PHUSE: Applying Advanced Data Privacy Methods to Real-World Data (RWD)

This repository is a shared workspace for the PHUSE project team to collaborate on:

• Benchmark specifications (what we evaluate)
• Reference implementations (how we run the benchmarks)
• Documentation that helps stakeholders review scope, structure, and progress

What this repo is not:

• A place to store raw patient-level datasets. We only keep dataset metadata in GitHub.

Quick links

• Dataset registry (33 datasets): datasets/README.md
• Privacy methods: privacy_methods/
• Research & paper reviews: research/
• Documentation by data type: docs/
• Next steps & roadmap: next_steps.md
• Contributing guide: CONTRIBUTING.md

Repo structure

phuseprivacy/
├── privacy_methods/                 # Privacy method implementations
│   ├── d1_structured_data/          # Structured (tabular) clinical data
│   │   ├── baseline_deidentification/
│   │   ├── differential_privacy/
│   │   ├── federated_learning/
│   │   └── synthetic_data/
│   └── d2_unstructured_data/        # Clinical notes / unstructured text
│       ├── deidentification/
│       ├── llm_privacy_controls/
│       └── privacy_attacks/
│
├── datasets/                        # Dataset registry — metadata only (33 datasets)
│   ├── README.md                    # Full reference with links & access info
│   ├── datasets.yaml                # Machine-readable registry
│   ├── 01_mimic-iv_icu_ehr/
│   ├── 02_mimic-iv-note_clinical_notes/
│   ├── ...
│   └── 33_openmhealth_wearables/
│
├── docs/                            # Documentation (mirrors privacy_methods/)
│   ├── d1_structured_data/
│   │   ├── baseline_deidentification/
│   │   ├── differential_privacy/
│   │   ├── federated_learning/
│   │   └── synthetic_data/
│   └── d2_unstructured_data/
│       ├── deidentification/
│       ├── llm_privacy_controls/
│       └── privacy_attacks/
│
├── research/                        # Paper reviews & implementation reports
│   ├── d1_structured_data/          # Papers on structured data methods
│   ├── d2_unstructured_data/        # Papers on unstructured text methods
│   └── planning/                    # Tech Implementation Plan & strategy
│
├── experiments/                     # Local experiment outputs (ignored by git)
├── data/                            # Local data directory (ignored by git)
└── results/                         # Local outputs directory (ignored by git)

Folder conventions

All three core folders — privacy_methods/, docs/, and research/ — share the same D1/D2 hierarchy:

Data Type	Folder	Methods
D1 Structured (tabular)	d1_structured_data/	baseline_deidentification, differential_privacy, federated_learning, synthetic_data
D2 Unstructured (text)	d2_unstructured_data/	deidentification, llm_privacy_controls, privacy_attacks

Contributing

See CONTRIBUTING.md for the full guide. In short:

1. Work in a group branch
2. Open a Pull Request into main
3. A maintainer reviews and merges

License

See LICENSE.