What is HIPAA Expert Determination?
HIPAA Expert Determination is one of two ways to de-identify protected health information (PHI) under the HIPAA Privacy Rule. A person with knowledge of statistical and scientific methods analyzes the data and finds that the risk of identifying anyone, alone or combined with other available information, is very small for the intended recipient. The expert documents the methods and results.
For example, a hospital network wants to share two years of patient encounters with a research partner studying seasonal admissions. The Safe Harbor method would cut every admission date down to the year, which ruins the study. Instead, the hospital hires an expert who shifts dates by a random offset per patient, groups rare diagnoses and checks how unique each record is. The expert concludes the risk is very small for that partner and writes a report. This is a summary, not legal advice.
How Expert Determination works
- Define the data and the recipient. Risk depends on who receives the data and what other data they can reach.
- Measure risk. The expert looks at quasi-identifiers such as dates, ZIP codes, age and rare conditions, and at how unique each record is.
- Reduce risk. Techniques include generalization, suppression, date shifting, pseudonymization and adding noise, applied until the risk is very small.
- Document. The report records the analysis, the methods, the result and any conditions, such as limits on how the recipient may use the data.
- Revisit. The determination should be reviewed when the data, the recipient or the outside data available changes. HHS guidance allows experts to set a time limit.
Expert Determination vs Safe Harbor
| Aspect | Expert Determination | Safe Harbor |
|---|---|---|
| Method | Statistical risk analysis by an expert | Remove 18 listed identifiers |
| Data detail | Can keep dates, locations and other detail | Year-only dates, limited ZIP codes |
| Cost and time | Higher; needs an expert and a report | Lower; rule-based |
| Scope | Tied to a dataset and recipient | Holds wherever the rule is met |
HIPAA does not set a numeric threshold for “very small” risk, and it does not require a specific degree or certification for the expert. The expert’s experience and documentation carry the weight.
Why Expert Determination matters for AI systems
Model training and analytics often need the details Safe Harbor removes, such as exact intervals for time-series work or locations for regional trends. Expert Determination lets teams keep that detail under a documented risk analysis. Two cautions apply to AI projects. Free-text notes carry identifiers that are hard to measure, so they need careful detection first. And a determination made for one recipient does not cover a new use, such as sending the same data to a different model provider.
How Protecto helps
Protecto detects PHI in structured and unstructured data and replaces direct identifiers with consistent tokens, so the same patient keeps the same token across records and the data stays usable for analysis. Protecto also shows where sensitive data flows and produces reports on how data policies are enforced, which an expert can use as input to the analysis. See AI training data de-identification.
Frequently asked questions
Who qualifies as an expert under HIPAA?
A person with appropriate knowledge of and experience with generally accepted statistical and scientific methods for making information not individually identifiable. HIPAA does not require a specific degree or certification.
Does an Expert Determination expire?
HIPAA sets no expiration date, but HHS guidance allows experts to set one. Many determinations are tied to a dataset and recipient and are reviewed when either changes.
Is Expert Determination better than Safe Harbor?
HIPAA does not prefer either. Expert Determination keeps more detail but costs more. Safe Harbor is simpler and suits cases where dates and locations are not needed.
Related terms
HIPAA Safe Harbor method · Quasi-identifier · K-anonymity · Differential privacy · Protected health information (PHI) · Data de-identification