Tidy Itc Workflow

Tidy ITC Workflow

Apply tidy modelling principles from "Tidy Modeling with R" (TMwR) to indirect treatment comparison analyses for consistent, reproducible, and maintainable code.

When to Use This Skill

• Setting up a new ITC analysis project
• Building reproducible analysis pipelines
• Creating standardized interfaces across ITC methods
• Ensuring code quality and maintainability
• Reviewing code for tidy modelling compliance

Core Principles from TMwR

1. The "Pit of Success" Philosophy

• Software should facilitate proper usage by design
• Users should "fall into winning practices" naturally
• Interface must protect users from methodological errors

2. Workflow-Centric Architecture

Every ITC analysis follows this structure:

Data → Validation → Preparation → Analysis → Diagnostics → Reporting

3. Consistent Interfaces

All ITC functions should have predictable patterns:

# Standard function signature pattern
itc_function(
  data,                    # Primary data input
  outcome_var,             # Outcome variable name
  treatment_var,           # Treatment variable name
  covariates = NULL,       # Optional covariates
  method = "default",      # Method specification
  alpha = 0.05,           # Significance level
  seed = NULL,            # For reproducibility
  verbose = TRUE,         # Progress messages
  ...                      # Additional method-specific args
)

# Standard return structure
list(
  results = tibble(...),   # Main results as tibble
  diagnostics = list(...), # Model diagnostics
  model = fitted_model,    # Raw model object
  data_summary = list(...),# Data summary
  call = match.call(),     # Original call
  parameters = list(...)   # Analysis parameters
)

ITC Workflow Structure

Step 1: Project Setup

# Recommended project structure
project/
├── R/
│   ├── 01_data_prep.R
│   ├── 02_analysis.R
│   ├── 03_sensitivity.R
│   └── 04_reporting.R
├── data/
│   ├── raw/
│   └── processed/
├── output/
│   ├── figures/
│   └── tables/
├── renv.lock          # Package versions
└── _targets.R         # Pipeline definition (optional)

Step 2: Environment Setup

# Load packages with explicit namespacing preference
library(tidyverse)
library(meta)       # Pairwise MA
library(netmeta)    # NMA
library(maicplus)   # MAIC
library(multinma)   # ML-NMR

# Set global options
options(
  dplyr.summarise.inform = FALSE,
  mc.cores = parallel::detectCores() - 1
)

# Set seed for reproducibility
set.seed(12345)

Step 3: Data Validation

# Validate IPD structure
validate_ipd  0) {
    errors  0) {
    stop(sprintf("Missing AgD fields: %s", paste(missing, collapse = ", ")))
  }

  # Check numeric fields are positive
  numeric_fields  65, ]

# Good
AGE_THRESHOLD  AGE_THRESHOLD, ]

2. Missing Validation

# Bad
result  0)
result <- maic_anchored(
  weights_object = weights,
  ipd = ipd,
  pseudo_ipd = pseudo_ipd
)

3. Unreproducible Operations

# Bad
bootstrap_ci <- boot::boot.ci(boot_result)

# Good
set.seed(12345)
boot_result <- boot::boot(data, statistic, R = 1000)
bootstrap_ci <- boot::boot.ci(boot_result)

Resources

• TMwR Book: https://www.tmwr.org/
• tidymodels: https://www.tidymodels.org/
• NICE DSU TSD 18: Population-adjusted indirect comparisons

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: choxos
• Source: choxos/BiostatAgent
• License: MIT

Install and usage instructions live in the source repository linked above.