Glossary
Definitions as used in this knowledge base. Where a term is contested or used differently in different traditions, that is noted rather than resolved.
The first sections cover architecture, standards and security — the vocabulary of building systems. The AI section carries the vocabulary of model development, evaluation and post-deployment assurance, in the stricter sense regulators use it. The later sections cover care delivery, virtual care, quality measurement, payment and operations — the vocabulary of the people who will decide whether those systems are bought, adopted and paid for. A good number of the payment and programme terms are specific to the United States and are marked as such; the mechanism underneath usually generalises even where the programme does not.
Systems and records
EMR — Electronic Medical Record. The digital record within one organisation. Its scope is that organisation's care.
EHR — Electronic Health Record. A record intended to span organisations and a person's lifetime. In practice the terms are used interchangeably by vendors; the distinction that matters architecturally is scope.
PHR — Personal Health Record. A record the individual controls, which may draw from provider records.
SHR — Shared Health Record. A longitudinal clinical store assembled from multiple point-of-service systems in a health information exchange. See business domain services.
HIS — Health Information System. Any system holding health information; often used loosely.
HMIS — Health Management Information System. Aggregate reporting and analysis for programme management. Frequently DHIS2.
RHIS — Routine Health Information System. The routine, usually monthly, data collection from facilities.
LIS / LIMS — Laboratory Information (Management) System.
LMIS — Logistics Management Information System. Supply chain. See supply chain.
PACS — Picture Archiving and Communication System. Imaging storage and retrieval. VNA — Vendor Neutral Archive, a PACS-independent archive.
RIS — Radiology Information System. Orders, scheduling and reporting for imaging.
POS — Point of Service. In OpenHIE, any system where care happens and data is created. See point-of-service systems.
CHW — Community Health Worker. See community health.
Exchange and interoperability
HIE — Health Information Exchange. Both the act of exchanging health information across organisations and the institution that enables it. See health information exchange.
IOL — Interoperability Layer. The mediating component through which point-of-service systems exchange. See interoperability layer.
OpenHIE — Open Health Information Exchange. A community reference architecture for national exchange. See OpenHIE.
OpenHIM — Open Health Information Mediator. The OpenHIE reference implementation of the interoperability layer. See OpenHIM.
ESB — Enterprise Service Bus. A mediating integration component; the enterprise-architecture ancestor of the interoperability layer.
ETL / ELT — Extract, Transform, Load (or Extract, Load, Transform). Pipelines moving data into an analytical store.
CDC — Change Data Capture. Streaming changes out of a source database as they occur.
IHE — Integrating the Healthcare Enterprise. An organisation publishing profiles: precise constrainings of standards for named workflows (PIX, PDQ, XDS, XCA, ATNA, MHD).
Standards
HL7 — Health Level Seven International. The standards body responsible for v2, CDA and FHIR.
HL7 v2. Pipe-delimited event messaging; still carries most hospital integration traffic. See HL7 v2.
CDA — Clinical Document Architecture. XML clinical documents. C-CDA is the US consolidated template set. See CDA.
FHIR — Fast Healthcare Interoperability Resources. HL7's resource-and-REST standard for health data exchange. See HL7 FHIR.
IG — Implementation Guide. A published constraining of FHIR for one context: profiles, value sets, capability statements. Nothing interoperates against "FHIR"; systems interoperate against an IG. See FHIR implementation guides.
Profile. A StructureDefinition constraining a resource — cardinality,
required elements, terminology bindings.
IPS — International Patient Summary. A minimal cross-border summary specification. IPA — International Patient Access, the equivalent for patient-facing app access.
SMART — Substitutable Medical Applications, Reusable Technologies. The SMART App Launch framework for authorising health apps. See SMART on FHIR.
CDS Hooks. A specification for calling out to decision services at defined points in clinical workflow.
DICOM — Digital Imaging and Communications in Medicine. The imaging standard. DICOMweb is its RESTful expression. See DICOM.
openEHR. A two-level modelling specification for persisting clinical data: a stable reference model plus clinician-authored archetypes. See openEHR.
OMOP CDM. The OHDSI common data model for observational research.
Terminology and classification
Terminology. A vocabulary describing clinical meaning for the record — SNOMED CT, LOINC.
Classification. A grouping into mutually exclusive categories for counting — ICD, ICF. You encode with a terminology and report with a classification.
