Does removing names de-identify patient data for AI?

By Verillian. Published . Updated .

No. HIPAA recognizes two de-identification methods, and a name is one of eighteen identifiers. Here is what each method requires.

A clinician in scrubs, her mask pulled down, works at a computer in a clinic room.

No. Taking the patient’s name out of a prompt does not make it de-identified under HIPAA. The Privacy Rule recognizes two methods. Safe harbor requires removing all eighteen listed types of identifier, of which a name is only one, and also that you have no actual knowledge the rest could still identify the person.

Expert determination requires a qualified expert to determine, and document, that the risk of identification is very small. Text that meets neither is still protected health information.

What does HIPAA mean by de-identified?

The standard is at 45 CFR 164.514(a) and (b). Health information that “does not identify an individual and with respect to which there is no reasonable basis to believe that the information can be used to identify an individual” is not individually identifiable health information.

A covered entity may determine that health information meets that standard only by one of two methods. The reason to use either one is set out in 164.502(d)(2). The requirements of that subpart do not apply to information that has been de-identified in accordance with 164.514. Those requirements still apply to information that does not meet the de-identification standard.

What does the safe harbor method require?

Under 164.514(b)(2), eighteen types of identifier of the individual, and of their relatives, employers, or household members, must all be removed. In addition, the covered entity must not have actual knowledge that the information could be used, alone or in combination with other information, to identify the person. The eighteen are:

  • names;
  • all geographic subdivisions smaller than a state, including street address, city, county, precinct, zip code and equivalent geocodes, except that the first three digits of a zip code may stay if the area those three digits cover has more than 20,000 people (and is otherwise changed to 000);
  • all elements of dates, except year, that relate directly to the individual, including birth, admission, discharge and death dates, and all ages over 89 and all date elements, year included, that indicate such an age, though ages over 89 may be grouped as “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 addresses (URLs);
  • IP addresses;
  • biometric identifiers, including finger and voice prints;
  • full face photographs and comparable images; and
  • any other unique identifying number, characteristic, or code, except a code permitted by 164.514(c).

Now read a prompt against that list. Consider an invented, illustrative line about a ninety-one-year-old admitted on March 3 after a fall at home, whose daughter Maria, in the same town, will collect her. The patient’s name is gone, and the line still carries an age over 89, an admission date, a relative’s name, and a town.

Clinical prompts are mostly free text, and free text is where identifiers hide, in dates, in places, in the names of relatives and employers, and in the last category, any other unique characteristic. Removing the patient’s name is one category out of eighteen. The others take a review of the whole text.

What is expert determination?

The other method, 164.514(b)(1), requires a person with appropriate knowledge of and experience with generally accepted statistical and scientific principles and methods for rendering information not individually identifiable.

That person, applying those principles and methods, must determine that the risk is very small that the information could be used, alone or in combination with other reasonably available information, by an anticipated recipient to identify an individual who is a subject of the information.

The expert must also document the methods and results of the analysis that justify the determination.

That is an analysis with a written record, done by a qualified person about particular information and a particular recipient. Removing names is not that analysis, and it is not the safe harbor list either, so a prompt that has only had names taken out meets neither method.

Who can de-identify protected health information?

Under 164.502(d)(1), a covered entity may use protected health information to create de-identified information, and may disclose protected health information for that purpose only to a business associate.

Sending identifiable text to an outside service so that the service can strip it is a disclosure of protected health information to that service, which the rule permits for this purpose only to a business associate. Our post on whether ChatGPT is HIPAA compliant covers where one common tool stands.

What about replacing names with codes?

Section 164.514(c) allows a covered entity to assign a code so that de-identified information can be re-identified later, as long as the code is not derived from or related to information about the individual and cannot be translated to identify them, and the covered entity does not use or disclose the code or the mechanism for re-identification for any other purpose.

A random placeholder such as “Patient A” with a mapping table that stays inside the organization can fit that description. A placeholder built from a medical record number, or a hash of one, does not, and under 164.502(d)(2) disclosing a code designed to enable re-identification is itself a disclosure of protected health information.

What does it take to de-identify a prompt?

  • Choose the method first, either safe harbor, with every one of the eighteen types removed and no actual knowledge that the rest identifies the person, or a documented expert determination.
  • Review the whole text, not only the name. Dates, places, ages over 89, and the names and details of relatives and employers all count.
  • Use random placeholders, and keep any mapping table inside the organization.
  • Treat anything that has not met one of the two methods as protected health information. Whether pasting it into an AI tool is a breach is the question in our post on pasting patient data into an AI tool.

What does Verillian do with identifiers in a prompt?

On each enrolled device, the identifiers your policy flags are replaced before the request leaves when your policy says to redact, and the decision goes in the record. Either method above is a review of the whole text, so a redacted prompt still goes through that analysis.

Sources

  • 45 CFR 164.514, (a) and (b), de-identification standard and methods, and (c), re-identification codes.
  • 45 CFR 164.502, (d)(1) and (d)(2), uses and disclosures of de-identified protected health information.

How redaction and denial work on the device

See Verillian govern AI on a real device.