OHDSI Home | Forums | Wiki | Github

How missing values are covered by PLP

(Eldar Allakhverdiiev) #1

When taking a look at results of PLP, we have found an interesting thing:
for some of our covariates based on measurement values CovariateCountWithOutcome ( as well as CovariateCountWithNoOutcome) are much less then expected.
For example, body mass index (BMI) values are around 13, while direct look at data shows values ~ 21.

As far as I understand the problem is that not all subjects included into the study do have this measurement results.So for the rest of subjects covariate values are considered as 0
Can anyone confirm if this is an actual behavior of the package?
@Rijnbeek, @jreps, any thoughts will be appreciated.

(Youjin Park) #2

Hi, I’m youjin.
Recently, I’m trying to develop a PLP model using measurement values too.
you can see what happens if there were missing measurement values in the Book of OHDSI chapter 13,

The book said PLP package inserts them as 0.

So, I used to (1) use the measurement values without missing and (2) exclude the measurement concepts with many missing. And sometimes, I (3) excluded the patients with missing when I use measurement value as covariates.

I think talking about data missing when we use measurement values in PLP is a good topic.