SNOMED CT — Systematized Nomenclature of Medicine, Clinical Terms. The comprehensive clinical terminology.
LOINC — Logical Observation Identifiers Names and Codes. Identifiers for laboratory tests and clinical measurements.
ICD — International Classification of Diseases. WHO's classification; ICD-10 and ICD-11.
ICF — International Classification of Functioning, Disability and Health.
ICPC — International Classification of Primary Care.
RxNorm. US normalised drug terminology. ATC — WHO's Anatomical Therapeutic Chemical drug classification.
UCUM — Unified Code for Units of Measure. Computable units.
DRG — Diagnosis Related Group. Case-mix grouping used for payment.
Value set. The permitted codes for a coded element. Binding strength — required, extensible, preferred or example — states how strictly.
ConceptMap. A defined translation between code systems, with an equivalence assertion.
TS — Terminology Service. The component holding code systems, value sets and maps. See terminology services.
Registries and identity
CR — Client Registry. The authoritative list of clients and their identifiers.
MPI — Master Patient Index. The same function, in hospital-sector language. EMPI — enterprise MPI, spanning organisations. See client registry.
FR — Facility Registry. The authoritative list of health facilities. See facility registry.
HWR — Health Worker Registry. Providers, roles and qualifications. See health worker registry.
Golden record. The registry's consolidated view of an entity, assembled from source records by survivorship rules.
Deterministic matching. Exact agreement on defined fields. Probabilistic matching. Weighted scoring across fields, with thresholds and a human review band.
Merge / unmerge. Combining two records judged to be the same person, and reversing that when the judgement was wrong.
Security and identity
OAuth 2.0. The IETF authorisation framework: delegated, scoped API access via tokens.
OIDC — OpenID Connect. An authentication layer over OAuth 2.0. See OAuth and OIDC.
PKCE — Proof Key for Code Exchange. Binds an authorisation code to the client that requested it. Required for all clients in current practice.
SSO — Single Sign-On. MFA — Multi-Factor Authentication.
RBAC / ABAC / PBAC / ReBAC — access control by role, attribute, policy or relationship. Health almost always needs more than RBAC.
PDP / PEP — Policy Decision Point / Policy Enforcement Point. Where an access decision is made, and where it is applied.
Purpose of use. The declared reason for an access — treatment, payment, public health, research — recorded and auditable.
Break-glass. Emergency override of access restrictions, always available and always reviewed.
Provenance. The record of where data came from and who asserted it.
AuditEvent. The FHIR resource recording who accessed which data, when and why.
PKI — Public Key Infrastructure. TLS — Transport Layer Security. HSM — Hardware Security Module.
Zero trust. No network location is trusted; every request is authenticated and authorised on its own merits. See NIST SP 800-207.
PET — Privacy Enhancing Technology. Any technical means of reducing the privacy risk of processing data — secure multi-party computation, homomorphic encryption, zero-knowledge proofs, secure enclaves, differential privacy, federated learning, synthetic data generation. They change what has to be trusted, rather than removing the need for a lawful basis and a governance decision. See consent and trust.
SIEM — Security Information and Event Management. SOC — Security Operations Centre.
RTO / RPO — Recovery Time Objective / Recovery Point Objective. How long recovery may take, and how much data may be lost. In health these are clinical safety parameters.
Architecture
Reference architecture. A component-level design for a class of systems, technology-neutral. Solution architecture is one system in one context, and names technologies.
ADR — Architecture Decision Record. One decision, with context, options, choice and consequences. See ADRs.
C4. Describing software at four levels of zoom: context, container, component, code. See C4 model.
TOGAF / ArchiMate / Zachman. General enterprise architecture framework, modelling language and classification schema respectively. Not health standards. See frameworks.
RM-ODP. ISO/IEC reference model describing systems from five viewpoints.
CQRS — Command Query Responsibility Segregation. Separate write and read models.
Event sourcing. An append-only event log as the source of truth.
Saga. A sequence of local transactions with compensating actions, replacing a distributed transaction.
Anti-corruption layer. An adapter preventing one system's model from contaminating another's.
FHIR facade. A FHIR API over a system that does not store FHIR. See FHIR facade.
Offline-first. The application works fully without a network; connectivity is an opportunity to synchronise. See offline-first.
Store-and-forward. Queue locally, transmit when possible.
