OMOP - Generating a cohort with Capr

This tutorial is similar to the last except in this one we’re generating a cohort using the Capr package instead of ATLAS. After the cohort is generated, it can be treated the exact same as an ATLAS-generated cohort. This tutorial will use a cohort created from the SynPuf synthetic dataset, using a cohort definition with the same criteria as the ATLAS cohort “[DEMO] Type 2 diabetes patients prescribed metformin within 30 days after type 2 diabetes diagnosis” More details on generating cohorts can be found here: https://ohdsi.github.io/CohortGenerator/ and here: https://ohdsi.github.io/TheBookOfOhdsi/Cohorts.html.

We’ll need to install one new package: Capr

renv::install("OHDSI/Capr")

We’ll use the same set up code as the intro vignette.

# ==============================================================================
# Packages =====================================================================
library(Capr)
library(DatabaseConnector)
library(ohdsilab)
library(DatabaseConnector)
library(keyring)
library(CohortGenerator)
library(tidyverse)

# DB Connections ===============================================================
synpuf_schema = "omop_cdm_synpuf_110k_531"
write_schema = paste0("work_", keyring::key_get("db_username"))

Again, we’ll create the connection details and add the path to the jdbc driver we installed in the intro vignette.

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"))

# Connect to the database. You won't need this connection until later, but for 
#now you just need the "con" information saved for the next step.
con =  DatabaseConnector::connect(connectionDetails)

# Make it easier for some r functions to find the database
options(con.default.value = con)
options(schema.default.value = synpuf_schema)
options(write_schema.default.value = write_schema)

Now, we’re ready to start using Capr. Just like ATLAS, the first thing you have to do when designing a cohort is create concept sets for the clinical concepts (conditions, drugs, procedures, etc.) involved in your cohort. Let’s create a concept set for type 2 diabetes using the concept ID 201826 and all of its descendants and a concept set for the drug metformin with only one concept ID 40164929.

t2d <- cs(
    descendants(201826),
    name = "Type 2 diabetes")

metformin <- cs(
    40164929,
    name = "Metformin")

Now we define our cohort using the same logic as ATLAS. The syntax may look a little different, but the ideas are the same. An entry event, inclusion criteria (attrition), and an exit event can be defined. The following cohort includes persons who have a condition occurrence of type 2 diabetes and also a drug exposure of metformin starting between 0 and 30 days after their type 2 diabetes diagnosis. Persons will leave this cohort at the end of continuous observation (default in both ATLAS and Capr)

t2dcohortDef <- cohort(
    entry = 
        entry(conditionOccurrence(t2d), primaryCriteriaLimit = "All"),
    attrition = 
        attrition("metformin within 0-30 days" = 
                                withAll(atLeast(1, drugExposure(metformin), 
                                                                duringInterval(eventStarts(0,30))))))

Now we have to save the cohort definition as a json file, then convert that json into a SQL query that can be run against the database.

#convert cohort definition to json
t2dcohortDef_json <- as.json(t2dcohortDef)

#convert json to SQL query
sql <- CirceR::buildCohortQuery(
  expression = CirceR::cohortExpressionFromJson(t2dcohortDef_json),
  options = CirceR::createGenerateOptions(generateStats = FALSE))

Finally, like we did for the ATLAS cohort, we have to create a cohort definition set, though this time we’re using our own SQL query rather than pulling one from ATLAS.

cohortDefinitionSet <- tibble::tibble(
  cohortId = 1,
  cohortName = "Type 2 Diabetes",
  sql = sql)

And that’s where the differences end! From here, we can run through the same process of generating the cohort.

# Set a naming convention for the cohort tables.
cohortTableNames <- getCohortTableNames(cohortTable = "synpuf_t2d")

# Create empty tables in your personal schema using the naming convention
#designated in the last step.
createCohortTables(
    connectionDetails = connectionDetails,
    cohortTableNames = cohortTableNames,
    cohortDatabaseSchema = write_schema)

# Generate your cohort for the Synpuf database.
cohortsGenerated <- generateCohortSet(
    connectionDetails = connectionDetails,
    cdmDatabaseSchema = synpuf_schema,
    cohortDatabaseSchema = write_schema,
    cohortTableNames = cohortTableNames,
    cohortDefinitionSet = cohortDefinitionSet)

Now to prove that our cohort generated successfully, let’s create a table containing the people in our cohort and count them. This number can be compared to the number of persons in the synpuf dataset in the identical ATLAS cohort “[DEMO] Type 2 diabetes patients prescribed metformin within 30 days after type 2 diabetes diagnosis”

# Create a table containing your new cohort.
cohort <- tbl(
    con,
    inDatabaseSchema(write_schema, "synpuf_t2d")) |>
  select(person_id = subject_id, cohort_start_date, cohort_end_date)

# How many people are in the cohort?
tally(cohort)