scenarios

Methods:

• Binary search: Efficient for single scenario
• Grid search: Better for multiple scenarios
• Compound criteria: Min power across scenarios ≥ target </Task>

Phase 3: Validation

<Task> subagent_type: tte-specialist condition: endpoint_type == "time-to-event" prompt: | Validate sample size for TTE endpoint: 1. Confirm power at recommended N using sim_fixed_n() 2. Run validation simulations (larger n_sims) 3. Compare to analytical approximation 4. Check event count assumptions </Task> <Task> subagent_type: cse-specialist condition: endpoint_type != "time-to-event" prompt: | Validate sample size using CSE: 1. Confirm power at recommended N 2. Generate detailed results across scenarios 3. Prepare presentation-ready output </Task>

Success Criteria

• Sample size found meeting target power
• Results validated with high-precision simulation
• Sensitivity across scenarios documented
• Clear recommendation with rationale

Final Deliverables

1. Recommended sample size with justification
2. Power at recommended N for all scenarios
3. Sample size sensitivity table
4. Optimal interval analysis
5. R code for replication

Configuration Options

• target_power: Target power (default: 0.90)
• alpha: Type I error rate (default: 0.025)
• search_method: "binary" or "grid"
• n_range: Sample size search range (e.g., c(50, 500))
• scenarios: Treatment effect scenarios
• robustness: "expected" (optimize for expected) or "conservative" (ensure power under conservative)