HINF 520 Week 6 Data Governance and Data Quality Example

Who Decides What the Data Mean? A Data Governance and Quality Program for a Health System After Its Diabetes Project Exposed Three Definitions, Silent Failures and a Biased Risk Score

[Student Name]

University of Phoenix

HINF/520: Data Management and Design in Health Administration

Week 6 Assignment

[Instructor Name]

[Date]

The health system, its council, policies and findings are composites written for a model paper; research findings come from the sources listed.

A year into the diabetes data work, the health system's chief executive listed what the project had exposed: three departments using three definitions of diabetes, a laboratory upgrade that silently removed a third of the A1c results, five reports of the same measure that disagreed and a vendor risk score used to select patients for care management that no one had tested. She asked the director of enterprise analytics for a data governance and quality program. This paper describes it.

What Data Governance Is

Data governance is the system of decision rights, roles, policies and processes that determines how data are defined, accessed, protected, checked and used. It is not a technology or a single committee. Its test is simple: when a question about data arises, is it clear who decides, by what rules and how the decision is recorded?

The Structure

The program has four layers. A data governance council, chaired by the chief operating officer, with the chief medical, nursing, financial and information officers, the privacy and compliance officers and the analytics director, sets policy and resolves disputes. Data owners, senior leaders accountable for each data domain, such as the chief nursing officer for nursing documentation and the laboratory director for results, approve definitions and uses. Data stewards, experienced staff in each domain, manage definitions, documentation and quality day to day. Data custodians in information technology operate the systems that store and move data.

Policy One: Definitions

Every measure and cohort used in management or external reporting must have a documented definition, approved by the relevant owner and stored in a central data dictionary with a version number. The diabetes definition and value set built earlier became the first entry. New reports must use approved definitions or request a new one through the steward.

Policy Two: Access

Access to identifiable data follows role and need, approved by the data owner and reviewed annually, with identifiers removed for analyses that do not require them.

Policy Three: Quality

Every data flow feeding management or external reports must have automated quality checks, and failures must be reported to the owner and steward within one business day.

A Framework for Quality Checks

The program adopted a harmonized data quality framework that organizes checks into three categories, conformance, completeness and plausibility, each assessed either by verification within the organization or by validation against an external standard (Kahn et al., 2016). When six national data networks mapped their 11,026 existing quality checks to this framework, 99.97% fit its categories, most commonly checks of plausibility, value conformance and completeness (Callahan et al., 2017). Using a common framework lets the system compare its checks with those of others and find gaps.

Applying the Framework

The analytics team built 140 checks for the diabetes data mart and other priority data. Conformance checks confirm that values use expected formats and codes, such as LOINC codes from the approved value set. Completeness checks confirm expected volumes by source and clinic, the kind that would have caught the laboratory upgrade within hours. Plausibility checks flag values outside possible ranges and sudden shifts in distributions.

Policy Four: Change

Changes to source systems that affect coded data must be reviewed for downstream effects, with data stewards on the notification list, following the lesson of the silent failure.

Policy Five: Appropriate Use

Accurate data can still mislead when used for a purpose they do not fit. Obermeyer et al. (2019) examined a commercial tool that ranked a large health system's patients for a high-risk care program. The tool was trained to forecast next year's spending, and spending was lower for Black patients than for equally ill white patients, largely because of unequal access to care. At any given score, Black patients therefore carried more chronic illness, and retraining the tool to predict health rather than spending would have sharply raised the number of Black patients offered help. Data can be complete, correct and plausible and still answer the wrong question.

Applying the Use Policy

Any new use of data to make decisions about patients, including predictive scores, must be approved by the data owner and reviewed by an algorithm review group that asks what the model predicts, whether that target matches the intended purpose and whether performance differs by race, ethnicity, sex, age, language or insurance. The vendor risk score used for care management was reviewed first. It predicted future cost; the group replaced it with a measure combining clinical severity and unmet needs.

Resolving a Dispute

The council's first test came quickly. The finance office wanted to keep counting patients with diabetes by billing codes, because its payer contracts used them, while the quality department wanted the validated clinical definition. The council's decision rights settled it: the endocrinology data owner approved the clinical definition for management reports, and finance kept the billing-based count for contract reconciliation only, labeled as such. Each department got what it needed, and executives no longer saw two unlabeled numbers side by side.

Metadata and Lineage

Governance needs a record of what data exist and where they come from. The data dictionary now holds each approved definition, its owner, steward, source systems, refresh schedule and quality checks, and a lineage map shows which source fields feed each report. When the laboratory plans a change, the steward can see within minutes which reports would be affected. Maintaining the dictionary is part of each steward's job description, with about four hours a week protected for the work.

Privacy and Compliance

Governance works with privacy and compliance programs rather than replacing them. The privacy officer sits on the council, and every new use is checked against privacy rules and patient consent requirements.

Culture and Training

Policies work only if people follow them. Stewards train report writers in the data dictionary, every report carries its definition and data source and a short course teaches managers how to read reports critically.

Measures

The program tracks the share of management reports using approved definitions, the number of quality check failures and the time to resolve them, incidents detected by checks before users noticed, access reviews completed on time and the number of predictive tools reviewed for fairness.

First-Year Plan

In the first year, the council will approve definitions for the 50 most used measures, extend quality checks to all external reporting flows, review all predictive tools in use and complete access reviews for identifiable data.

Conclusion

The diabetes project exposed problems that no single analyst or report could fix: competing definitions, silent failures, conflicting numbers and an untested algorithm. A governance program with clear decision rights, a recognized quality framework and a policy on appropriate use addresses each. Its purpose is to make sure that when the next board member asks a simple question, the health system gives one answer it can explain and trust.

References

Callahan, T. J., Bauck, A. E., Bertoch, D., Brown, J., Khare, R., Ryan, P. B., Staab, J., Zozus, M. N., & Kahn, M. G. (2017). A comparison of data quality assessment checks in six data sharing networks. eGEMs, 5(1), 8. https://doi.org/10.5334/egems.223

Kahn, M. G., Callahan, T. J., Barnard, J., Bauck, A. E., Brown, J., Davidson, B. N., Estiri, H., Goerg, C., Holve, E., Johnson, S. G., Liaw, S.-T., Hamilton-Lopez, M., Meeker, D., Ong, T. C., Ryan, P., Shang, N., Weiskopf, N. G., Weng, C., Zozus, M. N., & Schilling, L. (2016). A harmonized data quality assessment terminology and framework for the secondary use of electronic health record data. eGEMs, 4(1), 1244. https://doi.org/10.13063/2327-9214.1244

Obermeyer, Z., Powers, B., Vogeli, C., & Mullainathan, S. (2019). Dissecting racial bias in an algorithm used to manage the health of populations. Science, 366(6464), 447-453. https://doi.org/10.1126/science.aax2342