OHDSI Home | Forums | Wiki | Github

[Altas] Find out measurements for each person be on separate days no more than 2 years apart

Good morning!

I would like to find out patients who had 2 HbA1c measurements >= 6.5% on separate days no more than 2 years apart to identify diabetic patients by lab results. Attached figure shows what I’ve done in Atlas - unfortunately it failed. I can also get access to the data and cross compare results from both sides, so my questions are:

  1. I tried to trouble shoot and it turns out that the nested criteria doesn’t work at all. The number of identified patients remained the same, whether or not the nested criteria were included. I would appreciate any suggestions on how to address this issue.
  2. I’ve noticed that the use of criteria units requires caution. In order to capture information comprehensively, I specified two units, both of which are percentages because it seems that percentages have been categorized differently for reasons known only to ETLers… So I’m wondering if we need to get a protocol about ETL and to understand how things were mapped before we implement analysis to prevent this type of issue?

Thank you for your patience, and any advice and comments will be greatly appreciated!

Shanshan

The logical issue is that you included the index date of your outer measurement (when you said between 0 days before and all days after). That will include itself in the counts. So, don’t include the index date by specifying ‘between 1 days after and 730 days after’.

Yes, especially when dealing with Measurements. You can look at the Datasources → Measurements report to understand how a particular concept is associated with units. Additionally, take care how you use ‘descendant concepts’ in your concept set (in this case HBA1c), because different descendants may use different units…so in case of measurement, probably better to have very specific concept sets to look for the unit that it typically uses.

1 Like

It works!!!

Gotcha.

Thank you Chris, your answers have been really helpful…

1 Like
t