HIPAA de-identification: Safe Harbor vs Expert Determination
When health data stops being protected health information, the two methods the HIPAA Privacy Rule allows, the 18 identifiers, and the limited data set that is often confused with them.
Why does de-identification matter?
The HIPAA Privacy Rule protects protected health information (PHI): health information that identifies a person, or could reasonably be used to. Once data is de-identified by one of the two methods the rule sets out, it is no longer PHI, and the rule no longer limits how it is used or shared. That is what lets a hospital give a dataset to researchers, a vendor or a public registry without each patient's authorisation.
This is a legal classification under HIPAA, not a guarantee that no one could ever be re-identified. That is why Safe Harbor also requires no actual knowledge that the rest could identify anyone, and why Expert Determination asks the expert to judge the risk for the data and the recipient.
What does the Safe Harbor method require?
Two conditions, both required:
- The 18 identifiers listed below are removed, for the individual and for the individual's relatives, employers and household members.
- The organisation has no actual knowledge that the information left could be used, alone or with other information, to identify the person. If a record says "the hospital's chief executive, aged 54", removing the name is not enough.
Safe Harbor is a checklist: simple to apply and to audit, but it removes a lot, above all dates and detailed geography, which researchers often need.
What are the 18 Safe Harbor identifiers?
- Names.
- Geographic units smaller than a state: street address, city, county, precinct, ZIP code and their equivalent geocodes. The first three digits of a ZIP code may stay if all the ZIP codes sharing them cover more than 20,000 people; otherwise they become 000.
- All elements of dates (except the year) directly related to the person: birth, admission, discharge, death. For anyone over 89, the age and every date element that reveals it, the year included, must be removed; they may be 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 licence numbers.
- Vehicle identifiers and serial numbers, including licence plates.
- Device identifiers and serial numbers.
- Web addresses (URLs).
- IP addresses.
- Biometric identifiers, such as fingerprints and voice prints.
- Full-face photographs and any comparable images.
- Any other unique identifying number, characteristic or code.
Under either method, the organisation may give each record a re-identification code of its own, so that it can link the data back later, provided the code is not derived from information about the person (not a scrambled medical record number, for example) and the key to it is not disclosed.
What does the Expert Determination method require?
- A person with appropriate knowledge of and experience with generally accepted statistical and scientific methods for making information not individually identifiable.
- That person applies those methods and 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 someone.
- The methods and results of the analysis are documented.
HHS sets no single number for "very small"; the expert judges it for the data and the recipient. The method may keep things Safe Harbor would remove, such as dates of service or a 5-digit ZIP code, if the analysis shows the risk stays very small, for example by generalising other fields. The expert may also limit the conclusion to a certain recipient or a certain period.
Safe Harbor and Expert Determination side by side
| Safe Harbor | Expert Determination | |
|---|---|---|
| How it works | Remove a fixed list of 18 identifiers | Statistical analysis of the risk of identification |
| Who decides | The organisation applying Safe Harbor, by the list of identifiers | A qualified expert |
| Extra condition | No actual knowledge that the rest could identify anyone | Methods and results documented |
| Dates and ZIP codes | Year only; first 3 ZIP digits, under the population rule | Can stay, if the risk remains very small |
| Usefulness for research | Lower: much detail removed | Higher: detail kept where it is safe |
| Cost and effort | Low | Higher: an expert and an analysis |
How is a limited data set different?
A limited data set is a third thing, and it is still PHI. It removes direct identifiers (names, street addresses, telephone numbers, record numbers and the like) but may keep dates, city, state, the full ZIP code and ages. It may be used or shared only for research, public health or health care operations, and only under a data use agreement in which the recipient promises, among other things, not to identify or contact the people in it.
So the choice is: if the recipient needs dates and places and the project fits those purposes, a limited data set with a data use agreement; if the data must leave the rule entirely, Safe Harbor or Expert Determination.
What do candidates get wrong?
- "Encrypted data is de-identified." No. Encryption protects PHI, but it is still PHI, because whoever holds the key can read it. Encryption matters elsewhere: under the Breach Notification Rule, PHI encrypted according to HHS guidance is not "unsecured", so losing it is not treated as a breach of unsecured PHI and the duty to notify does not apply, as long as the decryption key was not lost or exposed with it. That still does not make the data de-identified. That is sometimes also called a safe harbor, a second, different use of the same words.
- Keeping the full date of birth, or the month of admission. Under Safe Harbor only the year may stay.
- Replacing names with initials. Initials are still part of a name and still identify.
- Treating a limited data set as de-identified. It is still PHI and needs a data use agreement.
- Asking the IT director to approve the dataset. Expert Determination needs a person with statistical expertise and a documented analysis, not someone's general sign-off.
Summary for revision
- Safe Harbor: remove 18 identifiers (of the person, relatives, employers and household members), with no actual knowledge that the rest identifies anyone.
- Dates: only the year may stay; for anyone over 89 the year goes too, and ages may be grouped as 90 or older.
- Expert Determination: a statistical expert finds the risk very small and documents the methods and results; details such as dates may stay.
- Limited data set: still PHI, only for research, public health or health care operations, under a data use agreement.
- Encryption is not de-identification: it matters for breach notification, not for whether the data is PHI.
Sources
- HIPAA Privacy Rule, 45 CFR 164.514: de-identification (a)-(c) and the limited data set (e).
- HHS Office for Civil Rights, "Guidance Regarding Methods for De-identification of Protected Health Information in Accordance with the HIPAA Privacy Rule" (2012).
- HHS guidance on rendering PHI unusable, unreadable or indecipherable (breach notification).
Related free practice questions
- De-identification by the Safe Harbor method
- The presumption of a breach, rebutted by a low probability of compromise
The site files this concept under the domain Healthcare Information and Systems Management.
Try exam-style questions
A 20-question trial exam with an explanation for every option and a score per domain, drawn from 40 free questions. No payment method needed.
Start the trial examQuestions about the exam: CPHIMS guide · What is CPHIMS? · Exam cost · Eligibility · Passing score · Renewal · Exam day · How long to study · Is it worth it? · What is CAHIMS?
Concepts explained: HL7 v2 vs FHIR · Capitation vs fee-for-service · Incident vs problem vs change · Backup types, RTO and RPO · Privacy Rule vs Security Rule · CDS five rights · Lewin vs Kotter · SDLC and go-live strategies · Testing types · Earned value (SPI, CPI) · Structure, process, outcome · HIE models · Six Sigma vs Lean · Levels of interoperability · Bandwidth vs latency · EMTALA · Risk responses · RACI matrix · Build vs buy · Sunk vs opportunity cost · Data governance
The four domains: Healthcare and Technology Environments · Clinical Informatics · Healthcare Information and Systems Management · Management and Leadership
Written by Adel Moustafa, About us
Written and checked against the sources above on 22 October 2026. An independent site, not affiliated with or endorsed by HIMSS. CPHIMS® is a registered trademark of HIMSS.