Readmission Dashboard: Data Quality Before a Public Claim
Valley Health Network's board dashboard reports 30-day unplanned readmission rates by hospital. The data team built the dashboard from two systems: the electronic medical record (EMR) and a separate claims warehouse. Patient identifiers differ between systems, and a deterministic match on surname, date of birth and postcode fails for 9% of records. Discharge destination is missing for 12% of records, and missing values are currently treated as 'home'. In July, the definition of an unplanned readmission changed from 'any readmission' to 'readmission for a related condition', reducing the reported rate by 2.3 percentage points. The dashboard shows a 0.4-point improvement in August, and the COO has cited it in a public speech. The board paper is due in five days.
1. Assess the data quality dimensions and the risk that the COO's claim is misleading.
Read the worked answer
Data quality assessment
- Completeness: discharge destination is missing for 12%; treating missing as home biases readmission rates because missing cases may differ systematically (for example, transfers or complex patients). The 9% unmatched records add to incompleteness.
- Accuracy and validity: deterministic matching on surname, date of birth and postcode fails for 9%; this can create false matches or missed links, distorting which readmissions are counted.
- Consistency: two source systems with different identifiers; the metric definition changed mid-year. The 2.3-point drop from the definition change is larger than the 0.4-point improvement, so August is not comparable to prior periods.
- Timeliness: the claims warehouse may lag, so recent months may be incomplete.
Risk to the COO's claim
The 0.4-point improvement is within the noise created by a 2.3-point definitional change, 12% missing destination and 9% unmatched records. Attributing it to performance is unsupported. The COO's public statement creates reputational and possibly regulatory risk if the data is challenged.
Judgement
The dashboard is useful for operational improvement but not for public claims of performance until the definition is stabilised and missing data is reduced. The direction of bias from missing-as-home is unknown; it could understate or overstate readmissions. The board should see a data quality statement alongside the metric.
2. Recommend what should be presented to the board and what should be disclosed about limitations.
Read the worked answer
What to present
- Restate the metric on a consistent basis. Recalculate July and August using both the old and new definitions, and show both. Do not compare across the change.
- Show a data quality panel: 9% unmatched, 12% missing discharge destination, claims lag. State the direction of likely bias if known; if not, say unknown.
- Withdraw or qualify the 0.4-point claim. The COO should correct the public statement or clarify that it was based on preliminary data.
- Recommend a remediation plan: probabilistic matching or a master patient index, mandatory discharge destination field, frozen metric definitions with version control, and reconciliation between EMR and claims.
Judgement
The board needs decision-useful information, not a polished number. Presenting a range or a confidence statement is better than false precision. The COO's claim should be corrected promptly; a small correction now is less damaging than a challenge later. The board paper should distinguish performance improvement from measurement change. If the board wants a public claim, require independent validation of the metric first.
Control recommendation
Assign a data owner for the readmission metric, document the definition, and require sign-off before any external use. Add a data quality KPI to the board dashboard itself.