Using Characterization Package

Jenna Reps

2024-08-07

Source: vignettes/UsingPackage.Rmd

Introduction

This vignette describes how you can use the Characterization package for various descriptive studies using OMOP CDM data. The Characterization package currently contains three different types of analyses:

  • Aggregate Covariates: this returns the mean feature value for a set of features specified by the user for i) the Target cohort population, ii) the Outcome cohort population, iii) the Target population patients who had the outcome during some user specified time-at-risk and iv) the Target population patients who did not have the outcome during some user specified time-at-risk.
  • DechallengeRechallenge: this is mainly aimed at investigating whether a drug and event are causally related by seeing whether the drug is stopped close in time to the event occurrence (dechallenge) and then whether the drug is restarted (a rechallenge occurs) and if so, whether the event starts again (a failed rechallenge). In this analysis, the Target cohorts are the drug users of interest and the Outcome cohorts are the medical events you wish to see whether the drug may cause. The user must also specify how close in time a drug must be stopped after the outcome to be considered a dechallenge and how close in time an Outcome must occur after restarting the drug to be considered a failed rechallenge).
  • Time-to-event: this returns descriptive results showing the timing between the target cohort and outcome. This can help identify whether the outcome often precedes the target cohort or whether it generally comes after.

Setup

In this vignette we will show working examples using the Eunomia R package that contains simulated data. Run the following code to install the Eunomia R package:

install.packages("remotes")
remotes::install_github("ohdsi/Eunomia")

Eunomia can be used to create a temporary SQLITE database with the simulated data. The function getEunomiaConnectionDetails creates a SQLITE connection to a temporary location. The function createCohorts then populates the temporary SQLITE database with the simulated data ready to be used.

connectionDetails <- Eunomia::getEunomiaConnectionDetails()
## attempting to download GiBleed
## attempting to extract and load: C:\Users\cknoll1\AppData\Local\Temp\5\RtmpoFseQQ/GiBleed_5.3.zip to: C:\Users\cknoll1\AppData\Local\Temp\5\RtmpoFseQQ/GiBleed_5.3.sqlite
Eunomia::createCohorts(connectionDetails = connectionDetails)
## Cohorts created in table main.cohort
##   cohortId       name
## 1        1  Celecoxib
## 2        2 Diclofenac
## 3        3    GiBleed
## 4        4     NSAIDs
##                                                                                        description
## 1    A simplified cohort definition for new users of celecoxib, designed specifically for Eunomia.
## 2    A simplified cohort definition for new users ofdiclofenac, designed specifically for Eunomia.
## 3 A simplified cohort definition for gastrointestinal bleeding, designed specifically for Eunomia.
## 4       A simplified cohort definition for new users of NSAIDs, designed specifically for Eunomia.
##   count
## 1  1844
## 2   850
## 3   479
## 4  2694

We also need to have the Characterization package installed and loaded

remotes::install_github("ohdsi/FeatureExtraction")
remotes::install_github("ohdsi/Characterization", ref = "new_approach")
## 
## Attaching package: 'dplyr'
## The following objects are masked from 'package:stats':
## 
##     filter, lag
## The following objects are masked from 'package:base':
## 
##     intersect, setdiff, setequal, union

Examples

Aggreagate Covariates

To run an ‘Aggregate Covariate’ analysis you need to create a setting object using createAggregateCovariateSettings. This requires specifying:

  • one or more targetIds (these must be pre-generated in a cohort table)
  • one or more outcomeIds (these must be pre-generated in a cohort table)
  • the covariate settings using FeatureExtraction::createCovariateSettings or by creating your own custom feature extraction code.
  • the time-at-risk settings
  • riskWindowStart
  • startAnchor
  • riskWindowEnd
  • endAnchor

