Data Science
HIPAA de-identification: Safe Harbor vs. Expert Determination
Short answer
HIPAA allows two ways to de-identify health data. Safe Harbor removes 18 specified identifiers and requires no actual knowledge that the rest could identify someone. Expert Determination has a qualified expert document that the re-identification risk is very small. Safe Harbor is simpler; Expert Determination keeps more analytic detail, such as dates and locations.
- Author
- Kumbirai Muhwati
- Published
- Last reviewed
- Reading time
- 4 min
What “de-identified” means under HIPAA
Health information that does not identify an individual, and gives no reasonable basis to identify them, is not protected health information (45 CFR 164.514(a)). Once data is de-identified, the Privacy Rule no longer restricts how it is used. That is why de-identification sits at the start of most analytics and AI pipelines. HIPAA recognizes two methods, both in 45 CFR 164.514(b).
Safe Harbor: the 18 identifiers
Under Safe Harbor, the following identifiers of the individual, and of their relatives, employers or household members, must be removed:
- Names.
- All geographic subdivisions smaller than a state, including street address, city, county, precinct, ZIP code and equivalent geocodes. The first three digits of a ZIP code may be kept if the area formed by all ZIP codes sharing those digits contains more than 20,000 people; otherwise they must be changed to 000.
- All elements of dates (except year) directly related to the individual, including birth, admission, discharge and death dates, and all ages over 89 and date elements that indicate such an age, unless grouped into a single category of 90 or older.
- Telephone numbers.
- Fax numbers.
- Email addresses.
- Social Security numbers.
- Medical record numbers.
- Health plan beneficiary numbers.
- Account numbers.
- Certificate and license numbers.
- Vehicle identifiers and serial numbers, including license plate numbers.
- Device identifiers and serial numbers.
- Web URLs.
- IP addresses.
- Biometric identifiers, including finger and voice prints.
- Full-face photographs and comparable images.
- Any other unique identifying number, characteristic or code, except a re-identification code assigned under the rule’s conditions.
There is a second condition: the covered entity must not have actual knowledge that the remaining information could be used, alone or in combination, to identify the individual. A rare diagnosis in a small community can fail this test even after all 18 identifiers are gone.
Expert Determination
Under Expert Determination, a person with appropriate knowledge of and experience with generally accepted statistical and scientific methods determines that the risk is very small that the information could be used, alone or with other reasonably available information, by an anticipated recipient to identify an individual. The expert documents the methods and results.
HIPAA does not name a required credential. HHS’s 2012 guidance describes relevant expertise from statistics, mathematics or other scientific fields. A determination applies to a specific dataset, recipient and release setting, and experts often set an expiry date after which the data must be re-assessed.
Side by side
| Safe Harbor | Expert Determination | |
|---|---|---|
| How it works | Remove 18 identifiers; no actual knowledge that the rest identifies anyone | A qualified expert analyzes the risk and documents that it is very small |
| Data utility | Lower: dates reduced to year, geography to state or three-digit ZIP | Higher: shifted dates or finer geography can be kept where the risk allows |
| Cost and time | Low, and largely automatable | Higher: expert time and documentation |
| Dates and ZIP codes | Year only, with ages over 89 grouped; three ZIP digits only for areas over 20,000 people | Decided case by case, for example by date shifting or generalization |
| Free text such as clinical notes | Every identifier must be found and removed, which is hard to do reliably | The risk assessment includes identifiers that remain in text |
| Best for | Simple datasets shared broadly | Research, real-world evidence and model training that need detail |
Limited data sets: a third option
A limited data set removes 16 direct identifiers listed in 45 CFR 164.514(e) but may keep dates, city, state, five-digit ZIP code and ages. It is still PHI. It may be used only for research, public health or health care operations, and only under a data use agreement that limits how it is used and prohibits re-identifying or contacting individuals.
Re-identification risk in clinical notes and model training
Clinical notes hide identifiers in unexpected places: names of relatives, dates written out in prose, rare occupations and facility names. Automated de-identification tools reduce these but rarely remove every one, so measure the residual identifiers on a reviewed sample and record the error rate.
AI models add a second risk. Large models can memorize and reproduce rare strings from their training data. Train on de-identified data by default, remove duplicates and rare unique strings, and test for extraction before a model is released.
How de-identification fits the GDPR
The GDPR draws the line differently. Pseudonymized data, where identifiers are replaced but a key exists, is still personal data (GDPR Recital 26). Only anonymous data, where people can no longer be identified by any means reasonably likely to be used, falls outside the GDPR. A dataset de-identified under HIPAA Safe Harbor is not automatically anonymous under the GDPR, so international programs need to assess both.
Frequently asked questions
- What are the 18 HIPAA identifiers?
- Names; small geographic subdivisions; date elements other than year; phone and fax numbers; email addresses; Social Security, medical record, health plan, account and certificate or license numbers; vehicle and device identifiers; URLs; IP addresses; biometric identifiers; full-face photos; and any other unique identifying number, characteristic or code.
- Can de-identified data be used to train AI models?
- Yes. De-identified data is not PHI, so HIPAA does not restrict its use, although contracts may. Document the method and test the trained model for memorized data before release.
- Who can perform an Expert Determination?
- A person with appropriate knowledge of and experience with generally accepted statistical and scientific methods. HIPAA does not require a specific degree or certificate.
- Is a limited data set de-identified?
- No. A limited data set is still PHI and may only be used for research, public health or health care operations under a data use agreement.
Sources
- 45 CFR 164.514, de-identification and limited data sets (eCFR)
- HHS: Guidance regarding methods for de-identification (2012)
- GDPR, Regulation (EU) 2016/679, including Recital 26 (EUR-Lex)
General information, not legal or medical advice.