Counts needed for negative controls?

Researchers
1

Dear everyone,

I would like to hear your opinion on a recommendation from the Book of OHDSI. In the section about method validity some criteria about choosing negative controls are mentioned.
One of them states: “The negative controls should exist in data, ideally with sufficient numbers.”
My question is, what do you consider sufficient numbers or how would evaluate it?
Unfortunately, I’ve been unsuccessfully in trying to find any resources with recommendations.

Kind regards,

Andreas

2

Hi @awrosen! Unfortunately, we currently have no way to tell beforehand what the required amount of data for negative controls is.

We did implement functions to estimate the uncertainty in the empirical calibration, which is driven by the number of negative controls and the power per negative control. So afterwards you can have a sense of whether the numbers were high enough.

As a rule of thumb I’d recommend you want your negative controls to have at least as much power as your hypothesis of interest.

3

Dear @schuemie,

Thanks for the fast reply! That sounds like a good rule of thumb.

KR

Andreas

4

I think not only the counts are relevant though :wink:
It is also important to consider how well the specific control outcome was captured in the respective data source :slight_smile: