Files
vigilcare-clinical/docs/ops/phi-encryption-runbook.md
T
2026-06-23 00:16:39 +08:00

199 lines
7.3 KiB
Markdown

# PHI Encryption & Access Logging — Operations Runbook
Phase 32 encrypts sensitive patient demographics at rest and records who accessed PHI. This runbook covers key management, migration, rotation, compliance, and troubleshooting.
---
## Architecture summary
| Layer | Mechanism |
|---|---|
| Encryption at rest | ASP.NET Data Protection API (AES-256-GCM) via EF Core value converters |
| Encrypted columns | `first_name`, `last_name`, `date_of_birth`, `allergies`, `emergency_contact_name`, `emergency_contact_phone` |
| Plaintext (by design) | `mrn` — exact-match lookup only |
| Name search | HMAC-SHA256 `name_search_token` index — search without decrypting all rows |
| Access audit | Append-only `phi_access_logs` table; written by `PhiAccessLogService` on patient Get/List/Register/Update |
Configuration lives in `appsettings.json` under `PhiEncryption` and `DataProtection`. Implementation: `PhiEncryptionService`, `PatientPhiConverterConfigurator`, `PatientService`.
---
## Secrets and key storage
**Never commit:**
- `data-protection-keys/` (Data Protection key ring)
- Production `PhiEncryption:SearchTokenKey`
- Any Key Vault / KMS credentials
`data-protection-keys/` is listed in `.gitignore`. Treat loss of this directory as **unrecoverable data loss** for encrypted PHI columns.
| Environment | Data Protection key ring | Search HMAC key (`SearchTokenKey`) |
|---|---|---|
| Development | `./data-protection-keys/` on disk (`DataProtection:KeyPath`) | `appsettings.json` (dev placeholder only) |
| Production | Azure Key Vault XML blob or AWS KMS-backed store | Key Vault / Secrets Manager secret — **not** appsettings |
---
## Initial deployment / encrypting existing rows
New patients are encrypted automatically on save. Existing plaintext rows need a one-time re-save.
### Option A — CLI (recommended)
```bash
./scripts/encrypt-existing-patient-phi.sh
```
Or directly:
```bash
dotnet run --project VigilCareClinicalAPI -- encrypt-phi
```
Loads every patient through EF, applies value converters, and recomputes `name_search_token`.
### Option B — Startup hosted service
`PatientPhiMigrationService` runs on application start and backfills search tokens for rows that are not yet encrypted. It is registered in `Program.cs`. Disable or remove after the first successful production deploy to avoid redundant work on every restart.
---
## Verifying encryption
### Automated tests
```bash
dotnet test VigilCareClinicalAPI.Tests --filter "FullyQualifiedName~PhiEncryption"
```
### Live stack (API + Postgres running)
```bash
./scripts/run-phase32-verification.sh
```
### Manual DB check
Encrypted `first_name` values will **not** equal the patient's plaintext name:
```sql
SELECT id, mrn, left(first_name, 30) AS encrypted_prefix
FROM patients
LIMIT 5;
```
Prefix often starts with `CfDJ8` (Data Protection default).
### PHI access logs
```bash
# Admin token required (AuditRead permission)
curl -H "Authorization: Bearer $TOKEN" \
"http://localhost:5270/api/v1/phi-access-logs?patientId=<uuid>"
```
Prometheus metric: `phi_access_logs_total{access_type="VIEW|LIST|SEARCH|CREATE|UPDATE"}`.
