Power Analysis Workflow

Overview

Calculate statistical power for clinical trial designs across multiple scenarios, analysis methods, and design parameters.

Workflow Phases

Phase 1: Requirements Gathering

<Task> subagent_type: simulation-architect prompt: | Gather requirements for power analysis: 1. What is the primary endpoint type? (continuous, binary, time-to-event, count) 2. What is the trial design? (two-arm, multi-arm, multi-endpoint) 3. What are the treatment effect assumptions? (conservative, expected, optimistic) 4. What is the target power level? 5. What is the Type I error rate (alpha)? 6. Are there any special considerations? (non-proportional hazards, multiplicity, stratification)

Based on responses, determine whether to use simtrial (TTE) or Mediana (CSE) framework. </Task>

Phase 2: Model Selection

<Task> subagent_type: tte-specialist condition: endpoint_type == "time-to-event" prompt: | Set up simtrial-based power analysis: 1. Configure fail_rate for piecewise exponential model 2. Set up enrollment rates 3. Define data cutting strategy (events or calendar) 4. Select analysis method (logrank, weighted, MaxCombo) 5. Prepare sim_fixed_n() or sim_gs_n() parameters </Task> <Task> subagent_type: cse-specialist condition: endpoint_type != "time-to-event" OR multi_scenario == TRUE prompt: | Set up Mediana CSE-based power analysis: 1. Build DataModel with appropriate distribution 2. Define multiple sample size or event count scenarios 3. Configure treatment effect scenarios 4. Build AnalysisModel with tests 5. Build EvaluationModel with power criteria </Task>

Phase 3: Power Calculation

<Task> subagent_type: power-optimizer prompt: | Execute power simulations: 1. Run simulations across all scenarios 2. Calculate power at each scenario combination 3. Generate power curves 4. Identify scenarios meeting target power 5. Summarize operating characteristics </Task>

Phase 4: Review

<Task> subagent_type: code-reviewer prompt: | Review power analysis code and results: 1. Validate simulation parameters 2. Check statistical assumptions 3. Verify reproducibility (seeds) 4. Confirm power calculations are correct 5. Flag any potential issues </Task>

Success Criteria

• Power calculated for all specified scenarios
• Results validated by code reviewer
• Clear power curves or tables generated
• Assumptions documented
• Reproducible code provided

Final Deliverables

1. R code for power analysis
2. Power summary table by scenario
3. Power curves (if applicable)
4. Documentation of assumptions
5. Sensitivity analysis results

Configuration Options

• framework: "simtrial" or "mediana" or "auto"
• n_sims: Number of simulations (default: 10000)
• alpha: Type I error rate (default: 0.025)
• target_power: Target power level (default: 0.90)
• scenarios: List of treatment effect scenarios
• parallel: Enable parallel computation (default: TRUE)