When a Data Quality Dashboard and Clinicians Disagree on Blood Pressure Control
A primary care network's quality dashboard reports 58% hypertension control, but clinicians insist it is closer to 75%. The measure counts the most recent BP in the prior 12 months and requires a value below 140/90. Investigation reveals: (1) BP values imported from patient home devices are stored in the same field as clinic readings; (2) some clinics record repeat readings in free-text notes rather than structured fields; (3) a subset of records has BP entered with systolic and diastolic swapped; (4) patients with no BP in 12 months are counted as uncontrolled rather than excluded. Leadership wants the number 'fixed' before a payer meeting next week.
1. How would you diagnose and prioritize these data quality issues without simply producing a more favorable number?
Read the worked answer
Frame this as a data quality assessment, not a number-tuning exercise. Use a standard dimension lens (accuracy, completeness, validity, timeliness) and quantify each issue's impact.
Diagnosis:
- Provenance (accuracy): home-device readings mixed with clinic readings change what the measure means. Quantify how many qualifying readings are home vs. clinic and whether the measure specification permits home readings.
- Completeness/structure: free-text repeat readings mean structured data undercounts control. Sample notes to estimate how many patients would flip.
- Validity: swapped systolic/diastolic produces physiologically implausible values (e.g., systolic 80, diastolic 160). Run range and plausibility rules to count affected records.
- Denominator definition: counting patients with no recent BP as uncontrolled is a specification choice, not a data error. Confirm against the measure steward's definition and document the decision.
Prioritization: fix validity (swaps) and provenance (separate home readings) first because they are tractable and affect interpretation; address free-text capture through workflow change, not retroactive guessing.
For the payer meeting, present the corrected number with a methods note showing each adjustment and its magnitude. A defensible 71% with transparent exclusions beats an unexplained 75%.
2. What ongoing controls would prevent these issues from recurring?
Read the worked answer
Move from one-time cleanup to sustained data quality management.
- Validation rules at entry: range and cross-field checks reject swapped or implausible BP (e.g., systolic must exceed diastolic; flag values outside physiologic bounds).
- Structured capture: require repeat readings in discrete fields; retire free-text BP entry where feasible.
- Source tagging: store device/source (clinic, home, other) as an attribute so measures can filter by provenance.
- Measure specification governance: document denominator/exclusion logic and get clinical sign-off so the definition is stable and auditable.
- Monitoring: a recurring data quality scorecard tracking completeness, validity-failure rate, and provenance mix, reviewed monthly by a data governance group.
- Feedback loop: route persistent outliers back to the clinics generating them rather than silently correcting centrally.