Privacy Sensitive Project

Overview

The Privacy Sensitive Project is designed for research involving sensitive or protected data. It provides enhanced privacy controls with separate private and public directories, and stricter .gitignore rules that prevent private data from being committed.

Directory Structure

my-study/
├── functions/           # Custom R functions (auto-loaded)
│   └── helpers.R
├── inputs/
│   ├── private/         # Sensitive data (gitignored)
│   │   ├── raw/
│   │   ├── intermediate/
│   │   └── final/
│   └── public/          # Non-sensitive data
│       ├── raw/
│       ├── intermediate/
│       └── final/
├── notebooks/           # Analysis notebooks
├── outputs/
│   └── private/
│       └── notebooks/   # Rendered notebooks (gitignored)
├── scripts/             # Automation scripts
├── framework.db         # Metadata database
└── settings.yml         # Project configuration

Additional directories are created on first use:

• outputs/private/cache - Cached computations
• outputs/private/figures - Private figures
• outputs/private/tables - Private tables
• outputs/public/ - Shareable results

Privacy Controls

What Gets Gitignored

The privacy sensitive .gitignore template excludes:

• inputs/private/ - All private input data
• outputs/private/ - All private outputs
• .env - Environment variables with credentials
• framework.db - May contain data hashes

Public vs Private Outputs

Use the public parameter when saving outputs:

# Save to private (default)
save_table(results, "analysis_results")

# Save to public for sharing
save_table(summary_stats, "summary", public = TRUE)

Workflow

1. Set Up Environment

library(framework)
scaffold()

2. Organize Data by Sensitivity

Place sensitive data in private directories:

data:
  # Private: contains PHI
  inputs.private.raw.patients:
    path: inputs/private/raw/patients.csv
    type: csv
    locked: true

  # Public: aggregated, de-identified
  inputs.public.final.demographics:
    path: inputs/public/final/demographics.csv
    type: csv

3. Work with Private Data

# Load private data
patients <- data_read("inputs.private.raw.patients")

# Process and de-identify
demographics <- patients |>
  group_by(age_group, region) |>
  summarize(n = n())

# Save de-identified data to public
data_save(demographics, "inputs.public.final.demographics")

4. Generate Public Outputs

# Private analysis output
save_figure(detailed_plot, "patient_outcomes")

# Public summary for publication
save_figure(summary_plot, "figure_1", public = TRUE)

Configuration

Example settings.yml for a privacy sensitive project:

default:
  project_type: project_sensitive

  directories:
    notebooks: notebooks
    scripts: scripts
    functions: functions
    inputs_private_raw: inputs/private/raw
    inputs_private_intermediate: inputs/private/intermediate
    inputs_private_final: inputs/private/final
    inputs_public_raw: inputs/public/raw
    inputs_public_intermediate: inputs/public/intermediate
    inputs_public_final: inputs/public/final
    cache: outputs/private/cache

  packages:
    - name: dplyr
      auto_attach: true
    - name: ggplot2
      auto_attach: true

  data:
    inputs.private.raw.patients:
      path: inputs/private/raw/patients.csv
      type: csv
      locked: true

Best Practices

Lock Sensitive Source Data

Use locked: true to ensure source data hasn't changed:

data:
  inputs.private.raw.patients:
    path: inputs/private/raw/patients.csv
    type: csv
    locked: true  # Error if file changes

De-identify Before Publishing

Always aggregate or de-identify data before moving to public directories:

# Never do this
save_table(raw_patient_data, "patients", public = TRUE)

# Do this instead
public_summary <- raw_patient_data |>
  group_by(category) |>
  summarize(mean_value = mean(value), n = n())

save_table(public_summary, "summary", public = TRUE)

Review Before Committing

Check git status before committing to ensure no private data is staged:

git_status()

---