DATA
Charles Trinh — De-identification and Anonymisation | Las Vegas, NV · Las Vegas, Nevada · Henderson · Clark County, Nevada · North Las Vegas, Nevada

De-identification is a method, a measurement and a decision. The method is the transformation; the measurement is the residual risk; the decision is who accepts that risk. All three belong in the same document.
De-identification is not a checkbox
A file with names replaced by Patient_1 is de-identified under some methods and not under others, and the distinction matters to a legal reviewer. Safe Harbor under HIPAA requires removing a specific set of identifiers, including dates more specific than the year and every geographic unit smaller than a three-digit ZIP prefix. Expert determination requires a qualified person to conclude that the risk of re-identification is very small. A tool that does not know which method it is applying produces neither; the output looks identical either way while meeting the standard of neither.
I start every de-identification project with the method decision written down: which framework applies, which identifiers are being handled, what the residual risk will be measured against, and who accepts it. The method is decided before the pipeline is written. When the method is ambiguous, the file is treated as not de-identified until someone with the right authority says otherwise. This ordering prevents the common failure where a team publishes data under one framework and an auditor tests it against another, which is a finding that arrives after publication and is hard to correct.
The artefact I produce is the de-identification report: the method, the pipeline that applied it, the residual risk measurement, the acceptance decision with a name and a date, and the re-identification controls that remain. That report is the document an external reviewer, an institutional review board, or a regulatory audit expects to see. It also names the re-identification risks that cannot be removed and the controls that mitigate them, so a reviewer reads the residual as a known quantity rather than discovering it and concluding the process was careless.
Structured, unstructured and free text
Clinical data breaks into two main shapes: structured codes and free text. Structured codes are easier, but only because the risks are more visible. A patient identifier, a date of birth, a full postcode, an encounter identifier, and a device serial number are the obvious ones. A facility code inside a rural setting, or an unusual diagnosis code that identifies the patient in a small population, is the less obvious one. I classify each structured field by identifier category and apply the method's rule to the field, not to the table.
Free text is harder and the common failure is a human-made regex that looks plausible. The pipeline must handle notes, discharge summaries, radiology reports and letters, with the variations in punctuation, abbreviations, spelling and case that clinical text uses. I use statistical and dictionary-based processing in combination, with a manual review sample of every extract. The sample is large enough to estimate the residual leak rate, and small enough to review in a day. The leak estimate is reported even when it is zero; an unmeasured zero is not a result the reviewer can trust.
For structured FOMs the US Core conformance work and the C-CDA exchange work define the field boundaries. I use those rather than the source database layout, because the FHIR and C-CDA field models are designed with privacy in mind and the source layout often is not. Handling by resource type also makes the de-identification pipeline portable across source systems., which is the same standard applied to the structured side of the same extract.
Dates, geography and rare events
Dates are consistently underhandled. Replacing all dates with a random value destroys temporal relationships and reduces the dataset's research value, so the common method is shifting per patient. Each patient gets a random offset within an agreed range and all of their dates shift by that offset. The method is publicly documented, the offset range is reported, and the date shift itself is stored securely so a re-identification request with authority can reverse it.
Geography needs care. Replacing each postcode with a three-digit prefix is a safe-harbor-aligned method, but a small population in a rural area becomes unique. In those areas I roll up further, or suppress rows below a threshold. The threshold is chosen against the actual population density around the address, not a global constant. The same postcode prefix can safely identify thousands in an urban area and a single family in a rural one.
Rare events need the same test. A patient with a rare diagnosis, a rare procedure and a specific age band can be re-identified even with every identifier removed, because the combination is unique in the source. I measure the uniqueness of the de-identified dataset against the source, quantify the residual risk per row, and where the residual is above the acceptance line, I apply additional measures: further aggregation, category collapse, or suppression of the affected rows. The residual risk assessment records which measures were chosen and why.
Measuring residual re-identification risk
Residual risk is a probability, and I estimate it. The framework differs by population and by recipient, but the shape is: what does an attacker with access to auxiliary data know, and how many people does that match in the de-identified set. For a public research release, the population is everyone and the auxiliary data is public. For a limited-recipient research extract, the population is the recipient and the auxiliary data is what the recipient has. The threshold I accept differs between the two, and the de-identification report says which threshold applies.
I compute uniqueness per row against the source population and report the proportion of rows above the threshold. Above the threshold, the report names the measures taken. Below the threshold, the report says the method is acceptable. The decision to accept the residual is made by the sponsor of the release, in writing, with a named person and a date. A pipeline that measures risk but does not name who accepted it leaves the decision implicit and the release unowned.
The measurement runs against the current source population, not a historical sample, because a piece of the population that used to hide an individual can stop hiding them if the population grows or shrinks. I re-run the measurement whenever the extraction scope changes or the source population changes by more than a threshold. A dataset de-identified in 2024 against the 2024 population may not meet the 2026 standard if the population has moved.
The pipeline as an artefact
The de-identification pipeline is code, in a repository, with a version and a test harness. The code handles each resource type, each identifier category, and each method. The tests are synthetic inputs with known identifiers and known safe outputs. The pipeline is run by the team that owns the source, not by the analyst who requested the extract, which separates the requester from the transform.
Every run produces a manifest: the source snapshot, the code version, the method parameters, and the outputs. The manifest is signed and stored with the outputs. When an external reviewer asks whether a specific dataset was processed by a specific method, the manifest answers directly. When a question arrives about a dataset from two years ago, the manifest and the code are what allow reproduction without calling the person who ran it.
Re-identification requests (when a recipient needs to trace a record back, under appropriate authority), reversal of date shifts, and access to source data all go through the same governance route. The healthcare data governance work names who approves and what is recorded. The de-identification pipeline keeps the reversal keys in a separate system from the de-identified outputs. If someone has both the dataset and the keys in the same place, the de-identification is a label, not a control.
