OHDSI Home | Forums | Wiki | Github

"ILLCONDITIONED" error while conducting PLE package

Researchers
#1

Hello, I’m Chul Hyoung Park from Ajou Univerity Hospital. I’m having trouble executing PLE package.
This package runs normally in some databases, but I get an error when I run it on other databse.
It’s an “ILLCONDITIONED” when using argument(s): list(cohortMethodDataFile = "~~

image
image1861×113 20.8 KB

I thought it was covariate matching problem, so several trials were made by setting various covariates (including large scale PSM).
I can’t find any information about the error or how to solve it, but if anyone knows how to fix it, please help. Thank you very much for everyone.

#2

The error message indicates the propensity model cannot be fitted. Unfortunately, these are the hardest problems to debug. Most likely there is something in the data that makes treatment almost perfectly predictable. It isn’t any single variable by itself, since that would trigger a different, more informative error message.

If you’re somewhat handy with R, I would suggest loading the CohortMethodData object (CmData_l1_t2002_t2003.zip) using the loadChortMethodData() function in CohortMethod, and try running createPs() with a fixed strong prior (e.g. createPrior("laplace", variance = 0.001)), and see which covariates end up still having large coefficients.

What are your target and comparator cohorts?

1 Like
#3

@schuemie Thank you very much for answering my question!
I think I can solve the problem as the way you said.
Asthma patients taking statins is in the target cohort, and asthma patients not taking statins is in the comparator cohort.

t