choxos/ITC-agents

Latest recorded source activity: December 8, 2025 at 22:29

SkillsMP catalog refreshed: November 2, 2026 at 18:02

skills collected: 5

GitHub stars: 0
GitHub forks: 0

Install with an AI assistant

Copy this prompt into the AI assistant you're using.

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/choxos/ITC-agents.

Repository README

Repository README source: README.md

Read the source README

Expand full README

# ITC Modelling Agents: Evidence Synthesis Automation for R > **🎯 Expert ITC Scientists** — 7 specialized agents with world-class methodology knowledge following NICE DSU guidance > > **📊 Complete Method Coverage** — Pairwise MA, NMA, MAIC, STC, and ML-NMR with 8 R packages supported > > **🔬 Deep Methodology Skills** — 6 progressive disclosure skills for evidence synthesis best practices A comprehensive production-ready system combining **7 specialized ITC agents**, **2 multi-agent workflow orchestrators**, **6 methodology skills**, and complete coverage of indirect treatment comparison methods for [Claude Code](https://docs.anthropic.com/en/docs/claude-code). ## Overview This plugin provides everything needed for rigorous indirect treatment comparison modelling in R: - **7 Specialized Agents** - Domain experts covering method selection, analysis execution, and code review - **6 Methodology Skills** - Deep knowledge packages following NICE DSU TSDs, Cochrane, and PRISMA guidelines - **2 Workflow Orchestrators** - Multi-agent coordination for complete ITC analyses and code reviews - **8 R Packages Supported** - `meta`, `metafor`, `bayesmeta`, `netmeta`, `gemtc`, `maicplus`, `stc`, `multinma` - **Safe Code Amendment** - Original files never modified; corrections saved to subfolders ### Key Features - **World-Class Expertise**: Agents embody best practices from NICE DSU TSDs, Cochrane Handbook, and leading HTA methodologists - **Assumption Transparency**: Every analysis explicitly checks and documents methodological assumptions - **Tidy Modelling Compliance**: All code follows TMwR patterns for reproducibility and consistency - **Comprehensive Review Options**: Choose between detailed comprehensive or concise summary reviews - **Safe Amendment Mode**: Corrected code saved to subfolders with REVIEW_NOTES.md documenting all changes - **Progressive Disclosure**: Skills load methodology knowledge only when needed for token efficiency ### How It Works The plugin provides specialized agents for each ITC methodology: - **Method selection** - ITC Architect guides you to the appropriate method based on data and assumptions - **Analysis execution** - Specialist agents (pairwise, NMA, MAIC, STC, ML-NMR) implement rigorous analyses - **Code review** - ITC Code Reviewer identifies methodological errors and optionally amends code - **Methodology guidance** - Skills provide deep knowledge on demand **Example**: Asking about MAIC automatically activates the `maic-methodology` skill with ESS thresholds, weight diagnostics interpretation, and anchored vs unanchored decision frameworks. ## Quick Start ### Step 1: Add the Plugin Add this plugin to Claude Code: ```bash /plugin add choxos/ITC-agents ``` ### Step 2: Start Using Agents The agents are now available for ITC analyses: ```bash # Get method recommendation "I have IPD from our trial and AgD from a competitor. Help me choose an ITC method." # Run specific analysis "Conduct a MAIC analysis with these data, checking ESS and covariate balance." # Review existing code "Review my NMA code for methodological issues - comprehensive review please." ``` ## ITC Methods Covered | Method | R Packages | When to Use | |--------|------------|-------------| | **Pairwise Meta-Analysis** | `meta`, `metafor`, `bayesmeta` | Direct comparison with multiple trials | | **Network Meta-Analysis** | `netmeta`, `gemtc` | Connected network allowing indirect comparisons | | **MAIC** | `maicplus` | IPD vs AgD with population differences (weighting) | | **STC** | `stc` | IPD vs AgD with population differences (regression) | | **ML-NMR** | `multinma` | Combined IPD/AgD network with population adjustment | ## Agents ### Strategic Planning | Agent | Model | Purpose | |-------|-------|---------| | **itc-architect** | Opus | Master strategist for evidence assessment and method selection | ### Method Specialists | Agent | Model | Packages | Expertise | |-------|-------|----------|-----------| | **pairwise-meta-analyst** | Sonnet | `meta`, `metafor`, `bayesmeta` | Heterogeneity, publication bias, forest plots | | **nma-specialist** | Sonnet | `netmeta`, `gemtc` | Consistency, rankings, league tables | | **maic-specialist** | Sonnet | `maicplus` | Weight estimation, ESS, Bucher method | | **stc-specialist** | Sonnet | `stc` | Effect modifiers, centering, outcome regression | | **ml-nmr-specialist** | Sonnet | `multinma` | Integration points, population adjustment | ### Quality Assurance | Agent | Model | Purpose | |-------|-------|---------| | **itc-code-reviewer** | Opus | Methodological review with optional code amendment | ## Methodology Skills Progressive disclosure skills provide deep methodology knowledge: | Skill | Content | |-------|---------| | **tidy-itc-workflow** | TMwR patterns, reproducibility, data validation, result tibbles | | **pairwise-ma-methodology** | Fixed vs random effects, I²/τ² interpretation, publication bias | | **nma-methodology** | Transitivity, consistency assessment, ranking interpretation | | **maic-methodology** | ESS thresholds (>50%), weight diagnostics, covariate selection | | **stc-methodology** | Effect modifier identification, centering, MAIC comparison | | **ml-nmr-methodology** | Integration points, marginal vs conditional effects | ### Skill Activation Skills activate automatically based on context: ``` User: "What ESS is acceptable for my MAIC?" → Activates maic-methodology skill → Provides ESS interpretation table (>70% good, 50-70% acceptable, <30% poor) → Suggests strategies for improving ESS ``` ## Workflow Commands ### `/itc-analysis` - Complete ITC Analysis Orchestrates multi-agent workflow from start to finish: ``` /itc-analysis Design and execute a complete MAIC analysis comparing Treatment A to Treatment B using our IPD and published competitor data. ``` **Workflow Phases:** 1. **Evidence Assessment** (itc-architect) - Map studies, identify data types 2. **Method Selection** (itc-architect) - Recommend approach with justification 3. **Data Preparation** (itc-code-reviewer) - Validate and prepare data 4. **Primary Analysis** (appropriate specialist) - Execute analysis with diagnostics 5. **Code Review** (itc-code-reviewer) - Validate implementation 6. **Final Reporting** (itc-architect) - Synthesize results ### `/itc-review` - Code Review Workflow ``` /itc-review Review my MAIC analysis code: review_type: comprehensive