OHDSI Home | Forums | Wiki | Github

Heel: Number of death records outside valid observation period (and "umbrella data quality rules")

CDM Builders
#1

The Heel rule 1 logic on analysis 510 requires that all deaths are observed inside an observation period.

In the DQ study results, this rule fires quite often on many datasets.

I think if death event is imported from an external source (e.g., Social Security Death Index), it may be fine (perhaps just a notification) that it falls outside an observation period.

I was hoping to untangle at least some of the rule_1/analysis pairs into separate rules.

Rule 1 and rule 2 are checking a lot of DQ issues at the same time and the drill down logic would be different. If the rule_id is the same, it is hard to support the drill down.

In this case, it could also be changed to a warning. Since a patient can move out of a given state and death is later matched from a national source.

What other users think about rule_id 1 and analysis 510 and “umbrella rules” (rules that check 5+ (sometimes quite varying) data quality issues in one “batch” in general?

ERROR: 510-Number of death records outside valid observation period; count (n=xxxxx) should not be > 0
#2

All events in the data should appear within an of observation period. You have 2 choices:

  1. Go through he people who died and extend their observation period to cover the death date.
  2. Create a 1 day observation period around each death so it’s covered.

The CDM-oriented question you should be asking yourself is: for those people who died outside of an OP, do you think you have confidence that you knew about their medical history leading up to their death? If yes, go with 1. If no, go with 2.

-Chris

#3

I’d like to continue the thread of breaking up “umbrella rules” . . .

I recently got these two errors from ACHILLES:

ERROR: 706 - Distribution of age by drug_concept_id (count = 2); min value should not be negative
ERROR: 906 - Distribution of age by drug_concept_id (count = 2); min value should not be negative

Looking at the SQL it is easy to tell one is for DRUG_EXPOSURE (706) and the other for DRUG_ERA (906).

It will be helpful if the descriptions called this out so you don’t have to go to the SQL to figure it out. Thank you!

Additionally logged here:

1 Like
t