search

ohdsi-studies / ReproducibilitySodhi2023

[under development] study to evaluate claims by Sodhi JAMA 2023
0 stars 1 forks source link

MAX ITERATIONS error on analysis 2 - too large a covariate space #7

Open mdlavallee92 opened 1 year ago

mdlavallee92 commented 1 year ago

When running analysis 2 (the replication of the sodhi study using the covariates for alcohol, smoking, hyperlipidemia, and abdominal surgery), cyclops returns an error of MAX ITERATIONS. The study still provides output despite this error.

In the sodhi study it seems these 4 covariates were indicated based on presence of a larger codeset list. In our replication attempt we have instead added each concept as an individual covariate instead of grouping them. There are thousands of concepts included as covariates all descendants of the 4 covariate types mentioned above. I suspect this is what is leading to the MAX ITERATION error since the model can't converge with such a large X space.

Is it possible to use cohorts as included covariates in the model instead of all these concepts? That seems like it would be a better replication of sodhi and would likely alleviate this convergence issue.

msuchard commented 1 year ago

see https://github.com/OHDSI/FeatureExtraction/blob/main/vignettes/CreatingCovariatesBasedOnOtherCohorts.Rmd on how to use cohorts are covariates.

also, that's not a large X space (unless you are not using regularization). make sure cross-validation is turned on and regularization is employed with many covariates.