Using chmsflow at an RDC

Overview

Statistics Canada Research Data Centres (RDCs) provide access to CHMS master files in a controlled environment with no internet access. This means you cannot install packages from CRAN or GitHub inside the RDC. This vignette explains how to set up chmsflow at an RDC and how to organize your workflow between the RDC and your regular development environment.

Workflow

Working with chmsflow at an RDC follows a three-step model:

1. Develop outside the RDC – Write and test your R scripts using chmsflow’s bundled dummy data on your own computer. The dummy data mirrors the structure of the real CHMS data, so your code will run the same way inside the RDC.
2. Have the RDC analyst set up the environment – The RDC analyst places the R packages, CHMS data files, and your R scripts into the RDC environment (typically on your P drive).
3. Run inside the RDC – Go into the RDC and run your tested scripts against the real CHMS master file data.

This separation means you can iterate on your analysis code without using RDC time, and only go in when the code is ready to run on real data.

Setup inside the RDC

1. Set the library path

Since there is no internet access at the RDC, all R package dependencies (chmsflow, recodeflow, dplyr, haven, purrr, etc.) must be pre-installed in a folder on your P drive. The RDC analyst can do this for you. Then point R to that folder:

.libPaths("<path_to_your_P_drive_package_folder>")

Note that renv does not work at the RDC because renv::restore() requires internet access to download packages.

2. Load packages

library(chmsflow)
library(recodeflow)
library(dplyr)

3. Load additional scripts and worksheets

If you have custom functions or modified metadata files, load them with source() and read.csv():

source("my_custom_functions.R")
my_variable_details <- read.csv("my_variable_details.csv")

4. Load CHMS data

Load CHMS data files and name the objects to match chmsflow’s naming conventions. This is important because rec_with_table() uses the object name to match the databaseStart column in variable-details.csv.

• Name main cycle data as cycle1 through cycle6
• Name medication data as cycle1_meds through cycle6_meds

# Cycles 1-6: Stata format
cycle1 <- haven::read_stata("cycle1.dta")
cycle2 <- haven::read_stata("cycle2.dta")
cycle3 <- haven::read_stata("cycle3.dta")
cycle4 <- haven::read_stata("cycle4.dta")
cycle5 <- haven::read_stata("cycle5.dta")
cycle6 <- haven::read_stata("cycle6.dta")

# Medication data: Stata for cycles 3-6, SAS for cycles 1-2
cycle1_meds <- haven::read_sas("cycle1_meds.sas7bdat")
cycle2_meds <- haven::read_sas("cycle2_meds.sas7bdat")
cycle3_meds <- haven::read_stata("cycle3_meds.dta")
cycle4_meds <- haven::read_stata("cycle4_meds.dta")
cycle5_meds <- haven::read_stata("cycle5_meds.dta")
cycle6_meds <- haven::read_stata("cycle6_meds.dta")

Some cycles use uppercase column names. Lowercase them so rec_with_table() can match the variable names in variable-details.csv:

names(cycle1) <- tolower(names(cycle1))
names(cycle4) <- tolower(names(cycle4))
names(cycle6) <- tolower(names(cycle6))

names(cycle1_meds) <- tolower(names(cycle1_meds))
names(cycle4_meds) <- tolower(names(cycle4_meds))
names(cycle6_meds) <- tolower(names(cycle6_meds))

5. Run your analysis

From here, the workflow is identical to what you tested with dummy data. See Analysis walkthrough for a complete example, or Recoding medications for the medication recoding pipeline.

Tips

• Test thoroughly outside the RDC. RDC time is limited. Use chmsflow’s dummy data to verify your code runs end-to-end before going in.
• Bring all dependencies. Check what chmsflow imports by running tools::package_dependencies("chmsflow", recursive = TRUE) outside the RDC, and make sure all of those packages are on your P drive.
• Keep medication data separate. Do not merge medication files into the main cycle files before loading. chmsflow expects them as separate objects for recode_meds_cycles1to2() and recode_meds_cycles3to6().

Next steps

• Run a complete analysis – Follow the Analysis walkthrough to estimate hypertension prevalence across cycles 3–6 using dummy data, then bring the same code into the RDC.
• Recode medications – If your analysis needs medication variables, see Recoding medications for the full pipeline.
• Understand the methodology – For background on why harmonization is non-trivial, see Methodology.