HA535 Unit 2 Merged Extract Data Quality Review Example

Twelve Point One or Twelve Point Five: A Data Quality Review of a Composite Three-Hospital System's Merged Inpatient Extract, Counting Duplicates, Gaps and Miscoded Stays and Showing How an Unlinked Patient Crosswalk Hides Readmissions

[Student Name]

Purdue University Global

HA535: Data Analytics for Health Care Managers

Unit 2 Data Quality Review

[Instructor Name]

[Date]

The system, the extract and every count are composite and bracketed for the writer's own data.

The Extract

The file holds [29,560] encounter rows from three hospitals, drawn from two electronic record instances stitched together after the rural hospital joined the system. It will feed this term's work on length of stay, benchmarking, utilization and forecasting.

The Framework

Data quality is not a single property. A widely used framework treats it as fitness for use, with dimensions such as accuracy, completeness, consistency and timeliness judged from the user's point of view (Wang & Strong, 1996). A review of electronic health record data quality assessment found that completeness, correctness and concordance are the dimensions most often checked, and that methods should match the intended use (Weiskopf & Weng, 2013).

How the Problems Were Found

Each check was written as a simple rule and run against every row: duplicate rows by comparing all fields; missing values by counting blanks per field and hospital; impossible values by testing discharge dates against admission dates; miscoding by comparing the encounter class with the billing record; and linkage by matching name, date of birth and sex across the two record instances. Every rule and its count is kept in a log so the review can be repeated next quarter.

Findings, Counted

1. Duplicates: [494] exact duplicate rows from an interface resend, [1.67] percent of the file. Fix: remove before any count.

2. Missing discharge disposition: [0.5] percent at the flagship, [0.7] at the suburban hospital and [6.2] at the rural hospital. Effect: the rural hospital's discharges to skilled nursing will be undercounted.

3. Miscoded stays: [278] rural observation stays coded as inpatient admissions. Effect: inflated rural admissions and shortened rural length of stay.

4. Impossible dates: [29] rows discharged before they were admitted. Fix: correct from source or exclude.

5. Unlinked patients: of [799] people treated under both record instances, the crosswalk misses [176], so their returns to a different instance are counted as new admissions.

The quietest finding matters most: [176] people with two record numbers make readmissions look lower than they are.

Fitness-for-Use Table

Length of stay: fit after duplicate removal and correction of impossible dates and rural observation stays. Readmissions: unfit until the crosswalk is repaired; after linkage, fit. Discharge disposition: fit for the flagship and suburban hospitals, unfit for the rural hospital without chart review. Benchmarking: fit only if each hospital's data are cleaned the same way. Forecasting census: fit after duplicate removal.

What Each Problem Would Have Done

Left alone, the duplicates would have inflated volume by almost two percent and shortened average length of stay slightly, since resent rows were mostly short stays. The miscoded observation stays would have made the rural hospital look busier and quicker than it is. The missing dispositions would have hidden how many rural patients go to nursing facilities. None would have been visible in a summary table.

Why the Readmission Gap Matters

A [0.4]-point difference may look trivial, but it moves the system's readmission rate across the line of a peer median used in public reporting, and it changes which hospital appears to perform best. Small linkage errors produce systematic bias, not random noise, because the patients most likely to be missed are those who move between hospitals, often the sickest.

Operational Definitions

Some problems are not errors but differences in definition. The flagship counts a readmission from the date of discharge; the rural hospital's legacy system counted from the date of admission. Data used for improvement need clear operational definitions so that every site counts the same thing the same way (Provost & Murray, 2011). The review adopts the discharge-date rule for all three hospitals.

Repairs

1. Remove exact duplicates using a full-row key.

2. Rebuild the patient crosswalk with name, date of birth, sex and address matching, with manual review of near matches.

3. Recode the [278] observation stays from the billing record.

4. Correct or exclude the [29] impossible dates.

5. Request chart review of rural discharges with missing disposition.

Preventing Recurrence

The interface resend should be fixed at its source. Registration staff at the rural hospital need a required disposition field. The record instances should share one enterprise patient index, and a monthly duplicate report should go to health information management so fresh errors surface in the month they occur instead of at the annual audit.

Limits

The review checked the fields this term's work needs; other fields may carry other problems. Matching rules may still miss some patients.

Conclusion

The merged extract is usable once its duplicates, miscoded stays and broken crosswalk are repaired. Without those repairs, every later analysis would carry errors, and the readmission rate would be quietly wrong.

References

Provost, L. P., & Murray, S. K. (2011). The health care data guide: Learning from data for improvement. Jossey-Bass.

Wang, R. Y., & Strong, D. M. (1996). Beyond accuracy: What data quality means to data consumers. Journal of Management Information Systems, 12(4), 5-33. https://doi.org/10.1080/07421222.1996.11518099

Weiskopf, N. G., & Weng, C. (2013). Methods and dimensions of electronic health record data quality assessment: Enabling reuse for clinical research. Journal of the American Medical Informatics Association, 20(1), 144-151. https://doi.org/10.1136/amiajnl-2011-000681