action: amend output_folder: ./reviewed/ ``` **Options:** - `review_type`: `comprehensive` (detailed) | `summary` (bullet points) - `action`: `review_only` | `amend` (create corrected files) - `output_folder`: Path for amended files (default: `./amended/`) **Output Structure (amend mode):** ``` project/ ├── analysis.R # Original (NEVER modified) └── amended/ ├── analysis_amended.R # Corrected version ├── REVIEW_NOTES.md # Change summary └── backup/ └── analysis_original.R ``` ## Popular Use Cases ### Method Selection Guidance ``` Help me choose an ITC method. I have: - IPD from our Phase 3 trial (Treatment A vs Placebo, n=450) - Published AgD from competitor trial (Treatment B vs Placebo, n=380) - Populations differ in age (median 58 vs 65) and prior therapy (35% vs 52%) ``` **Response includes:** - Recommendation: Anchored MAIC or STC - Justification referencing NICE DSU TSD 18 - Assumption checklist - Sensitivity analysis plan ### MAIC Analysis with Diagnostics ``` Run anchored MAIC for response rate: - Match on: age, sex, ECOG status, prior therapy - Check ESS and covariate balance - Use bootstrap CIs (1000 iterations) - Compare with STC as sensitivity ``` **Agent activates:** - `maic-methodology` skill for ESS interpretation - Provides complete `maicplus` code with diagnostics - Flags if ESS < 50% with mitigation strategies ### NMA with Consistency Assessment ``` Conduct Bayesian NMA for this 8-treatment network: - Check transitivity across comparisons - Test consistency via node-splitting - Generate league table and SUCRA rankings - Use informative priors for heterogeneity ``` **Agent provides:** - Network visualization code - Complete `gemtc` implementation - Convergence diagnostics (R-hat, traces) - Interpretation of rankings with uncertainty ### Code Review with Amendment ``` Review my meta-analysis code for methodological issues: review_type: comprehensive action: amend library(meta) ma <- metabin(events.e, n.e, events.c, n.c, data = studies, sm = "OR") forest(ma) ``` **Review identifies:** - Missing: `method.tau = "REML"`, `hakn = TRUE` - Missing: prediction intervals, heterogeneity assessment - Missing: publication bias check (if ≥10 studies) - Creates amended code with all corrections ## Methodology References ### Primary Guidelines - **[NICE DSU TSD 2](https://www.sheffield.ac.uk/nice-dsu/tsds/full-list)** - Network meta-analysis - **[NICE DSU TSD 18](https://www.sheffield.ac.uk/nice-dsu/tsds/full-list)** - Population-adjusted indirect comparisons - **[Cochrane Handbook](https://training.cochrane.org/handbook)** - Systematic reviews - **[PRISMA-NMA](http://prisma-statement.org/Extensions/NetworkMetaAnalysis)** - Reporting guidelines ### Key Papers - Dias et al. (2018). Network Meta-Analysis for Decision Making - Phillippo et al. (2018). Methods for Population-Adjusted Indirect Comparisons - Phillippo et al. (2020). Multilevel Network Meta-Regression - Signorovitch et al. (2010). Matching-Adjusted Indirect Comparisons - Higgins et al. (2019). Cochrane Handbook, Chapter 10 ### R Package Documentation - [meta](https://cran.r-project.org/package=meta) - General Package for Meta-Analysis - [metafor](https://www.metafor-project.org/) - Meta-Analysis Package for R - [bayesmeta](https://cran.r-project.org/package=bayesmeta) - Bayesian Random-Effects Meta-Analysis - [netmeta](https://cran.r-project.org/package=netmeta) - Network Meta-Analysis using Frequentist Methods - [gemtc](https://cran.r-project.org/package=gemtc) - Network Meta-Analysis Using Bayesian Methods - [maicplus](https://hta-pharma.github.io/maicplus/) - Matching Adjusted Indirect Comparison - [multinma](https://dmphillippo.github.io/multinma/) - Multilevel Network Meta-Regression ## Architecture ### Repository Structure ``` ITC-agents/ ├── .claude-plugin/ │ └── marketplace.json # Plugin registration ├── plugins/ │ └── itc-modelling/ │ ├── agents/ │ │ ├── itc-architect.md # Method selection (Opus) │ │ ├── pairwise-meta-analyst.md # meta/metafor/bayesmeta │ │ ├── nma-specialist.md # netmeta/gemtc │ │ ├── maic-specialist.md # maicplus │ │ ├── stc-specialist.md # stc │ │ ├── ml-nmr-specialist.md # multinma │ │ └── itc-code-reviewer.md # Code review (Opus) │ ├── commands/ │ │ ├── itc-analysis.md # Multi-agent orchestration │ │ └── itc-review.md # Code review workflow │ └── skills/ │ ├── tidy-itc-workflow/ # TMwR patterns │ ├── pairwise-ma-methodology/ # MA methodology │ ├── nma-methodology/ # NMA methodology │ ├── maic-methodology/ # MAIC methodology │ ├── stc-methodology/ # STC methodology │ └── ml-nmr-methodology/ # ML-NMR methodology ├── docs/ # Additional documentation ├── LICENSE # MIT License └── README.md # This file ``` ### Design Principles 1. **Tidy Modelling Compliance** - All code follows TMwR patterns (workflows, recipes, consistent interfaces) 2. **Assumption Transparency** - Every method explicitly checks and documents assumptions 3. **Safe Code Handling** - Original files never modified; amendments in subfolders 4. **Progressive Disclosure** - Skills load deep knowledge only when activated 5. **Reproducibility First** - Seeds, package versions, session info always documented ### Model Assignment | Role | Model | Rationale | |------|-------|-----------| | **Strategic** (itc-architect) | Opus | Complex reasoning for method selection | | **Specialists** (5 agents) | Sonnet | Efficient execution of defined methodologies | | **Review** (itc-code-reviewer) | Opus | Deep analysis for error detection | ## Contributing Contributions welcome! Please ensure: 1. **Methodology** - Follows NICE DSU guidance and Cochrane standards 2. **R Patterns** - Matches package APIs exactly 3. **Tidy Principles** - TMwR-compliant code 4. **Assumptions** - Clearly documented and checked 5. **Testing** - Example code verified to run ### Adding New Content ```bash # Add a new agent plugins/itc-modelling/agents/new-specialist.md # Add a new skill plugins/itc-modelling/skills/new-methodology/SKILL.md # Update marketplace.json with new components .claude-plugin/marketplace.json ``` ## Resources ### Claude Code Documentation - [Claude Code Overview](https://docs.anthropic.com/en/docs/claude-code) - [Plugins Guide](https://docs.anthropic.com/en/docs/claude-code/plugins) - [Subagents Guide](https://docs.anthropic.com/en/docs/claude-code/sub-agents) - [Agent Skills](https://docs.anthropic.com/en/docs/agents-and-tools/agent-skills) ### ITC Methodology Resources - [NICE DSU Technical Support Documents](https://www.sheffield.ac.uk/nice-dsu/tsds/full-list) - [Cochrane Training](https://training.cochrane.org/) - [R-HTA Consortium](https://r-hta.org/) ## License MIT License - see [LICENSE](LICENSE) file for details. ## Author **Ahmad Sofi-Mahmudi** --- *Expert ITC agents built with Claude Code following best practices in evidence synthesis methodology.*

