De-identification

PHI removed before anything else.

Sandy removes patient identifiers from clinical text and records with its own models, before anyone or anything else sees them.

Book a walkthroughOn a sample of your own records.
As received

Pt John Rivera, born 04/12/1979, seen for right knee pain after a fall. Call back at (805) 555-0142.

What leaves Sandy

Pt Name, born Date, seen for right knee pain after a fall. Call back at Phone.

The same for every note, from every client and location you serve. Adding a client never adds an outside service.
3x fasterthan routing to an outside service
0.986overlap F1 on a 600-sample labeled set
In-houseraw PHI never goes to a third party
How it detects

Three signals, one clean record.

Structure

Field names, paths, and types flag IDs and records.

Patterns

Exact patterns catch dates, phone numbers, emails, and URLs.

Language

Sandy's own model reads the context to catch names and places in prose. Nothing goes to an outside model.

A record that fails is flagged, never passed on, and logs never store the text.

Validation

Measured against the alternatives.

Overlap F1 on a 600-sample labeled set, against the de-identification approaches teams usually reach for. Higher is better.

Sandy de-id pipeline
0.986
Competitor 1, with helper functions
0.977
Competitor 2, with helper functions
0.940
Competitor 2, direct extraction
0.864
Competitor 1, standard service
0.655
0.60Overlap F1 score1.00

See it run on your own records.

A short walkthrough on a sample of your data, so you can see exactly what it removes.

Book a walkthrough