Degraded mode. Defined reduced functionality when a dependency is unavailable.
Data
Canonical model. The single agreed representation used for exchange.
Master data. The reference data everything else points at — the registries and terminology.
Data lineage. Where a value came from, through which transformations.
Data catalogue / data dictionary. Technical inventory of datasets; business definitions of data elements.
Operational data store / data warehouse / data lake / lakehouse. Current-state transactional store; modelled analytical store; raw file storage; the combination with transactional table formats over object storage.
Episode of care. The unit spanning multiple encounters — a pregnancy, a TB treatment course.
Denominator. The population a rate is computed against; usually the most contested number in any health report.
De-identification. Removing or obscuring identifiers, with residual re-identification risk that must be assessed rather than assumed away.
AI
Regulatory usage here is stricter than everyday usage. Where the two diverge the stricter reading is given, because it is the one that decides whether a product may be placed on a market and who answers for its output.
Systems and capability
AI — Artificial Intelligence. A machine-based system that, for a given set of human-defined objectives, produces predictions, recommendations or decisions influencing a real or virtual environment: it perceives, abstracts what it perceives into a model, and infers from that model. AI system is the engineered whole around it — the model plus everything that gets data in, output out, and a person accountable.
ML — Machine Learning. Techniques for improving performance at a task from data rather than from explicit instruction. See machine learning.
ML algorithm / ML model. The algorithm is the procedure that learns; the model is the resulting construct that produces an inference for new input. The distinction carries weight in regulation, where what is authorised is a model, not a method.
Model weights. The learned parameters in which the model's capability is held.
Deep learning. ML using neural networks with several hidden layers, each transforming its input into a more abstract representation than the last. Neural network: the layered arrangement of connected nodes this depends on.
CNN — Convolutional Neural Network. An architecture for grid-like data, images above all, in which filters slide across the input to detect local patterns while preserving spatial relationships. The workhorse of imaging AI.
GAN — Generative Adversarial Network. A generator and a discriminator trained against each other until the generator's output is indistinguishable from the training distribution. Used to synthesise images.
Foundation model. A model trained on very large, largely unlabelled data, applicable well beyond what it was specifically trained for, and intended as a base that others adapt by fine-tuning. LLM — Large Language Model: the language case, trained to predict and generate natural language.
Generative AI. Models that emulate the structure of their input data in order to produce new content — text, image, audio, video.
Multimodal. Processing several data types together — text, image, waveform, genomics, sensor data — because the relation between them carries information that none of them carries alone.
Assistive AI. The product informs; a person decides. Autonomous AI: the product decides and acts without human intervention. These are two ends of a spectrum rather than two categories, and where a product sits on it determines what evidence it needs and who is answerable for the result.
Learning approaches
Supervised learning. Training on labelled data. Unsupervised learning: training on unlabelled data to find latent structure. Semi-supervised: a small labelled set carrying a large unlabelled one. Self-supervised: labels derived from the structure of the unlabelled data itself — the approach that makes foundation models possible.
Labelling / annotation. Attaching descriptive information to data without altering the data. Usually the expensive part, and the part where clinical judgement actually enters the model.
Reinforcement learning. Learning from reward and penalty through interaction with an environment, balancing exploitation of what is known against exploration of what is not.
Transfer learning. Adapting a model built for one task to a second, related one, keeping what generalises. Fine-tuning is the usual mechanism.
Federated learning. Training across sites where raw data never leaves the site: each trains locally and only model updates are sent for aggregation. It removes the need to pool patient records; it does not remove the need for governance. See data governance.
Ensemble methods. Combining several models — bagging, boosting, stacking — to perform better than any one of them alone.
Feature engineering. Selecting, transforming and constructing the input variables that best represent the underlying pattern, using domain knowledge.
Locked model. A model whose parameters cannot change in use: the same input gives the same output, every time. Continual (adaptive) learning: a model with a defined process for changing its behaviour as new data arrives. The regulatory question is not which is better, but whether the change process was specified, bounded and validated in advance.
Data, evaluation and assurance
Training / tuning / test data. The data a model learns from; the data used to choose between architectures and hyperparameters; and the held-out data used to estimate performance afterwards. Test data must be independent of both others, or the reported performance means nothing. Note that regulators avoid calling the tuning phase "validation", which has a narrower legal meaning.
