cp-R Shiny

A Shiny web-application port of cp-R 0.4 (Chemical Pathology R), originally a PyQt4 desktop application by Dr. Daniel T. Holmes, MD.

Features

• Least Squares (optionally 1/x² weighted), Passing-Bablok, and Deming (optionally weighted, constant CV) regression methods
• Right-click a point in the interactive plot to exclude it from (or re-include it in) the analysis
• Bland-Altman difference plot
• Interactive plotly widgets with per-point tooltips showing data-table row numbers
• Static image downloads in JPG, PNG, TIFF, BMP, PDF, and PS formats
• Fully interactive data entry via rhandsontable (paste from spreadsheet, sort, insert/delete rows)

R Package Dependencies

Package	Purpose
shiny	Web framework
rhandsontable	Interactive data-entry grid
shinyjs	JS helpers
jsonlite	Colour list serialisation
plotly	Interactive plots
mcr	Passing-Bablok estimates and analytical CIs
boot	Bootstrap CIs and confidence bands (Passing-Bablok, Deming)

All packages are auto-installed on first run if missing.

Running locally

# From R or RStudio:
shiny::runApp("path/to/cpr-shiny")

# From the terminal:
Rscript -e "shiny::runApp('cpr-shiny')"

Docker deployment

Build

docker build -t cpr-shiny .

Run

docker run --rm -p 3838:3838 cpr-shiny

Then open http://localhost:3838 in your browser.

Run with persistent logs

docker run --rm -p 3838:3838 \
  -v /path/to/logs:/var/log/shiny-server \
  cpr-shiny

Docker Compose

services:
  cpr-shiny:
    build: .
    ports:
      - "3838:3838"
    restart: unless-stopped

Cloud deployment (shinyapps.io)

library(rsconnect)
rsconnect::deployApp("path/to/cpr-shiny")

Directory layout

cpr-shiny/
├── app.R                 <- Shiny UI + server
├── R/                    <- sourced automatically by Shiny at startup
│   ├── analysis.R        <- LS / Passing-Bablok / Deming / Bland-Altman calculations and stats text
│   ├── plots_base.R      <- static plots (preview and downloads)
│   └── plots_plotly.R    <- interactive plots
├── tests/
│   └── compare_mcr.R     <- regression results checked against mcr
├── Dockerfile            <- Docker build definition
└── README.md

All analysis runs inside the Shiny process: packages load once at startup, results are kept in memory, and download files are drawn only when requested. Bootstraps use a fixed seed, so identical inputs give identical results, and run across CPU cores where the platform supports forking (Linux, macOS).

Validation against mcr

tests/compare_mcr.R checks the app's regression functions in R/analysis.R against mcr::mcreg (tested with mcr 1.3.3.1). It uses test_data.csv and synthetic datasets with constant error, proportional error, integer ties and a negative slope, plus Deming error ratios of 1, 0.25 and 4. Run it from the project root:

Rscript tests/compare_mcr.R

It exits non-zero if any check differs by more than 1e-6.

Method	mcr equivalent	Result
Least Squares: estimates and 95% CIs	LinReg	Identical
Weighted Least Squares (1/x²): estimates and slope CI	WLinReg	Identical
Passing-Bablok: estimates and analytical CIs	PaBa	Identical (the app calls mcreg directly)
Passing-Bablok bootstrap estimator for the confidence band (pb_coef)	PaBa, slope.measure = "tangent"	Identical
Deming, unweighted and weighted, all error ratios	Deming, WDeming	Identical (to the convergence tolerance)
Deming bootstrap BCa CIs	Deming, method.ci = "bootstrap"	Close; not exact because both are bootstraps

Error-ratio convention: the app's Deming input is Var(y)/Var(x). mcr's error.ratio is the inverse, Var(x)/Var(y), so an app value of 4 corresponds to error.ratio = 0.25.

Limitations inherited from mcr: mcr refuses Passing-Bablok fits on data containing negative values. For those data the app reports an error rather than a fit.

License

GNU GPL v3 — see gpl-3.0_license.pdf in the original cp-R distribution.