How would you detect duplicate safety reports without merging genuinely separate cases?

Why

GVP Module VI Addendum I describes how duplicates are found. Searches compare patient data such as age, sex and initials, product data and reaction data, often grouped first by primary source country, sex and age, as EudraVigilance does. Case identifiers in ICH E2B fields, such as the worldwide unique case identification number, give a quick first check.

The algorithm only suggests candidates. Real duplicates differ because reporters describe events differently and coders choose different MedDRA terms, so matching should be on consistent medical concepts. The risk runs both ways.

Missing duplicates inflates counts and can create false signals. Merging separate cases hides real ones. Two patients of the same age at the same hospital with the same reaction to the same drug may be a cluster, which is exactly what signal detection should see.

So check distinguishing facts: dates of birth, onset dates, dose, lot number and reporter. When cases are merged, a master case keeps all source information and references the duplicates.

How to work it through

1. Run an automated search on patient, product, reaction and country fields at case entry.
2. Check E2B case identifiers for the same worldwide case number.
3. Review suspected pairs manually against the full narratives and source documents.
4. Compare distinguishing details such as onset date, dose, lot and reporter before deciding.
5. Merge confirmed duplicates into a master case that keeps all source information.
6. Log decisions, including pairs judged not to be duplicates, for audit.

Where the answer stops holding

Sparse reports with missing age or sex are hard to match, and broadening the search to null values raises false matches. Privacy masking can remove the very details needed to tell cases apart. Algorithms tuned on one database may perform poorly on another with different reporting patterns.