Using the Eunomia data were we previous generated four cohorts, we can use cohort ids 1,2 and 4 as the targetIds and cohort id 3 as the outcomeIds:

exampleTargetIds <- c(1, 2, 4)
exampleOutcomeIds <- 3

If we want to get information on the sex, age at index and Charlson Comorbidity index we can create the settings using FeatureExtraction::createCovariateSettings:

exampleCovariateSettings <- FeatureExtraction::createCovariateSettings(
  useDemographicsGender = T,
  useDemographicsAge = T,
  useCharlsonIndex = T
)

There is an additional covariate setting require that is calculated for the cases (patients in the target cohort with have the outcome during the time-at-risk). This is called caseCovariateSettings and should be created using the createDuringCovariateSettings function. The user can pick conditions, drugs, measurements, procedures and observations. In this example, we just include condition eras groups by vocabulary heirarchy. We also need to specify two related variables casePreTargetDuration which is the number of days before target index to extract features for the cases (answers what happens shortly before the target index) and casePostOutcomeDuration which is the number of days after the outcome date to extract features for the cases (answers what happens after the outcome). The case covariates are also extracted between target index and outcome (answers the question what happens during target exposure).

caseCovariateSettings <- Characterization::createDuringCovariateSettings(
  useConditionGroupEraDuring = T
)

If we want to create the aggregate features for all our target cohorts, our outcome cohort and each target cohort restricted to those with a record of the outcome 1 day after target cohort start date until 365 days after target cohort end date with a outcome washout of 9999 (meaning we only include outcomes that are the first occurrence in the past 9999 days) and only include targets or outcomes where the patient was observed for 365 days or more prior, we can run:

exampleAggregateCovariateSettings <- createAggregateCovariateSettings(
  targetIds = exampleTargetIds,
  outcomeIds = exampleOutcomeIds,
  riskWindowStart = 1, startAnchor = "cohort start",
  riskWindowEnd = 365, endAnchor = "cohort start",
  outcomeWashoutDays = 9999,
  minPriorObservation = 365,
  covariateSettings = exampleCovariateSettings,
  caseCovariateSettings = caseCovariateSettings,
  casePreTargetDuration = 90,
  casePostOutcomeDuration = 90
)

Next we need to use the exampleAggregateCovariateSettings as the settings to computeAggregateCovariateAnalyses, we need to use the Eunomia connectionDetails and in Eunomia the OMOP CDM data and cohort table are in the ‘main’ schema. The cohort table name is ‘cohort’. The following code will apply the aggregated covariates analysis using the previously specified settings on the simulated Eunomia data, but we can specify the minCharacterizationMean to exclude covarites with mean values below 0.01, and we must specify the outputFolder where the csv results will be written to.

runCharacterizationAnalyses(
  connectionDetails = connectionDetails,
  cdmDatabaseSchema = "main",
  targetDatabaseSchema = "main",
  targetTable = "cohort",
  outcomeDatabaseSchema = "main",
  outcomeTable = "cohort",
  characterizationSettings = createCharacterizationSettings(
    aggregateCovariateSettings = exampleAggregateCovariateSettings
  ),
  databaseId = "Eunomia",
  runId = 1,
  minCharacterizationMean = 0.01,
  outputDirectory = file.path(getwd(), "example_char", "results"), executionPath = file.path(getwd(), "example_char", "execution"),
  minCellCount = 10,
  incremental = F,
  threads = 1
)

You can then see the results in the location file.path(getwd(), 'example_char', 'results') where you will find csv files.

Dechallenge Rechallenge

To run a ‘Dechallenge Rechallenge’ analysis you need to create a setting object using createDechallengeRechallengeSettings. This requires specifying:

  • one or more targetIds (these must be pre-generated in a cohort table)
  • one or more outcomeIds (these must be pre-generated in a cohort table)
  • dechallengeStopInterval
  • dechallengeEvaluationWindow