Reference standard (ground truth). The best available determination of the true state, against which model output is scored. Where that determination is itself a human judgement, its variability is part of the result and should be reported as such.
Performance metrics. Accuracy, precision, sensitivity, specificity, F1, AUC-ROC. Which of them matters is a clinical question, not a statistical one: the cost of a missed case and the cost of a false alarm are rarely equal.
Model fitting. Adjusting parameters until predictions approach the targets. Overfitting: learning the noise along with the pattern, so performance collapses on unseen data. Underfitting: failing to capture the pattern at all.
Model calibration. Whether predicted probabilities match observed frequencies — a model that says 20% should be right about one time in five. A model can discriminate well and be badly calibrated, and calibration is usually what a clinical decision actually rests on.
Model robustness. Holding the specified level of performance under noisy input, unseen data, different equipment and deliberate manipulation.
Data drift. Change over time in the distribution of input a deployed model receives — different demographics, different practice, different devices — degrading performance with nothing visibly failing. The general term drift covers change in input or output distributions after deployment.
AI performance monitoring. Regularly collecting and analysing data from the deployed system to detect degradation, drift, misuse and safety concerns. Its premise is that a model's accuracy is a property of a moment and a population, not a fixed property of the model.
Algorithmic bias. Systematic deviation in a model's output for some people or groups relative to others, arising from the training data, from development choices, from collection methods, or from the human decisions already recorded in the data. See AI ethics.
Explainability. A representation of the mechanism underlying a system's operation — answering why it produced this output. Interpretability: what the output means in the context of what the system is for. Related, routinely conflated, and not the same requirement.
Model card / data card. Structured disclosure documents: for a model, its characteristics, intended context of use and evaluation results; for a dataset, its composition, collection protocol and measured properties. The minimum an organisation should demand before deploying someone else's model.
Synthetic data. Data generated artificially to reproduce the structure and relationships of real patient data without containing any real person's information. Useful for development and for cohort assembly; not free of disclosure risk, which must be assessed.
Watermarking. Embedding a traceable pattern in generated output so its origin, provenance and subsequent modification can be verified.
Model registry. Versioned models with lineage, evaluation results and approval status — the operational record of which model is answering in production, and who signed it off.
In clinical use
SaMD — Software as a Medical Device. Software intended for a medical purpose in its own right, rather than as part of a hardware device. A regulatory classification that most clinical AI falls under once it does more than inform. See clinical AI.
Human-in-the-loop. The model proposes; a person disposes; the override is recorded. In its stricter ML sense the interaction is iterative and feeds back into the model — unlike supervised learning, where the human contribution ends at labelling.
AI-powered diagnostics. Models supporting or automating a diagnostic judgement.
Predictive analytics. Estimating a future event from historical data. Predictive risk modelling is the health-specific case: the likelihood of a readmission, a deterioration, a defaulted treatment course.
NLP — Natural Language Processing. Extracting structured meaning from unstructured clinical text — notes, discharge summaries, referral letters — and generating text in return.
Ambient clinical intelligence. Voice capture that documents the encounter in the background while the clinician attends to the patient. Speech-to-text is the narrower transcription component; an encounter summary is the generated output the clinician must still read and sign.
Conversational AI. Chat or voice agents handling triage, intake, scheduling or follow-up. The clinical risk sits in what they are permitted to decide, not in the language model.
RAG — Retrieval-Augmented Generation. Retrieving relevant documents and generating an answer grounded in them, with citations.
MCP — Model Context Protocol. An emerging protocol for connecting AI assistants to tools and data. Not a health interoperability standard — see MCP in healthcare.
Digital twin. A virtual construct mirroring the structure and behaviour of a physical thing — a patient, a device, a service — kept updated from it and used to inform decisions about it. The bidirectional link is the point; a static simulation is not a twin.
Health AI governance. The policies, approval gates, monitoring and accountability structures deciding which models may be deployed, on whom, and who answers when they are wrong. See AI ethics.
Care delivery
Terms from service delivery and care models. Several are named after United States programmes; where that is so it is said, because the concept usually travels further than the programme does.
Ambulatory care. Care without an overnight admission — consultations, tests, minor procedures. Outpatient care is used interchangeably. Inpatient care involves admission.
ASC — Ambulatory Surgery Centre. A facility for same-day surgical and diagnostic procedures.
