Did the risk analysis cover all of it?
45 CFR 164.308(a)(1)(ii)(A) sets the duty. A covered entity or business associate must assess thoroughly the risks to the electronic protected health information it holds.
Healthcare. Healthtech. Patient data.
Billing tables, research extracts and support exports can carry patient data out of the clinical system. IRIS finds it in the stores you connect.
Four paid engagements and three proofs of concept.
45 CFR 164.308(a)(1)(ii)(A) sets the duty. A covered entity or business associate must assess thoroughly the risks to the electronic protected health information it holds.
Under the Safe Harbor method, 164.514(b)(2)(i)(C), a dataset still holding birth dates or full admission dates isn’t de-identified, whatever its file name says.
Under GDPR Article 9 and UK GDPR Article 9, health data is a special category, so each use outside care needs one of the conditions Article 9(2) lists.
Records for a provider’s billing, research and support stores, with one candidate rejected and one imaging archive left out.
| Source | Location | Detected type | Why | Sensitivity | Decision |
|---|---|---|---|---|---|
| Billing database | billing.claims.patient_ssn | US Social Security number | Patient identifiers kept on every claim line for insurer matching | Restricted | Keep; limit reads to the billing team |
| Research extract (SharePoint) | Research/Cardiology/cohort_2026.xlsx | Date of birth | Named “deidentified”, but full birth dates remain, so Safe Harbor isn’t met | Restricted | Reduce dates to the year, then re-scan before it is shared |
| Patient portal database | portal.intake.medical_conditions | Health information (medical record) | An intake-form column copied from the clinical record | Restricted | Keep; check that the analytics role can’t read it |
| Support tool export (Amazon S3) | support/exports/patients_2026-10.csv | Date of birth and email address | A patient contact list exported to match callers to records | Confidential | Delete the export; the support owner keeps lookups in the portal |
| Scheduling database | scheduling.slots.booked_on | Date of birth (candidate) | Past dates in a slot table, read as birth dates | None | Rejected by the reviewer: booking dates, not birth dates |
| Imaging archive | Medical images and their headers | Not scanned | Not connected for this scan; the imaging team owns the archive | Unknown | Excluded: your team records it as a gap for the imaging team to assess |
In the US, HHS’s 2010 risk analysis guidance has the analysis cover all electronic protected health information, whatever the medium, beginning with where it is stored, received, kept or sent. The Safe Harbor method names the identifiers to remove, from (A) to (R).
In the EU, health data falls under GDPR Article 9, and in the UK under UK GDPR Article 9. India’s DPDP Act sets no separate category for health data, so its sensitivity has to come from your own classification.
Our position: run the scan before the risk analysis, and give the analysis the coverage record as its list of where patient data was looked for.
“A joined customer profile can put medical history in front of a payments team,” Dipal Panchal wrote in a published post. Check which teams a joined profile reaches.
Shows the roles that can read each table holding health information, from table and view grants, so a payments role on a clinical table stands out.
Before a research or vendor extract is shared, check it against the Safe Harbor list for dates, ZIP codes and record numbers left behind.
Classifies the extract in its folder and names each identifier type it finds there, so a date-of-birth column is caught before the file is sent.
Classifies the databases and warehouses beside the clinical record, where patient identifiers and health details can be copied for billing and analytics.
Text recognition reads scanned PDFs and images in English and Hindi, so a faxed referral saved as a PDF is read and classified, and an unreadable page is listed with the reason.
Checks sampled values for encryption, masking or hashing, once the check is activated on your install, as evidence for the addressable encryption specification in 164.312(a)(2)(iv).
IRIS can be deployed in your own infrastructure, on-premises or in any cloud, so your data stays within your environment.
Capabilities vary by edition.
IRIS finds where patient data sits and who can reach it. It doesn’t de-identify a dataset, decide who has a treatment need, or write the risk analysis; your privacy and security officers do.
It reads only the stores you connect, so an imaging archive or a hosted clinical system left unconnected stays outside its record. Name those as gaps in the analysis, so its scope is stated in writing. Each scan is a point-in-time reading, and IRIS reads a sample of each table, and small tables in full, so the analysis cites the scan dates it relied on.
What each law asks of patient data held outside the clinical record.
IRIS scores HIPAA, GDPR and India’s DPDP Act. For UK GDPR, UAE PDPL and Saudi PDPL, IRIS maps its evidence to their controls. A scored view for these in healthcare isn’t in v1; ask us if you need one.
Start where patient data can be copied for work outside care: the billing database and the research share. The how-it-works page follows that scan to its findings and to the coverage record your risk analysis can cite.
Yes, on a database engine IRIS supports, where your vendor allows a read-only account. You can also start with the copies around it, as the table above does.
No. It names the identifier types it finds, which shows where an extract falls short of Safe Harbor. The de-identification decision stays with your privacy officer.
Yes. HIPAA is one of the six frameworks IRIS scores, and your auditor judges whether each control is met.
Name the billing database your privacy officer worries about. We’ll show what IRIS finds there and who can reach it.