Adaptive Trial Simulator#

Statistical simulation platform for designing and validating adaptive clinical trial designs in silico. Enables optimization of interim analysis strategies, sample size adaptation, and early stopping rules while maintaining Type I error control.

Features#

• Design Simulation: Monte Carlo validation of adaptive designs
• Sample Size Re-estimation: Adapt sample size based on interim data
• Early Stopping Rules: Futility and efficacy boundary optimization
• Type I Error Control: Validate alpha spending strategies
• Multi-Arm Designs: Drop-the-loser and seamless Phase II/III
• Power Optimization: Identify designs with maximum power efficiency

Usage#

Basic Usage#

bash

# Run standard group sequential design
python scripts/main.py

# Adaptive design with sample size re-estimation
python scripts/main.py --design adaptive_reestimate

# Optimize design parameters
python scripts/main.py --optimize

Parameters#

ParameterTypeDefaultRequiredDescription--designstrgroup_sequentialNoTrial design type--n-simulationsint10000NoNumber of Monte Carlo simulations--sample-sizeint200NoInitial sample size per arm--effect-sizefloat0.3NoEffect size (Cohen's d)--alphafloat0.05NoType I error rate--powerfloat0.80NoTarget statistical power--interim-looksint1NoNumber of interim analyses--spending-functionstrobrien_flemingNoAlpha spending function--reestimate-methodstrpromising_zoneNoSample size re-estimation method--outputstrresults.jsonNoOutput file path--visualizeflagFalseNoGenerate visualization charts--optimizeflagFalseNoSearch for optimal design parameters

Advanced Usage#

bash

# Full adaptive design with visualization
python scripts/main.py \
  --design adaptive_reestimate \
  --n-simulations 50000 \
  --sample-size 250 \
  --effect-size 0.35 \
  --interim-looks 2 \
  --spending-function obrien_fleming \
  --visualize \
  --output adaptive_results.json

Design Types#

Design TypeDescriptionUse CaseGroup SequentialFixed interim looks with stopping boundariesStandard adaptive trialsAdaptive Re-estimateSample size adjustment based on interim dataUncertain effect sizeDrop the LoserMulti-arm trials dropping inferior armsPhase II dose selection

Spending Functions#

FunctionCharacteristicsEarly BoundaryO'Brien-FlemingConservative earlyHigh Z-scores earlyPocockAggressive earlyLower Z-scores throughoutPower FamilyModerate (ρ=3)Balanced approach

Output Example#

json

{
  "design_config": {
    "design_type": "adaptive_reestimate",
    "sample_size_per_arm": 200,
    "effect_size": 0.3,
    "alpha": 0.05,
    "target_power": 0.8
  },
  "simulation_results": {
    "power": 0.8234,
    "type_i_error": 0.0481,
    "expected_sample_size": 385.2,
    "early_stop_rate": {
      "efficacy": 0.1523,
      "futility": 0.0841
    }
  }
}

Technical Difficulty: HIGH#

⚠️ AI自主验收状态: 需人工检查

This skill requires:

• Python 3.8+ environment
• NumPy, SciPy, and Matplotlib packages
• Understanding of clinical trial statistics

Dependencies#

bash

pip install -r requirements.txt

Requirements#

text

numpy>=1.20.0
scipy>=1.7.0
matplotlib>=3.4.0

Risk Assessment#

Risk IndicatorAssessmentLevelCode ExecutionPython scripts with mathematical calculationsMediumNetwork AccessNo network accessLowFile System AccessWrites simulation resultsLowInstruction TamperingStatistical parameters could affect resultsMediumData ExposureNo sensitive data exposureLow

Security Checklist#

• No hardcoded credentials or API keys
• No unauthorized file system access
• Output does not expose sensitive information
• Input parameters validated
• Error messages sanitized
• Dependencies audited

Prerequisites#

bash

pip install -r requirements.txt
python scripts/main.py --help

Evaluation Criteria#

Success Metrics#

• Simulations run without errors
• Type I error controlled at nominal level
• Power estimates are accurate
• Visualizations generated correctly

Test Cases#

1. Basic Simulation: Default parameters → Valid results
2. Different Designs: All design types → Appropriate behavior
3. Optimization Mode: --optimize flag → Finds optimal parameters
4. Visualization: --visualize flag → Charts generated

Lifecycle Status#

• Current Stage: Draft

• Next Review Date: 2026-03-15

• Known Issues: Type checking warnings with numpy arrays

• Planned Improvements:

  • Bayesian adaptive designs
  • Multi-arm multi-stage (MAMS) support
  • Enhanced visualization options

References#

Available in references/:

• Adaptive design statistical theory
• Regulatory guidance documents
• Alpha spending function literature
• Sample size re-estimation methods

Limitations#

• Statistical Complexity: Requires biostatistics expertise
• Simulation Time: Large simulations may take hours
• Simplified Models: Does not capture all real-world complexities
• Regulatory Consultation: Results should be validated with regulators

---

⚠️ DISCLAIMER: This tool provides simulation results for research and planning purposes only. All clinical trial designs should be reviewed by qualified biostatisticians and regulatory experts before implementation.