Clinical and payer systems, where the record is legally significant, interoperability is decades of overlapping standards, and interrupting a clinician has a real cost.
19 published across 7 topics.
The record model, master patient index, identity resolution, and merge reversibility.
Patient matching across healthcare systems uses deterministic rules, probabilistic linkage and human review to connect records without a shared identifier. Blocking, thresholding and reversible links matter because false merges are clinically riskier than false splits. It also connects record linkage to the point an interviewer is testing.
Not as one field. Administrative sex, recorded sex at birth, gender identity, name and pronouns to use, and the sex parameter a clinical calculation needs are separate elements with separate provenance and separate consumers, and each has to be able to change without rewriting the others.
Containment first, because clinical decisions are being made on a chart holding two people's data. Then separate the merge itself from everything it caused - the notifications sent onwards, the orders, results and doses now attributed to the wrong person - and treat unmerge as a clinical exercise with human adjudication.
HL7 v2, FHIR, CDA, integration engines, profiles and extensions.
FHIR is a resource-oriented REST API with a defined set of resources, constrained for real use by profiles and implementation guides. HL7 v2 is the older pipe-delimited messaging standard that still carries most of the event traffic inside hospitals, so the two coexist and an integration engine translates between them.
Because a standard constrains structure and leaves the meaning local. Optionality, local identifiers and code lists, custom extensions and different workflow triggers mean conformance to the standard is not compatibility - which is why the fix is your own profile plus a test suite, not more mapping.
A code means nothing without its code system, its version and a mapping the receiver trusts. Local catalogues dominate, LOINC, SNOMED CT and ICD answer different questions, value set expansions drift with versions, and every local-to-standard map is lossy, owned and perpetually going stale.
Orders and results, e-prescribing, clinical decision support, and alert fatigue.
An order is a request with an owner, a state machine and a required acknowledgement of its result, and an alert on that pathway earns attention only if it is specific, actionable, delivered to whoever can act, and rare enough that firing means something. Measure override rates and retire alerts that fail.
Plan for four modes rather than one: a continuously fed read-only viewer on independent infrastructure, printed ward-level essentials, paper ordering with pre-allocated temporary identifiers, and a recovery process that backloads events at their original times and reviews everything that queued during the window.
Treat the paper list as the requirements document you never got. It is holding something the system cannot express - ordering by clinical worry, tasks with no order behind them, the sentence explaining why a patient is being watched - and the hazard is that it is private to one person.
Diagnosis and procedure coding, charge capture, and the revenue cycle.
When a diagnosis code set changes mid-year, healthcare billing systems must select the version by date of service, then update payer rules, authorisations, contracts and reports that copied those codes. Trend data needs an explicit bridge.
Documentation of the visit is translated by coders into diagnosis and procedure codes, charges are captured against those codes from the charge master, and the revenue cycle then prices, submits, tracks and posts the resulting claim until the account is closed. Each step depends on the documentation being specific.
Only the record can settle it. Rule out mix, population and grouper changes, then run a blind re-code of a stratified sample, report disagreement rate and direction, attribute each disagreement to coder, documentation, template or incentive, and fix the cause rather than the codes.
Claim submission and adjudication, eligibility, prior authorisation, and denials.
Separate the adjudication engine from effective-dated, immutably versioned configuration, resolve config by date of service rather than by today, hold accumulators as a reversible ledger of contributions, adjust decisions by replacement rather than edit, and release config through a replay harness.
Adjudication validates the member and provider, applies benefit and contract rules, prices the claim and returns a remittance explaining each line. Denials are prevented at the front end by eligibility and prior authorisation checks, and managed at the back end by triaging root causes rather than reworking claims.
Make every hop acknowledge itself and reconcile those acknowledgements actively - envelope receipt, file-level acceptance, then a claim-level acknowledgement from the payer. Until the last of those arrives, the claim is sent but not received, and the difference is a timely-filing denial.
Protected health information, minimum-necessary access, consent, and audit obligations.
You do not delete first. Erasure is a qualified right that yields where other law requires retention, so you triage each dataset by its lawful basis, erase what has no basis to be kept, suppress and restrict what must be retained, and put the residual judgement to legal and records management.
Role-based access permits this by design, so the control cannot be authorisation alone. You need a recorded relationship between clinician and patient, a break-glass path that costs something to use, read events in the audit log, and someone whose job is to review the exceptions.
Clinical trial data, validated systems, and the evidence a regulated change requires.
A change to a validated system needs a documented requirement, an assessed risk and impact, approved test evidence that the change does what was specified, an intact audit trail over any affected data, and an approval record showing who authorised release. Working code is not evidence.
The worksheet has become source data, so it must be retained, attributable and legible, and the electronic entry must carry the date of the visit separately from the date of entry. What makes the data usable is the audit trail and the query history, not the value in the field.