Day 04

Note that there’s good and bad variance.

Strategus is the OHDSI framework that lets you execute studies “reliably”. Bunch of configurable R scripts that you will use to run your analysis.

Epi people will have a few terms like “On treatment” and “Intention to treat” TODO: these may not be best of names… try to map these to things you know.

NOTE: You always build separate cohorts. And then you use Propensity Scores to match people! That is, you’re not using the same Patient ID to match across cohorts.

NOTE: When building a cohort, try and keep the ideas of ‘disease’ and ‘study’ reusable. You’re trying to specify “What is this disease?” E.g. with Ulcerative Colitis, you may run five different studies with different sets of people (adults or otherwise, female or otherwise, etc.) This is the same with the time period for the study as well. You specify these at the time of setting up your analysis.

Generally speaking and for Equipoise, $>50\%$ overlap is great. If it’s $< 25\%$ you really shouldn’t be performing that study. You’re basically studying edge cases in the overlapping populations.

RShiny is used for the reporting/diagnostic UI.

Note: KEEPER is only for conditions and not for procedures. You can generate samples based on the cohort for a procedure and study them if you’d like.

MACE — Major Adverse Cardiovascular Event. MI and Stroke are always there. Heart Failure may be another. But people quabble about the rest (there’s 3-point, 4-point, etc.)

TODO: Negative Controls?

Note: You must click the “Generate” button to generate your cohort! It’s not (as you’d erroneously assumed) a ‘counting’ tab. Clicking it populates the cohort table.

“Clean Window” is the span of time that once you’ve had the event, you cannot have the event again (you specify this.) You’re basically saying “I only care that you get it, I don’t care how many times it happens after the first time I observe it.” So if you have hypertension, a cleanWindow would be 999999.