Validate the data mastering results

Background and Strategy

Once each data mastering object has been run the case-level results should be reviewed.

As a reminder, the grain size of the mastering objects is one record per distinct Source ID - Source Local identifier value, with the "matched" data model key value provided on each record.

For local identifiers that failed to match to any others, the "matched" data model key value will be the original identifier prepended with the Source ID of the system of origin. When two or more identifiers have met the matching critera, they will share the same matched data model key value.

The primary purpose of this validation is to confirm that identifiers that should have matched did in fact match, and vice versa.

Key Diagnostics / Heuristics

1. In Case Review, check the aggregate statistics of the "matched" identifier field (e.g., Matched Patient ID); is the count of distinct values exactly the same as the number of records in the table? If so, no matches have been made. If this is expected, great; if not, investigate.

2. If there have been some matches, sort the Case Review data by the matched identifier field and scan down the list. If you see two records with the same matched identifier (post-sorting, those rows will be clustered together) review the features of each by scrolling to right; does it look like a legitimate match, or a false positive?

3. If you have trouble finding matches by manually scanning the results, create an Ad-Hoc Derived Field using the Count of Instances pattern to generate the count of records sharing the same matched identifier value; then filter on this new field to isolate records with 2 or more instances of the matched value.

Detailed Implementation Guidance

1. You will likely find it useful to create a board to memorialize some of the diagnostic Case Review formats listed above. Call the board "Useful Case Review Formats" or similar, make it visible (and, if appropriate, editable) for other users, and pin it to the left navigation bar.