4 Comments

So, stratification does not need other data, but the multivariate adjustment model uses external data. What external data is used? I mean, does it have to be an RCT meta-analysis with a dose-response curve, or can it be non-randomized/without a dose-response curve?

Expand full comment

Any study comparing groups with respect to the same exposure and outcome is OK. And both stratification and multivariable adjustment are rooted in real-world data; however, the regression component of multivariable adjustment is using real-world data to create artificial data (i.e. the regression line that does not perfectly go through every real-world data point, unless R=1 or R=-1). Does that answer your question?

Expand full comment

Yes, I think it mostly does. If I got it right, it does not need a dose-response curve and neither does it need to be an RCT (I remembered that we don't even have RCTs on smoking and it’s adjustable, so I could have figured).

But what do you mean by rooted in real-world data?

Just that the results are reflective of reality and non-fictional?

Expand full comment

All I mean is that the regression line is based on real-world data points measured in the study (e.g. how much someone's LDL-C changed in response to changing their saturated fat intake). I only point this out because some people criticise the regression line for being "fake/artificial" as if it's just been pulled from thin air, rather than based on what was observed.

Expand full comment