Triage. Ordering patients by urgency rather than by arrival. Self-triage tools move the first step to the patient, which is a redistribution of clinical risk and should be designed as one.
Care coordination. Deliberately organising a patient's care activities across the people and organisations involved.
Care management. Ongoing support for patients — usually those with chronic conditions — intended to improve outcomes and avoid avoidable utilisation. Care navigation is the patient-facing half: helping someone move through a system that assumes they already understand it.
Care team collaboration tools. Platforms through which a distributed care team communicates about a shared patient. See community health.
Care gap. A recommended screening, immunisation or follow-up that is due and has not happened. Care gap closure is the work of finding and resolving these, and depends entirely on a trustworthy denominator.
Discharge planning. Preparing a patient to leave a facility with follow-up arranged. TCM — Transitional Care Management is the structured version across the days after discharge; its measure of success is the readmission that did not occur.
CCM — Chronic Care Management. Continuing, largely non-face-to-face support for patients with multiple chronic conditions. A US Medicare-reimbursed service, but a common care model everywhere.
Disease management programme. A coordinated approach to one condition across its whole course.
Collaborative care model. Team-based mental health care linking a primary care provider, a behavioural health specialist and a care manager. BHI — Behavioural Health Integration is the wider practice of delivering mental health care inside general care settings rather than beside them.
PCMH — Patient-Centred Medical Home. A model in which a primary care practice holds continuing responsibility for comprehensive, team-based care.
IDN — Integrated Delivery Network. A single organisation owning facilities across levels of care for a defined population.
ACO — Accountable Care Organization. A US grouping of providers jointly accountable for the cost and quality of a defined patient population, sharing in the savings. The structural idea — shared accountability for a population — is what generalises.
FQHC — Federally Qualified Health Center. US community providers serving underserved areas regardless of ability to pay. Safety-net providers is the general term for organisations carrying that role in any system.
Hospital-at-home (virtual ward). Hospital-level care delivered in the patient's home, with monitoring and a defined escalation route.
Low-acuity care pathways. Deliberately lighter workflows for presentations that do not need the full clinical apparatus.
Anticipatory care planning. Planning ahead for a person's expected health and social needs rather than responding after the fact.
Medication adherence. How closely actual medicine-taking follows what was prescribed. Medication reconciliation is the check that the list is the same across settings — the transition being where the errors are.
MTM — Medication Therapy Management. Pharmacist-led review to optimise a patient's drug therapy. Meds-to-beds delivers discharge medicines to the bedside before the patient leaves.
Smart medication packaging. Packaging with sensors that record when a dose was taken and can prompt when one was not.
Evidence-based guidelines. Clinical recommendations derived from systematic review of evidence. Clinical effectiveness is how well an intervention works in ordinary practice, as distinct from how well it worked in the trial.
Credentialing and privileging. Verifying a clinician's qualifications, and authorising them for specific procedures at a specific facility. See health worker registry.
Virtual care and remote monitoring
DHT — Digital Health Technology. A system using computing platforms, connectivity, software or sensors for health purposes. Deliberately broad: it spans general wellness through to regulated medical devices, and covers technologies used as a medical product, in one, or alongside one — including those used to run clinical research. The regulatory weight falls on intended use, not on the technology.
Virtual care. Any care delivered remotely — video, telephone, chat, secure messaging, asynchronous review. Telehealth is used interchangeably; telemedicine usually means the narrower clinician-to-patient consultation.
Telehealth parity laws. Legislation requiring remote care to be reimbursed at the same rate as the in-person equivalent. Where it does not exist, reimbursement, not technology, is what limits virtual care.
RPM — Remote Patient Monitoring. Physiological data collected from the patient outside a facility and reviewed by a care team. RTM — Remote Therapeutic Monitoring covers adherence and therapy response rather than physiology.
mHealth — Mobile Health. Use of mobile devices for care delivery and public health. In low-connectivity settings this is usually offline-first by necessity.
Wearables. Consumer or clinical devices producing continuous signals — heart rate, activity, sleep, glucose.
Digital biomarkers. Quantifiable, device-derived measures used as indicators of a physiological or behavioural state.
Home-based diagnostics. Testing performed by the patient at home with results returned digitally.
Digital therapeutics. Software that is itself the intervention — delivering and evidencing a treatment, not merely supporting one. Clinically validated and usually regulated.