Using the Eunomia data were we previous generated four cohorts, we can use cohort ids 1,2 and 4 as the targetIds and cohort id 3 as the outcomeIds:

exampleTargetIds <- c(1, 2, 4)
exampleOutcomeIds <- 3

If we want to create the dechallenge rechallenge for all our target cohorts and our outcome cohort with a 30 day dechallengeStopInterval and 31 day dechallengeEvaluationWindow:

exampleDechallengeRechallengeSettings <- createDechallengeRechallengeSettings(
  targetIds = exampleTargetIds,
  outcomeIds = exampleOutcomeIds,
  dechallengeStopInterval = 30,
  dechallengeEvaluationWindow = 31
)

We can then run the analysis on the Eunomia data using computeDechallengeRechallengeAnalyses and the settings previously specified, with minCellCount removing values less than the specified value:

dc <- computeDechallengeRechallengeAnalyses(
  connectionDetails = connectionDetails,
  targetDatabaseSchema = "main",
  targetTable = "cohort",
  settings = exampleDechallengeRechallengeSettings,
  databaseId = "Eunomia",
  outcomeTable = file.path(getwd(), "example_char", "results"),
  minCellCount = 5
)

Next it is possible to compute the failed rechallenge cases

failed <- computeRechallengeFailCaseSeriesAnalyses(
  connectionDetails = connectionDetails,
  targetDatabaseSchema = "main",
  targetTable = "cohort",
  settings = exampleDechallengeRechallengeSettings,
  outcomeDatabaseSchema = "main",
  outcomeTable = "cohort",
  databaseId = "Eunomia",
  outcomeTable = file.path(getwd(), "example_char", "results"),
  minCellCount = 5
)

Time to Event

To run a ‘Time-to-event’ analysis you need to create a setting object using createTimeToEventSettings. This requires specifying:

  • one or more targetIds (these must be pre-generated in a cohort table)
  • one or more outcomeIds (these must be pre-generated in a cohort table)
exampleTimeToEventSettings <- createTimeToEventSettings(
  targetIds = exampleTargetIds,
  outcomeIds = exampleOutcomeIds
)

We can then run the analysis on the Eunomia data using computeTimeToEventAnalyses and the settings previously specified:

tte <- computeTimeToEventAnalyses(
  connectionDetails = connectionDetails,
  cdmDatabaseSchema = "main",
  targetDatabaseSchema = "main",
  targetTable = "cohort",
  settings = exampleTimeToEventSettings,
  databaseId = "Eunomia",
  outcomeTable = file.path(getwd(), "example_char", "results"),
  minCellCount = 5
)

Run Multiple

If you want to run multiple analyses (of the three previously shown) you can use createCharacterizationSettings. You need to input a list of each of the settings (or NULL if you do not want to run one type of analysis). To run all the analyses previously shown in one function:

characterizationSettings <- createCharacterizationSettings(
  timeToEventSettings = list(
    exampleTimeToEventSettings
  ),
  dechallengeRechallengeSettings = list(
    exampleDechallengeRechallengeSettings
  ),
  aggregateCovariateSettings = exampleAggregateCovariateSettings
)

# save the settings using
saveCharacterizationSettings(
  settings = characterizationSettings,
  saveDirectory = file.path(tempdir(), "saveSettings")
)

# the settings can be loaded
characterizationSettings <- loadCharacterizationSettings(
  saveDirectory = file.path(tempdir(), "saveSettings")
)

runCharacterizationAnalyses(
  connectionDetails = connectionDetails,
  cdmDatabaseSchema = "main",
  targetDatabaseSchema = "main",
  targetTable = "cohort",
  outcomeDatabaseSchema = "main",
  outcomeTable = "cohort",
  characterizationSettings = characterizationSettings,
  outputDirectory = file.path(tempdir(), "example", "results"),
  executionPath = file.path(tempdir(), "example", "execution"),
  csvFilePrefix = "c_",
  databaseId = "1",
  incremental = F,
  minCharacterizationMean = 0.01,
  minCellCount = 5
)

This will create csv files with the results in the saveDirectory. You can run the following code to view the results in a shiny app:

viewCharacterization(
  resultFolder = file.path(tempdir(), "example", "results"),
  cohortDefinitionSet = NULL
)