When de-identification is not enough
Some data cannot be safely de-identified for the intended release. Very rare diagnoses, small populations, unusual demographics and specific geography can each make a release unsafe. The de-identification project concludes "safe for this release" or "not safe for this release"; it does not conclude "safe in general". When the data is not safe for the intended recipient, I say so in writing and offer the alternatives: aggregate tables, a restricted-access environment, or an on-site analysis under supervision.
Restricted-access environments are often the right answer for rare data and research that needs record-level detail. The data is not extracted; the analysis is brought to the data, under governance controls that include the analysis being reviewed before release. The environment itself is a governance artefact: access logs, code review and output review. The alternative of "extract and promise" is what got the industry into trouble in several well-documented incidents, and it is the option I avoid proposing.
The de-identification pipeline is also useful for internal use where the team building the analytics does not need identifiers. A test environment, a development sandbox, a shared analytics warehouse all benefit from the same treatment. Extending the pipeline to internal environments is a common request once the first project lands, and I build it with that extension in mind from the start. Internal and external pipelines share the same code and the same tests.
Proof
- ONC (g)(10) self-test PASS, 317 of 317 subtests, April 2026 — public suite URL
- 67 public hostnames, 51 serving, one AWS account
- 104 clearinghouse enrolments across 53 jurisdictions
- USCDI v7 proposal — Patient Digital Identity Credential — listed at isp.healthit.gov
- SAM.gov UEI MVKQW9DKKGB8 · CAGE 20KM5 · NPI 1194645564
- AWS Business Associate Addendum active since 18 August 2026
Where this work is delivered
Engagements run from Las Vegas and reach across the valley: clients in Las Vegas, NV; Las Vegas, Nevada; Henderson; Clark County, Nevada; and North Las Vegas, Nevada — plus remote delivery nationwide.
How an engagement runs
1. Trace one real case
Follow a single patient or claim end to end with a timestamp at every hop, before any document is written.
2. Written findings
A findings list naming the defect, the system, the owner and the date.
3. Monthly review
Review the list monthly and check whether the dates held.
Frequently asked
What is this service?
This page describes charles trinh — de-identification and anonymisation — the scope, the failure modes it prevents and how the work is measured. It sits in the Security and Compliance category at trinh.io.
Where is the work delivered?
Engagements run from Las Vegas, Nevada across the valley — Las Vegas, Henderson, North Las Vegas, Spring Valley, Paradise, Winchester, Sunrise Manor, Enterprise, Summerlin, Clark County — and remotely anywhere in the United States. On-site work in the Las Vegas valley, remote delivery elsewhere.
What does it cost?
Standard rate $195 per hour, committed $165 per hour at twenty hours or more per week for three months or more; $1,600 standard day, $1,350 committed day. Scoped in writing before work starts, fixed fee where the deliverable is defined.
Who is Charles Trinh?
A technology architect in Las Vegas, Nevada, working in healthcare identity, consent and interoperability. Not a physician, not a medical practice, and not affiliated with any other Charles Trinh.
How does an engagement start?
With one trace of a real case and a written findings list naming the defect, the system, the owner and the date. The first call is twenty minutes and free: 702-559-3489.
Engagement
Scoped in writing before work starts; fixed fee where the deliverable is defined. First call is 20 minutes and free. 702-559-3489 · charles@trinh.io
Talk to Charles
- Phone: 702-559-3489
- Email: charles@trinh.io
- Based in Las Vegas, Nevada · working remotely across the United States
Record
Profile
Healthcare technology architect working where federal regulation becomes production infrastructure. Built and operates a FHIR R4 clinical platform — SMART App Launch authorization server, patient and provider portals, SQLite — on a hardened single-host AWS estate of 67 public hostnames. The platform returned a self-test PASS on the ONC (g)(10) Standardized API test suite in April 2026, 317 of 317 subtests, publicly verifiable. Designed a consent architecture in which a permissioned ledger records patient grant, refusal and revocation as immutable events while all clinical data stays off-chain and encrypted per patient — no PHI on the ledger by design. Filed a proposed USCDI v7 data element with ONC on verifiable patient identity. Translates the Federal Register, HHS/CMS guidance and HITRUST requirements into architecture, controls and audit evidence.
Selected engineering scale
- Self-managed estate of 67 public hostnames, 51 serving, on one AWS account.
- FHIR R4 API with a SMART App Launch authorization server; patient and provider portals; an AI service layer; Hyperledger Fabric peer, orderer and CA; IPFS; a 337-table SQLite clinical store.
- Two permissioned distributed ledgers in production, one operating a consent registry that records grant, refusal and revocation with no PHI on chain.
- X12 005010 pipelines across eight transaction sets.
- Designed, built, deployed and operated end to end by one person.
Credentials and standards work
- Inferno (g)(10) self-test PASS, 317 of 317 subtests, April 2026 — the ONC Standardized API test suite. Standards validated: US Core 6.1.0, USCDI v3, SMART App Launch 2.0.0, Bulk Data 2.0.0. Public suite URL.
- USCDI v7 public comment submission, April 2026 — proposed Patient Digital Identity Credential under the Provenance data class, listed publicly on the ONC Interoperability Standards Platform.
- SAM.gov UEI MVKQW9DKKGB8 · CAGE 20KM5 · NPI 1194645564. AWS Business Associate Addendum active 18 August 2026.
Technical environment
AWS us-east-1 (compute, storage, messaging, SMS, email, voice) · Linux and systemd hardening · nginx · Fabric and EVM ledgers · FHIR R4 · X12 005010 · HL7 · SQLite · self-hosted mail, drive, search and monitoring · Git-based change control and CI release gates.