clinical-research-analysis-framework

Statistical Analysis Framework

Overview

This skill provides a consultation-first workflow for statistical and machine learning analyses. Rather than jumping to implementation, it guides users through structured decision-making to ensure:

• Research intent is fully captured before any code runs
• Method selection is informed by trade-offs, not defaults
• Assumptions and limitations are explicitly acknowledged
• Execution is stepwise with checkpoints for user feedback
• All work is reproducible with complete audit trails

When to Use This Skill

• Planning a new analysis on MIMIC, eICU, or similar data
• Choosing between statistical inference vs predictive modeling
• Needing guidance on appropriate methods for a research question
• Executing analyses with reproducibility requirements
• Validating ML models or statistical findings

---

Guiding Philosophy

This Skill Guides, Not Limits

The methods and checklists in this skill are starting points, not boundaries. Claude's statistical knowledge extends far beyond what is explicitly listed here. When a user's problem calls for a method not covered in the references — whether that's g-computation, Bayesian hierarchical models, causal forests, doubly robust estimation, joint longitudinal-survival models, or something else entirely — Claude should:

1. Suggest it with clear explanation of why it fits better
2. Explain the intuition so the user understands the approach
3. Compare it to simpler alternatives with trade-offs
4. Let the user decide based on understanding, not just trust

The Collaboration Model

User describes problem
        ↓
Claude suggests methods (from full knowledge, not just this skill)
        ↓
Claude explains: "Here's why this fits your situation..."
        ↓
User asks questions, pushes back, explores alternatives
        ↓
Claude refines: "Given that concern, consider instead..."
        ↓
User makes informed choice
        ↓
Proceed with understanding, not blind execution