Preferably an exercise with specific data and using Atlas tools.
There’s a section in the Book of OHDSI on selecting negative controls: Chapter 18 Method Validity | The Book of OHDSI
I have already read that section of the book of OHDSI.
I meant a videotutorial or additional material.
I’ve found this
PLE: Empirical calibration (ehden.eu)
But still interested in more tutorials.
I’m still interested in knowing further details about negative controls.
I’ve read that there are two types of negative controls: exposure controls and outcome controls.
What kind of controls are used in Atlas?
How are they introduced in the model?
I guess they are used to adjust the p-values but I don’t know how.
Negative controls are used for two primary reasons:
- As a study diagnostic. If we observe too much bias for the negative controls, we cannot trust our study to produce unbiased estimates for our exposures or outcomes of interest, so we should terminate the study.
- Calibrate our statistics, such as p-values, confidence intervals, and log likelihood ratio (for MaxSPRT).
How to select negative controls and use them in a study in ATLAS is described in the Book of OHDSI
In HADES, you can have CohortGenerator generate cohorts for your negative controls, and in the various causal effect estimation packages such as CohortMethod and SelfControlledCaseSeries you can include them as outcomes (or exposures), where you can specify
trueEffectSize = 1 to indicate they are negative controls.
What if I don’t have negative controls, or I don’t know them? It seems that Atlas doesn’t allow you to perform any analysis if you don’t have controls.
I have another related question.