A small example
Three people each receive two tests. The dataset contains six test records but three people. If both tests are positive for one person, counting positive tests does not directly give the number of people with positive results.
An episode count adds another rule about which results belong together.
Define the deduplication method
| Unit | Rule needed |
|---|---|
| Test | What constitutes a separate test record? |
| Person | How are identities linked? |
| Episode | Which dates and criteria group records? |
The appropriate episode definition is disease- and purpose-specific. A universal arbitrary window can merge distinct events or split one event incorrectly.
Corrections can resemble new records
An amended result may arrive as another row. Without status and identifiers, it can be counted twice. Preserve the relationship between original and corrected records.
Label the dashboard clearly
“Positive results” and “people with positive results” should not be used interchangeably. Comparisons across places require compatible counting rules and coverage.
Deduplication is not merely a cleaning step hidden before analysis. It defines the event being measured and can materially change the trend. Document it so readers can understand why a laboratory total differs from a public case count.
