Sooner or later, two reports that should agree won't. A practice's quality dashboard shows most diabetic patients current on A1c screening; the payer's care-gap report shows a backlog.

The difference may be upstream: outside lab results are not consistently landing in structured fields, problem lists have been carried forward without maintenance, and duplicate patient records have distorted several denominators.

This is a common shape of a data-quality problem. It shows up as two reports arguing, and it was created months earlier at a keyboard.

Governance scales down

AHIMA defines data governance as the administration — through defined procedures and plans — that assures the availability, integrity, security and usability of an organization's data.¹ Enterprise data-management practice adds the vocabulary of owners, stewards, definitions and quality dimensions.²

Even without a formal data office, a medical group needs answers to five questions: what data matter most, who owns them, who stewards them day to day, what standards apply, and how problems get corrected.

The distinction between owner and steward is worth keeping. An owner is accountable for how a data domain is defined and used. A steward keeps it usable in daily work. Stewardship belongs close to where data are created and corrected — the front desk stewards demographics, a nurse manager stewards medication-list workflow, the coding lead stewards diagnosis specificity, the quality lead stewards payer care-gap files.

Eight domains cover many of the issues a practice argues about: patient identity and demographics, insurance and guarantor data, problem list and diagnoses, medication and allergy lists, orders and results, coding and charge data, payer attribution and care-gap files, and report definitions.

Start with the glossary, because many arguments are definitional

Many data problems are vocabulary problems in disguise. Active patient, no-show, new patient, panel, care gap, established patient, date of last visit — each can mean different things by payer, report, system or department, and each will be defended sincerely by whoever built the report.

A short business glossary gives the practice a common operating definition and makes differences among reports explicit. Six fields per term: the term, the definition, the source system, the owner, where it is used, and the review trigger.

Sample local definition — adapt to the report, contract or program

• Term: Active patient
• Definition: At least one completed visit in the past 24 months, excluding deceased and transferred patients
• Source: EHR / practice management system
• Owner: Operations lead
• Used in: Panel reports, outreach lists, access planning
• Review trigger: Annual, or on payer-contract change

The glossary does not need every field in every system. It needs the terms that drive leadership reports, quality measures, payer contracts, physician scorecards, staffing decisions and outreach. A first version may need only the 15 to 20 terms driving the practice's highest-consequence reports. Its main function is to keep meetings from becoming arguments about whose number is right.

Quality is made at the point of capture

Most data quality is created upstream in registration, rooming, documentation, coding, interface review and payer-file reconciliation. Analytics can detect defects and sometimes correct them, but it cannot reliably compensate for inconsistent capture, indexing or reconciliation.

Recognized quality characteristics translate well into practice examples.¹

• Accuracy: the value is correct and attached to the right patient.
• Completeness: the elements needed for care, payment, reporting or follow-up are present.
• Consistency: the EHR and the practice management system tell the same story.
• Currency: medication lists, problem lists and payer rosters are current enough to act on.
• Granularity: diagnosis specificity supports risk adjustment; screening data sit in discrete fields.
• Timeliness: information is captured or corrected soon enough to change a decision.

Failures cluster in predictable places, and the list is short enough to post:

• Duplicate patient records and identity mismatches
• Outdated demographic or insurance data
• Problem lists copied forward without review
• Medication and allergy lists never reconciled
• Screenings documented in free text rather than structured fields
• Results arriving as scanned documents rather than discrete data
• Payer attribution files imported but not reconciled
• Dashboard logic that shifted after a template or system update
• Coding workflows running on incomplete documentation

Audit one element, all the way through

A useful exercise is to trace one high-value data element from capture to report.

Blood pressure control works well because every step is visible. Before tracing the field, identify the exact measure version and specification governing which readings, dates, exclusions and encounter types count. Then ask: Where is the value entered — device integration, manual entry, which field? Which reading counts when three are taken? How is a repeat measurement after an elevated first reading handled, and does the report know the difference? How does the report pull the field — most recent, last of encounter, an average? Where do errors appear, and who sees them?

A manager who runs that trace once will understand the data environment in a way a policy alone cannot teach. The exercise also tends to surface at least one correctable defect in capture, workflow or report logic.

Reconcile on a schedule, validate before you trust

No practice runs on a single untouched source of truth. Information moves among the EHR, practice management system, lab and imaging feeds, HIEs, payer files, clearinghouses, registries and hand-maintained spreadsheets. Reconciliation compares sources and resolves mismatches before they become missed results, denied claims, wrong outreach lists or a dashboard that quietly misleads the board.

Frequency should follow consequence. Interface queues, unmatched results, duplicate records and failed transmissions need review on a defined short cycle, because delay converts a data problem into a safety problem. Payer attribution, quality-measure denominators and dashboard outputs can run on a monthly or reporting-cycle rhythm.

Validation is the separate discipline of asking whether a report still behaves as expected. Before a quality score, board dashboard or physician scorecard gets used, someone should verify source, denominator, exclusions, update timing and recent changes. A report that always worked can break silently after an EHR update, template change, interface modification or measure-specification revision — and it will keep producing a plausible number the whole time. Reports used for payment, compliance, physician feedback, board review or external submission deserve documented validation rather than assumed reliability.

A minimal validation record should identify the report owner and intended use; source systems and fields; logic or measure version; numerator, denominator and exclusions; refresh timing; a sample chart trace; the last validation date; and known limitations.

Own the exceptions

Every practice generates exceptions: unmatched results, duplicates, rejected interface messages, missing demographic fields, payer discrepancies, dashboard errors, claim edits, patient messages that never attached to a chart. The question is whether each has an owner and a correction path.

An exception process needs six things: who reviews the queue, how often, what threshold escalates, how errors get corrected, how root causes are found, and who confirms the fix held.

The issue log can be simple — issue, source, affected workflow, likely cause, owner, due date, action, status. Its value is in the pattern. If the same duplicate-record problem recurs monthly, the registration matching process needs attention rather than another correction. If quality data fail repeatedly because a field sits three clicks deep in a template, the workflow needs redesign. If payer files persistently conflict with local panels, the attribution process is the problem.

Monitor a small indicator set and make each review produce an assignment: duplicate-record rate, unmatched result count, payer-file discrepancies, report validation errors, missing structured fields, charge lag tied to documentation, correction requests by source. A red indicator with no owner produces concern without corrective action.

The payoff is unglamorous but large

Fewer conflicting reports. Cleaner outreach lists. Quality measures that survive an audit. Payer conversations where your numbers hold. And leaders who stop discounting the dashboard by an unspecified margin because they were burned by it once.

MGMA's 2025 Financials and Operations Data Report provides medical-practice benchmarking across financial and operating measures.³ Any benchmark comparison depends on definitional rigor on both sides. The same is true inside a practice: the number on the report was made weeks earlier — at registration, during rooming, in documentation or through an interface — by someone who had no reason to know it would end up in front of the board.

Notes

1. American Health Information Management Association, Healthcare Data Governance. https://www.ahima.org/media/pmcb0fr5/healthcare-data-governance-practice-brief-final.pdf
2. DAMA International, DAMA-DMBOK: Data Management Body of Knowledge. https://dama.org/learning-resources/dama-data-management-body-of-knowledge-dmbok/
3. MGMA, 2025 MGMA Financials and Operations Data Report, Sept. 4, 2025. https://www.mgma.com/2025-financials-and-operations