---
## Key rotation
**Always back up the current key ring before rotating.**
### Encryption keys (Data Protection)
1. Snapshot `data-protection-keys/` (or export Key Vault blob).
2. Deploy new key ring **alongside** the old one (Data Protection supports multiple keys; newest encrypts, all decrypt).
3. Change `PhiEncryption:ProtectorPurpose` to a new version string (e.g. `VigilCare.PatientPhi.v2`).
4. Re-run encrypt command to re-encrypt all patient rows with the new protector:
```bash
dotnet run --project VigilCareClinicalAPI -- encrypt-phi
```
5. Verify API reads and raw DB ciphertext look correct.
6. Retire old keys only after confirming all rows decrypt successfully.
### Search HMAC key (`SearchTokenKey`)
1. Store new key in secrets manager.
2. Update `PhiEncryption:SearchTokenKey` in configuration.
3. Re-run `encrypt-phi` (recomputes all `name_search_token` values).
4. Confirm name search still works (`GET /api/v1/patients?q=First+Last`).
Rotating one key without the other does not require touching the other, but both rotations need a full patient re-save.
---
## PHI access log retention (HIPAA)
`phi_access_logs` records who accessed patient demographics, when, from which path, and (hashed) search terms. HIPAA requires audit controls; retention guidance is **minimum 6 years**.
This phase creates the table and query API (`GET /api/v1/phi-access-logs`). Automated retention/archival policy is **out of scope** — plan for Phase 33+ (partitioning, cold storage, or purge job with legal review).
Query endpoints:
| Endpoint | Permission |
|---|---|
| `GET /api/v1/phi-access-logs` | `AuditRead` |
| `GET /api/v1/phi-access-logs/patients/{id}` | `PatientsRead` |
---
## Cross-system PHI (out of scope)
These paths are **not** covered by column encryption in PostgreSQL:
| System | Risk | Action |
|---|---|---|
| Elasticsearch patient documents | May index decrypted names at index time | Exclude PHI fields or accept decrypted-at-index policy |
| Kafka / data lake Parquet events | `patientName` in outbox payloads (e.g. encounter events) | Review lake schemas; redact or tokenize in a future phase |
| Application logs | Serilog must not log PHI fields | Audit log configuration |
---
## Troubleshooting
### `CryptographicException` / cannot decrypt patient
- Key ring missing or wrong `DataProtection:KeyPath`
- App deployed to new host without copying `data-protection-keys/`
- `ProtectorPurpose` changed without re-running `encrypt-phi`
**Fix:** Restore key ring from backup. Do not delete old keys until all data is re-encrypted.
### Name search returns no results
- `name_search_token` is null on older rows → run `encrypt-phi`
- `SearchTokenKey` changed without token recompute
- Query format: two-term search uses `First Last` (space-separated); single term matches first or last name only
### PHI access logs empty
- Caller not authenticated (`PhiAccessLogService` skips unauthenticated requests)
- `PhiEncryption:LogListAccess` is `false` (list/search aggregate logs suppressed)
- Integration/FHIR paths must still authenticate (Phase 31 Integration role)
### Column length errors on encrypt
Migration `WidenPhiEncryptedColumns` widens `first_name`, `last_name`, and emergency contact columns to `text`. Ciphertext is longer than plaintext. If new columns are added to encryption, ensure DB column types accommodate protected payload size.
### Tests fail with 500 on patient register
- Confirm migrations applied (`AddPatientNameSearchToken`, `AddPhiAccessLogs`, `WidenPhiEncryptedColumns`)
- Confirm `PhiEncryption:SearchTokenKey` is set in configuration
- Confirm `data-protection-keys/` is writable in the API working directory
---
## Related files
| File | Purpose |
|---|---|
| `VigilCareClinicalAPI/Services/PhiEncryptionService.cs` | Encrypt/decrypt + HMAC tokens |
| `VigilCareClinicalAPI/Services/PhiAccessLogService.cs` | Access audit writes |
| `VigilCareClinicalAPI/Commands/EncryptPhiCommand.cs` | Bulk re-save CLI |
| `VigilCareClinicalAPI/BackgroundServices/PatientPhiMigrationService.cs` | Startup token backfill |
| `scripts/encrypt-existing-patient-phi.sh` | Wrapper for encrypt CLI |
| `scripts/run-phase32-verification.sh` | End-to-end verification |
| `docs/plans/phase-32-plan.md` | Implementation plan and design rationale |