Virtual clinical trials. Studies run remotely through apps, connected devices and telehealth visits.
IoT in healthcare. Connected devices and sensors reporting on patients, equipment or environments.
Edge computing. Processing data near where it is produced, to cut latency and survive an unreliable link to the centre.
Patient engagement
Digital front door. The digital entry point through which a person first reaches a service — find, book, register, pay, message. Its failure mode is being a front door onto a building with no corridors.
Digital intake and check-in. Completing registration, history and consent before or on arrival rather than at a desk.
Patient engagement. The tools and practices that involve people in decisions about their own care. Patient activation measures the knowledge, skill and confidence they bring to it.
Patient experience. How care felt — communication, waiting, dignity, continuity — as distinct from clinical outcome. Patient feedback loop: the collection of that signal and, the harder half, acting on it.
PGHD — Patient-Generated Health Data. Data created by the patient outside clinical settings. It arrives with different provenance and different reliability from clinical data, and should be stored saying so.
Secure patient messaging. Encrypted asynchronous communication between a person and their care team.
PHR — Personal Health Record. See systems and records.
Health literacy tools. Aids that make health information usable by the person it concerns.
Digital consent. Obtaining and recording permission electronically. Informed consent management is the wider obligation that the person actually understood the options, risks and alternatives. See consent and trust.
Patient rights. The entitlements a person holds in the system — consent, privacy, and access to their own record.
Patient access mandates. Rules requiring providers and payers to make a person's data available to them digitally, typically through an API. See SMART on FHIR.
Behavioural nudging. Prompts designed to shift a decision. Effective, and therefore an ethical question rather than a design detail.
Human-centred design. Designing from the needs and behaviour of the people who will use the thing. Consumer-centric health design applies the same stance to health services specifically.
Clinical software and workflow
CDS — Clinical Decision Support. Filtered, situation-specific information delivered to a clinician at the point of decision. See CDS Hooks for the invocation mechanism.
CDI — Clinical Documentation Improvement. Work to make the record accurately reflect what was found and done — which also determines what can be coded and reimbursed.
Automated chart review. Software scanning records for documentation gaps, quality issues or coding opportunities.
Clinical surveillance software. Continuous monitoring of clinical data for deterioration, sepsis or infection risk, raising alerts against defined criteria.
Longitudinal patient record. A continuous record following a person across time, providers and settings — the point of a shared health record.
CCD — Continuity of Care Document. A CDA document summarising a patient at a transition of care.
e-Prescribing (eRx). Generating and transmitting prescriptions electronically to a dispensing point.
Structured data capture. Collecting information in a defined, coded form at the point of entry, so that it can later be counted rather than read.
Intelligent scheduling. Rules- or model-driven allocation of appointments, rooms and staff against demand.
Workforce management platform. Software for rostering, skill-mix and staffing levels.
Clinical workflow optimisation. Redesigning the process, not only digitising it. Workflow integration is the test of whether a tool sits inside the clinician's existing path or beside it — the usual reason good software goes unused.
IAM — Identity and Access Management. The systems governing who may reach which data. See security architecture.
SaaS in healthcare. Cloud-delivered software subscribed to rather than installed. See infrastructure.
Blockchain in healthcare. A distributed append-only ledger proposed for records, consent and supply chain provenance. Widely piloted; rarely the smallest thing that solves the problem.
Quality, population health and measurement
Quality improvement. Systematic, data-driven change to make care safer, more effective, more timely and more equitable.
Quality measures / quality metrics. The indicators used to assess care — screening rates, complication rates, patient-reported outcomes. CQMs — Clinical Quality Measures are the formally specified versions. Quality reporting programme: the structured submission of these to a regulator or payer.
HEDIS. A standardised US measure set for comparing health plan performance. NCQA is the non-profit that maintains it and accredits against it. Health plan star ratings and the HOS — Health Outcomes Survey are the US Medicare consumer-facing quality signals.
Patient safety. The prevention of harm caused by care itself.
Risk management. Identifying and reducing clinical, financial and reputational exposure.
Clinical registry. A curated dataset about a defined condition, procedure or cohort, maintained for outcome measurement and research.
RWE — Real-World Evidence. Evidence about a treatment's use and effect drawn from routine data rather than trials.
