reproducr-cmc

End-to-end reproducr pipeline for a simulated CMC statistical package — dissolution f2, ICH Q1E stability shelf-life, and ICH Q2(R1) assay method validation. renv environment locking, pharma-style QC report.

Domain	Chemistry, Manufacturing and Controls (CMC)
Analyses	Dissolution f2, stability, assay method validation
Regulatory framework	ICH Q1B, ICH Q1E, ICH Q2(R1)
Environment	renv — locked environment
Report style	pharma
Outputs certified	20
Audience	CMC statisticians, analytical chemists, regulatory affairs

See the full walkthrough: DEMO.md

---

Study overview

A simulated CMC statistical package for an immediate-release tablet formulation, covering three standard regulatory analyses:

1. Dissolution profile comparison (ICH Q1B / Q4B Annex 7)

• f2 similarity factor comparing reference and test dissolution profiles
• 6 time points (15–90 min), 12 vessels per arm
• Bootstrap 90% CI for f2
• Acceptance: f2 ≥ 50

2. Stability analysis (ICH Q1E)

• Linear regression of assay (% label claim) vs time
• Three storage conditions: LTS (25°C/60%RH), INT (30°C/65%RH), ACC (40°C/75%RH)
• Shelf-life estimated where 95% one-sided LCL crosses 90% LC acceptance limit

3. Assay method validation (ICH Q2(R1))

• Linearity: 5 concentration levels (50–150%), R² ≥ 0.999
• Repeatability: 6 replicates at 100%, CV ≤ 2%
• Intermediate precision: mixed model (nlme), CV ≤ 3%
• Accuracy: % recovery at 80%, 100%, 120% levels (98–102%)

---

Key results (simulated data)

Dissolution:

Metric	Value	Acceptance
f2 similarity factor	—	≥ 50
f2 90% CI (bootstrap)	—	—

Stability slopes (%LC/month):

Condition	Slope	95% LCL
LTS (25°C/60%RH)	—	—
INT (30°C/65%RH)	—	—
ACC (40°C/75%RH)	—	—

Method validation:

Test	Result	Acceptance
Linearity R²	—	≥ 0.999
Repeatability CV	—	≤ 2%
Intermediate precision CV	—	≤ 3%

Results populated by CI on each run — see .reproducr.rds for certified values.

---

Why CMC analyses need reproducibility tooling

CMC analyses feed directly into regulatory submissions (CTD Module 3 / NDA / ANDA). Silent breaking changes in statistical packages are especially consequential here because:

1. Regulatory precedent — results in a submission are compared against reanalysis during review. Any numerical difference requires explanation.

2. Long timelines — a stability programme may span 5 years from IND to NDA. Package upgrades over that period are inevitable.

3. Method transfers — the same analysis may be run at multiple sites (lab, CRO, sponsor). Locale differences and package version differences can produce different shelf-life estimates.

4. nlme sensitivity — lme() optimizer tolerance and default REML settings have changed across versions. Intermediate precision estimates are sensitive to these changes.

reproducr flags these risks at the call level and certifies the outputs so any numerical change between runs is immediately visible.

---

Running locally

# Clone the repo, then:
renv::restore()
source("analysis.R")

library(reproducr)
report <- audit_script("analysis.R", renv = TRUE)
risks  <- risk_score(report)
print(risks)

---

CI/CD

Workflow	Purpose
R-CMD-check.yml	Restore renv, run structural tests
reproducr-audit.yml	Audit, certify, detect drift, update badge

---

Part of the reproducr gallery

Example	Domain	renv	Report style
reproducr-ecology	Ecology / penguins	No	minimal
reproducr-clinical	Clinical trials / oncology	Yes	pharma
reproducr-rwe	Real world evidence	Yes	academic
reproducr-cmc (this repo)	CMC statistics	Yes	pharma