Loading the repository README…

Skills in this repository

Data Scientists Economists Survey Researchers

3 occupation categories · 100% classified

skill

occupation

description

updated

maic-methodology

Data Scientists

Deep methodology knowledge for MAIC including assumptions, weight diagnostics, ESS interpretation, and anchored vs unanchored decisions. Use when conducting or reviewing MAIC analyses.

2025-12-08

ml-nmr-methodology

Economists

Deep methodology knowledge for ML-NMR including IPD/AgD integration, population adjustment, numerical integration, and prediction to target populations. Use when conducting or reviewing ML-NMR analyses.

2025-12-08

nma-methodology

Survey Researchers

Deep methodology knowledge for network meta-analysis including transitivity, consistency assessment, treatment rankings, and model selection. Use when conducting or reviewing NMA.

2025-12-08

stc-methodology

Data Scientists

Deep methodology knowledge for STC including outcome regression approach, effect modifier selection, covariate centering, and comparison with MAIC. Use when conducting or reviewing STC analyses.

2025-12-08

tidy-itc-workflow

Data Scientists

Master tidy modelling patterns for ITC analyses following TMwR principles. Covers workflow structure, consistent interfaces, reproducibility best practices, and data validation. Use when setting up ITC analysis projects or building pipelines.

2025-12-08

Showing 5 of 5 collected skills.