Population health management. Aggregating and analysing data for a defined population in order to act on it. See public health surveillance.
Risk stratification. Grouping a population by risk so that effort goes where it changes the outcome.
Patient attribution. The rule assigning a person to a provider or group for accountability. Attributed lives is the resulting population count — and the rule, not the count, is where the arguments are.
RAF — Risk Adjustment Factor. A score predicting a patient's expected cost from diagnoses and demographics. HCC coding — Hierarchical Condition Category is the classification producing it.
CMI — Case Mix Index. The average relative resource intensity of a hospital's patients, derived from DRG assignment.
Health equity metrics. Measures that disaggregate outcomes and access by population group, so that an improving average cannot hide a widening gap.
SDoH — Social Determinants of Health. Housing, food, income, education, transport — the non-clinical conditions that determine most of the outcome.
CHNA — Community Health Needs Assessment. A structured assessment of the health needs of the population a provider serves.
Financing and payment
Payment model shapes system behaviour more reliably than any architecture does. Much of this vocabulary originates in the United States; the underlying mechanisms recur in most financing systems. See health financing.
FFS — Fee-for-Service. Payment per service delivered. Rewards volume.
Value-based care. Payment tied to outcomes rather than activity. Value-based purchasing is the payer-side programme implementing it.
Capitation. A fixed payment per enrolled person per period, whatever is delivered. Transfers risk to the provider.
Bundled payments. One payment covering all services in a defined treatment. Episode-based payment is the same idea framed around an episode of care.
Shared savings. Providers who deliver agreed quality below a cost benchmark keep a share of the difference.
APM — Alternative Payment Model. Any payment approach other than straightforward fee-for-service. Advanced APM denotes those carrying meaningful downside risk. APM risk adjustment corrects payment for the expected needs of the population enrolled.
Risk-based contracting. Agreements in which the provider carries financial risk. Performance-based risk sharing ties a vendor's or supplier's payment to achieved outcomes.
MACRA / QPP / MIPS. US legislation of 2015 and the Quality Payment Program it created, under which MIPS — Merit-based Incentive Payment System adjusts clinician payment on quality, cost and technology use.
CMS — Centers for Medicare & Medicaid Services. The US federal agency administering those programmes. CMMI is its innovation centre, which tests new payment and delivery models.
Medicare / Medicare Advantage / Medicaid. US public coverage for older people, its privately administered alternative, and the jointly funded programme covering people on low incomes. The ACA — Affordable Care Act is the 2010 law that expanded coverage and created the insurance marketplaces.
MLR — Medical Loss Ratio. The share of premium that must be spent on care and quality rather than administration and margin.
PBM — Pharmacy Benefit Manager. A third party administering drug benefits between payer, pharmacy and manufacturer.
Prior authorisation. Payer approval required before a service or medicine is provided. Prior authorisation automation is the attempt to make that approval fast enough not to be a clinical delay.
Utilisation management. Review of whether services are necessary, appropriate and proportionate.
Claims adjudication. The payer's decision to pay, deny or adjust a submitted claim. Claims processing workflow is the path from submission to settlement; claims denial management is the systematic handling of what comes back refused.
RCM — Revenue Cycle Management. The whole financial path from registration and eligibility through coding, billing and collection. Payment integrity is the assurance that what was paid should have been. Reimbursement optimisation is the provider-side work of reducing denials and under-claiming; reimbursement policy and reimbursement strategy are the payer's rules and the provider's response to them.
CPT — Current Procedural Terminology. The US procedural coding system used for billing.
Payor mix. The distribution of a provider's revenue across funding sources. Net patient revenue is what remains after contractual adjustments, discounts and bad debt. Operating margin is operating revenue less operating cost, as a share of revenue. CCR — Cost-to-Charge Ratio estimates actual cost from billed charges.
Deductible / co-insurance / out-of-pocket maximum. What the patient pays before cover begins, the percentage they continue to pay after it, and the ceiling beyond which the insurer pays in full.
In-network vs out-of-network. Whether a provider holds a contract with the person's plan. Network adequacy is whether the contracted network is sufficient to give actual access. Provider directory accuracy determines whether the person can find out — a registry problem before it is a policy one. See facility registry.
HRRP — Hospital Readmissions Reduction Program. A US programme reducing payment to hospitals with excess readmissions; the readmission penalty is its instrument. The HAC Reduction Program does the same for preventable hospital-acquired conditions.
