CMDP cheat sheet

Data governance assigns decision rights and accountability for data assets; stewardship executes those decisions operationally at the domain level.

A data governance council typically owns policy approval, while data stewards own definition, quality, and remediation within a business or clinical domain.

A data owner is accountable for a data domain (access approval, classification); a custodian implements technical controls on the owner's behalf.

Data classification tiers (e.g., public, internal, confidential, restricted/PHI) drive handling rules and must map to regulatory categories such as PHI and PII.

A data dictionary defines field names, types, allowed values, and ownership; a business glossary defines shared business terms and their approved definitions.

A RACI matrix documents who is Responsible, Accountable, Consulted, and Informed for each governance activity or data domain.

Governance KPIs commonly include data-quality scorecards, issue-resolution cycle time, and percentage of critical data elements with assigned stewards.

Critical data elements (CDEs) are the subset of fields whose quality materially affects reporting, safety, or regulatory obligations and receive prioritized controls.

ICD-10-CM codes diagnoses and ICD-10-PCS codes inpatient procedures; CPT/HCPCS are used for outpatient procedures and billing.

SNOMED CT is a comprehensive clinical terminology with concept-based, polyhierarchical structure; it is used for EHR problem lists and clinical documentation exchange.

LOINC standardizes laboratory and clinical observation identifiers; each LOINC term combines a component, property, timing, system, scale, and method.

RxNorm provides normalized names and identifiers for clinical drugs; NDC identifies packaged drug products for billing and supply chain.

HL7 v2 uses pipe-delimited segments and messages (e.g., ADT, ORU); HL7 FHIR uses RESTful resources (e.g., Patient, Observation) with JSON or XML.

DICOM governs medical imaging data and metadata, including patient, study, series, and instance identifiers used for image retrieval.

ICD-10-CM has more than 70,000 diagnosis codes; mapping between ICD and SNOMED is many-to-many and requires a maintained crosswalk.

UCUM provides a standard syntax for units of measure (e.g., mg/dL), which is required to interpret numeric observations correctly.

The six data-quality dimensions are completeness, accuracy, consistency, timeliness, validity, and uniqueness.

Completeness = (populated required fields / total required fields) x 100; report by field, record, and dataset level.

Validity checks enforce domain constraints: data type, range, format, code-set membership, and referential integrity.

A data integrity constraint includes primary key uniqueness, foreign key relationships, and check constraints enforced at the database layer.

Duplicate detection uses deterministic matching (exact keys) or probabilistic matching (e.g., Fellegi-Sunter weights on name, DOB, and sex).

Data profiling measures null rates, distinct counts, value distributions, and pattern conformance before rule design.

A data-quality rule should specify the dimension, threshold, severity, owner, and remediation action to be actionable.

Reconciliation compares source and target counts or sums (e.g., record counts, charge totals) to detect loss or duplication during loads.

HIPAA Privacy Rule governs use and disclosure of PHI; the Security Rule requires administrative, physical, and technical safeguards for ePHI.

The HIPAA minimum necessary standard requires limiting PHI use or disclosure to the amount reasonably needed for the purpose.

The 18 HIPAA identifiers (e.g., names, dates more specific than year, geographic subdivisions smaller than state, device identifiers) must be removed for de-identification under the Safe Harbor method.

The HIPAA Expert Determination method de-identifies data when a qualified statistician determines re-identification risk is very small.

Breach notification: covered entities notify affected individuals without unreasonable delay and no later than 60 days after discovery.

The HIPAA Security Rule's three safeguard categories are administrative, physical, and technical; encryption and access control are technical safeguards.

GDPR requires lawful basis for processing, data minimization, and honoring data-subject rights such as access, erasure, and portability.

Role-based access control (RBAC) grants permissions by job function; the principle of least privilege limits access to the minimum needed.

ALCOA+ principles require data to be Attributable, Legible, Contemporaneous, Original, Accurate, plus Complete, Consistent, Enduring, and Available.

Source data verification compares the clinical database against original source documents to confirm accuracy and completeness.

An audit trail must capture who changed what, when, and why, and must be immutable; corrections are made via addendum, not overwrite.

Case report form (CRF) design should collect only protocol-required data; each field maps to a database variable and edit check.

Query management tracks data discrepancies from detection through resolution; metrics include query rate per CRF page and time to close.

Informed consent must be obtained before any study procedure; documentation includes the consent version and date signed.

eSource and EDC systems must comply with 21 CFR Part 11 for electronic records and signatures, including validation and access controls.

Data clarification forms (DCFs) or electronic queries document the resolution of discrepancies between source and database.

Sensitivity = TP / (TP + FN); specificity = TN / (TN + FP); these trade off as the decision threshold moves.

Positive predictive value = TP / (TP + FP) and varies with prevalence, unlike sensitivity and specificity.

A confidence interval quantifies uncertainty around an estimate; a 95% CI excludes the null value when p < 0.05 for a two-sided test.

Categorical data are summarized with counts and percentages; continuous data with mean/SD or median/IQR depending on distribution.

A control chart distinguishes common-cause variation from special-cause variation using control limits, typically at +/-3 sigma.

Choosing the wrong chart type (e.g., pie chart for many categories, truncated axis) can misrepresent magnitude; use bar or line charts for comparison and trend.

Standardized rates (direct or indirect) adjust for confounders such as age and sex when comparing populations.

Data lineage for a report documents source systems, transformations, and refresh cadence so consumers can assess fitness for use.

HL7 v2 messages are triggered by events (e.g., A01 admit, A08 update); segments carry fields separated by pipes and components by carets.

FHIR resources use RESTful interactions (read, search, create) and are versioned; profiles constrain resources for a use case.

The US Core FHIR profiles define required elements for interoperability in US EHR certification and exchange.

A Health Information Exchange (HIE) enables exchange among organizations; models include directed exchange, query-based exchange, and consumer-mediated exchange.

Terminology mapping and value-set binding are required when exchanging coded data; unmapped local codes cause semantic interoperability failures.

Consent management affects HIE data sharing; opt-in, opt-out, and granular consent models change what data is exchanged.

A master patient index (MPI) links patient identities across systems; enterprise MPI (EMPI) spans organizations.

Interface engines route and transform messages, and their logs are the primary evidence for message delivery and error handling.

The data lifecycle phases are creation/capture, processing, storage, use, sharing, archiving, and destruction.

A data warehouse is subject-oriented, integrated, time-variant, and nonvolatile; a data lake stores raw data in native formats until use.

The medallion architecture organizes data into bronze (raw), silver (cleansed/conformed), and gold (curated/analytics-ready) layers.

Normalization (1NF-3NF) reduces redundancy in transactional systems; star schemas with fact and dimension tables optimize analytic queries.

Retention schedules specify minimum retention periods by record type and must align with legal, regulatory, and contractual requirements.

Data destruction must be verifiable (certificates of destruction) and documented; destruction of PHI must follow HIPAA safeguards.

ETL extracts, transforms, and loads data; ELT loads raw data first and transforms in the target, common in cloud warehouses.

A data catalog records metadata (technical, business, operational) to support discovery, lineage, and governance.