OMOP - Generating Covariates and Table 1 • ohdsilab

This vignette summarizes the code that can be found here: http://ohdsi.github.io/FeatureExtraction/articles/UsingFeatureExtraction.html

I recommending working through the full vignette for those interested. This vignette will show how to apply the code to the ohdsi-lab database.

We’ll need to install one new package: FeatureExtraction

renv::install("OHDSI/FeatureExtraction")

Load the typical packages

library(ohdsilab)
library(keyring)
library(DatabaseConnector)
library(ROhdsiWebApi)
library(CohortGenerator)
library(CohortDiagnostics)
library(FeatureExtraction)

Set the connection details

atlas_url = "https://atlas.roux-ohdsi-prod.aws.northeastern.edu/WebAPI"
cdm_schema = "omop_cdm_53_pmtx_202203"
write_schema = paste0("work_", keyring::key_get("db_username"))

Sys.setenv("DATABASECONNECTOR_JAR_FOLDER" = "insert path to jdbc driver here")

connectionDetails <- createConnectionDetails(
    dbms = "redshift",
    server = "ohdsi-lab-redshift-cluster-prod.clsyktjhufn7.us-east-1.redshift.amazonaws.com/ohdsi_lab",
    port = 5439,
    user = keyring::key_get("db_username"),
    password = keyring::key_get("db_password"))

This code will generate a table 1 for an established cohort of individuals with concepts of stroke and aphasia. This table was generated using CohortGenerator and ATLAS, and saved as “RC_aphasia” in my own schema. In the getDbCovariateData() funtion, we need to provide information about the database connection (connectionDetails), the omop database data, the specific user schema to look in (write_schema). “subject_id” becomes the row for subject - we could also name the “person_id” for example. -1 says to calculate covariates for all cohorts in the table (there’s only 1).

The aggregated argument defaults to FALSE, so we’ll get covariates at the person level here. We can aggregate them in the next step.

# This sets the covarate settings as the defaults
covariateSettings <- createDefaultCovariateSettings()

# This pulls the covariate data from the cohort table
covariateData <- getDbCovariateData(
    connectionDetails = connectionDetails,
    cdmDatabaseSchema = cdm_schema,
    cohortDatabaseSchema = write_schema,
    cohortTable = "RC_aphasia",
    cohortIds = -1,
    rowIdField = "subject_id",
    covariateSettings = covariateSettings)

summary(covariateData)

Calculating covariates can be pretty time consuming, so it’s good to save it once its generated. We can do this using

# save it like this. its an andromeda sql object, only nee dto do this once. 
saveCovariateData(covariateData, "covariates")

# we can then load later if needed it like this.
covariateData <- loadCovariateData("covariates")

According to the documentation, we should ‘tidy up’ the covariates. We can do this using the tidyCovariateData() function.

Normalize covariate values by dividing by the max and/or remove redundant covariates and/or remove infrequent covariates. For temporal covariates, redundancy is evaluated per time ID.

Again, we’ll save the result since it can take a while.

#this tidies up the covariates but I'm not sure what to use it for...
tidyCovariates <- tidyCovariateData(
    covariateData,
    minFraction = 0.001,
    normalize = FALSE,
    removeRedundancy = TRUE)

# aggregates summary statistics for table 1.
covariateData.agg <- aggregateCovariates(covariateData)
saveCovariateData(
    covariateData.agg,
    "covariates_agg")

# we can load it later like this:
covariateData.agg <- loadCovariateData("covariates_agg")

Finally, we can use the createTable1() function to put it all in a Table 1. I found that I prefer to have once column output. I haven’t found a terribly satisfying way of getting the results out in a nicely formatted table, so for now I’ve just saved it as a .csv file and formatted in excel.

result <- createTable1(covariateData.agg)
print(
    result,
    row.names = FALSE,
    right = FALSE)
result <- createTable1(
    covariateData.agg,
    output = "one column")
write.csv(
    result,
    "covariates.csv",
    row.names = FALSE)