Hospital reimbursement. The mix of mechanisms — case rates, bundles, fee-for-service, global budgets — by which a facility is actually paid.
Operations, procurement and partnership
Healthcare operations. The daily running of services, staff, supplies and finances.
Capacity planning. Estimating the staff, space and equipment required to meet expected demand.
Business case development. Setting out why an initiative should proceed — benefits, costs, risks and who bears them. ROI model: the financial frame used to argue it. Both are more honest when the assumptions are listed separately from the arithmetic.
Procurement process. How an organisation specifies, evaluates and buys. Architecture choices are usually made here, whether or not architects are present.
Value analysis committee. A multidisciplinary group assessing a proposed clinical product for cost, quality and operational fit.
Market segmentation. Dividing a population into groups with shared characteristics in order to target a service or product at them.
Strategic partnership. A formal collaboration between organisations towards a shared goal. Channel partnership: distributing through another organisation's route to market. Integration partner: a third party connecting systems and platforms to each other.
Health system consolidation. Mergers and acquisitions producing larger provider organisations — which changes what interoperability is for, since more of the exchange becomes internal.
IT governance. The framework aligning technology decisions with organisational goals, risk appetite and regulatory duty. See governance.
Regulation and compliance
Regulatory compliance. Meeting the legal and sector obligations that apply — privacy, safety, device regulation, reporting.
HIPAA. The US law setting standards for the protection of health information. BAA — Business Associate Agreement: the contract binding a vendor handling protected health information to those standards. Compare GDPR and ISO 27799.
Data security. The controls protecting health information against unauthorised access, loss or alteration. See security architecture.
Data transparency. Openness about what data is collected, how it is used, with whom it is shared and on what basis.
Anonymised / de-identified data. Data with identifiers removed. The two terms are used loosely and often interchangeably; neither removes re-identification risk, which must be assessed. See de-identification.
Machine-readable formats. Structured formats software can process, as opposed to documents a person can only read. Most transparency and price-disclosure rules turn on this distinction.
Health data harmonisation. Making data from different sources consistent enough to be analysed together. See terminology services.
ONC rule. US regulation from the Office of the National Coordinator on interoperability, patient access and information blocking.
TEFCA — Trusted Exchange Framework and Common Agreement. The US framework for nationwide exchange across networks. QHIN — Qualified Health Information Network: a network meeting its technical and legal conditions. Compare centralised and federated exchange.
Roles
Titles vary; the functions recur. Knowing which of these exists in an organisation, and which does not, tells you where a digital health decision will actually be made.
CIO — Chief Information Officer. IT strategy, infrastructure and security.
CTO — Chief Technology Officer. The technology platform and product stack.
CMIO — Chief Medical Information Officer. A clinician bridging medicine and IT: record usability, clinical workflow, informatics.
CNIO — Chief Nursing Informatics Officer. The equivalent for nursing practice and workflow.
CMO — Chief Medical Officer. Clinical policy, safety and performance. CNO — Chief Nursing Officer: nursing services and care quality.
COO — Chief Operating Officer. Daily operations and delivery.
CDO — Chief Digital Officer. Digital transformation and digital experience.
CXO — Chief Experience Officer. Patient, provider and user experience across the journey.
Chief Innovation Officer. New models, pilots and the path — often missing — from pilot to production.
Chief Population Health Officer. Outcomes, risk and cost across an attributed population.
CSO — Chief Strategy Officer. Long-term direction across markets and services. CCO — Chief Compliance Officer: regulatory and legal conformance.
Programme and policy
DPI — Digital Public Infrastructure. Shared national rails for identity, payments and data exchange. DPI-H — the health-specific framing; see digital public infrastructure.
DAK — Digital Adaptation Kit. WHO's software-neutral specification of a guideline area: personas, workflows, data dictionary, decision logic, indicators. See SMART Guidelines.
CQL — Clinical Quality Language. HL7's language for clinical decision logic and quality measures.
BPMN — Business Process Model and Notation. OMG's workflow notation, used in DAKs.
DHIS2 — District Health Information Software 2. The open-source HMIS platform. See DHIS2.
Global good / digital public good. Open-source software meeting defined criteria for reuse across countries.
Tier 1–4. This knowledge base's source classification: official, community, educational, experimental. See the overview.