deterministic · probabilistic · review queue
Charles Trinh — Patient Matching | Las Vegas, NV · Las Vegas, Nevada · Henderson · Clark County, Nevada · North Las Vegas, Nevada

Patient matching decides whether two records belong to one person. The cost of being wrong is asymmetric, and most systems are tuned as though it were not.
What this is
Matching is the operation every other healthcare data function depends on. If the index cannot tell that the Jane Doe at one hospital is the Jane Doe at another, the history is split and the clinician works from half a picture. If it decides two different Jane Does are one person, two people's records are merged and the error is a privacy incident rather than a data-quality one.
Deterministic matching uses an identifier that should be unique: a member ID, a medical record number within an organisation, a government identifier where lawfully held. Probabilistic matching compares a set of demographic fields with weights derived from how discriminating each field is, and produces a score. Production systems use both, deterministically first and probabilistically for the remainder.
The threshold is a policy decision disguised as a technical one. A low threshold merges more records and creates false matches; a high threshold leaves duplicates and fragments the record. The right answer depends on what the data is used for — a claims analytics pipeline tolerates more uncertainty than a clinical summary — and it should be written down rather than left at a default.
Most of the practical gain comes from data quality, not algorithms. Normalising names, parsing addresses into components, standardising dates and capturing a verified phone number move the match rate more than switching from one similarity function to another. That work is unglamorous and it is where I start.
Matching also has to survive the patient. Names change, addresses move, a family shares a phone number and a nickname appears in one system but not another. A matcher tuned on a static sample looks excellent and then degrades as the population moves. Measuring match quality on a rolling sample, rather than once at go-live, is what keeps the threshold honest.The clinical consequence of a threshold is not abstract. A duplicate record means a clinician sees half the medication history; a false merge means they see someone else's. When I set a threshold I state which of those two failures the organisation would rather have, because there is no setting that eliminates both, and a threshold nobody chose is a threshold nobody can defend.
How I do it
The first step is measurement on real data: sample pairs reviewed by a human, a measured false-match and false-non-match rate, and a written threshold with its rationale. Without that baseline, every later change is an opinion.
Then the ambiguity band. Scores between the automatic-match and automatic-reject thresholds go to a review queue with an owner and an SLA, because the ambiguous cases are exactly the ones a machine should not decide silently.
Finally the surrounding controls: a merge audit trail, an unmerge path, and a test that a merge can be reversed without losing the surviving record's history.
The measurement is repeated on a schedule with a fresh human-reviewed sample, and the threshold is adjusted in writing when the data drifts. A matching programme without that loop is a one-off project pretending to be a control.The output is not only a threshold but a rules document: which fields are compared, how each is normalised, which identifiers bypass scoring entirely, and what a reviewer does with a borderline pair. That document is what lets a second analyst reproduce the decision, and reproducibility is the property auditors actually test for.
Artifacts you keep
- Measured match and false-match rates on a reviewed sample
- Written thresholds with the rationale and the review band
- A review queue with an owner and a service-level target
- Data-quality fixes: name parsing, address components, date formats
- A reversible merge record with an audit trail
The asymmetric error nobody budgets for
Patient matching produces two failure modes and the industry tracks the cheaper one. A false negative creates a duplicate record: the history fragments, a second workup happens, and the cost is visible in imaging orders and repeated labs. A false positive merges two people into one chart.
The false positive is worse and it is structurally worse. No downstream security control detects it, because nothing was breached — the data is simply wrong in a location that looks authoritative, presented to a clinician making a decision with full confidence behind it. Reliability engineering has the right framing: a silent incorrect result costs more than a loud failure.
That asymmetry should set your thresholds, your review staffing, and your release gates. A matcher tuned to maximise match rate is a matcher tuned to merge, because merging is what raises the rate and the cost of the mistake is paid by somebody later in the process.
Build unmerge before merge
Most systems merge and few can separate cleanly. If a false positive is discovered after the records are combined and the system cannot undo it without destroying the audit trail, the organisation converts a correctable error into a permanent one and then chooses to live with it rather than admit it.
The unmerge path requires merge history: what was combined, when, by whom, on what basis, and which fields came from which source. Without it, separation is a manual reconstruction and nobody will authorise it. With it, unmerge is a transaction.
Test it before you need it. Merge two synthetic records, confirm the history entry, then unmerge and confirm both original records are intact with their provenance. An unmerge path that has never been exercised is a feature request waiting for an incident.
Four disciplines that separate good implementations
A deterministic tier that runs first and is trusted absolutely — where a verified identifier exists, that identifier decides and no probabilistic score overrides it. A probabilistic tier that logs the attribute contributions behind every score, because the score will be challenged and a black box cannot survive the challenge.
A human review queue for the middle band, staffed by someone who can see clinical context rather than a clerk comparing two forms, with a service level attached. A queue that sits for a week is a duplicate record with a head start.
And measurement per source rather than in aggregate. A blended match rate hides the single interface or registration site producing the problem, and aggregate dashboards will never localise it for you.
The one ratio worth tracking
Queries issued against responses with clinical content, reported per source system and per month. A low ratio is a matching problem, a data-quality problem or a participant-enablement problem, and per-source reporting tells you which within an afternoon. Aggregate reporting never will.
Pair it with duplicate rate per source and review queue age. Those three numbers describe whether matching is working, where it is failing, and whether the human safety net is keeping up with the volume the thresholds produce.
What to ask a matching vendor
Show one match decision with the attributes, weights and score that produced it, from your data rather than theirs. Give the false-positive rate separately from the false-negative rate. State the review queue volume per ten thousand records at the recommended threshold.
A vendor who answers all three has deployed this. A vendor who answers none has built a demonstration, and the demonstration will be indistinguishable from a working system until the first patient disputes a merged chart.
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 — patient matching — the scope, the failure modes it prevents and how the work is measured. It sits in the Identity and Consent 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.