Introduction

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

Target Baseline Covariates: this returns the mean feature value for a set of features specified by the user for i) the Target cohort populations.
Risk Factor Covariates: this returns the mean feature value for a set of features specified by the user for i) cases: target population patients who had the outcome during some user specified time-at-risk and ii) non-cases: the Target population patients who did not have the outcome during some user specified time-at-risk or just the target population (depending on mode).
Case Series Covariates: this returns the mean feature value for a set of features specified by the user for the cases: target population patients who had the outcome during some user specified time-at-risk. This is done at three different time periods, before target index (before), between target index and outcome index (during) and after outcome index (after).
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

First we need to install the Characterization package:

remotes::install_github("ohdsi/Characterization")

and then load it:

library(Characterization)
library(dplyr)

## 
## 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

In this vignette we will show working examples using a sample of the Eunomia R package GI Bleed simulated data. The function exampleOmopConnectionDetails creates a connection details object for a SQLITE database containing an example observational medical outcomes partnership (OMOP) common data model (CDM) data in a temporary location.

connectionDetails <- Characterization::exampleOmopConnectionDetails()

Examples

Target Baseline Covariates

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

one or more targetIds (these must be pre-generated in a cohort table)
a limitToFirstInNDays that removes target exposures that occur within this number of days of a prior exposure. Use 99999 to restrict to first target exposure.
a minPriorObservation that specifies the minimum number of days in the database a person needs to have at target index to be included.
the covariate settings using FeatureExtraction::createCovariateSettings or by creating your own custom feature extraction code.

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

exampleTargetIds <- c(1, 2, 4)

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 = TRUE,
  useDemographicsAge = TRUE,
  useCharlsonIndex = TRUE
)

If we want to create the aggregate features for all our target cohort restricted to the first ever target index and those where the patient was observed for 365 days or more prior to index, we can run:

exampleTargetBaselineSettings <- createTargetBaselineSettings(
  targetIds = exampleTargetIds,
  limitToFirstInNDays = 99999,
  minPriorObservation = 365,
  covariateSettings = exampleCovariateSettings
)

Next we need to use the exampleTargetBaselineSettings as the settings to computeTargetBaselineAnalyses, 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(
    targetBaselineSettings = exampleTargetBaselineSettings
  ),
  databaseId = "Eunomia",
  outputDatabaseSchema = "main",
  outputTable = 'example_char_cohort', 
  minCharacterizationMean = 0.01,
  outputDirectory = file.path(tempdir(), "example_char", "results"), 
  executionPath = file.path(tempdir(), "example_char", "execution"),
  minCellCount = 10,
  incremental = FALSE,
  nTargetJobs = 1,
  threads = 1
)

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

Risk Factor Covariates

To run an ‘Risk Factor Covariate’ analysis you need to create a setting object using createRiskFactorSettings. 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)
a limitToFirstInNDays that removes target exposures that occur within this number of days of a prior exposure. Use 99999 to restrict to first target exposure.
a minPriorObservation that specifies the minimum number of days in the database a person needs to have at target index to be included.
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 = TRUE,
  useDemographicsAge = TRUE,
  useCharlsonIndex = TRUE
)

If we want to create the aggregate features for all our cases/non-cases which are target cohorts restricted to those with/without 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 where the patient was observed for 365 days or more prior, we can run:

exampleRiskFactorSettings <- createRiskFactorSettings(
  targetIds = exampleTargetIds,
  outcomeIds = exampleOutcomeIds,
  limitToFirstInNDays = 99999, # limit to first target exposure
  riskWindowStart = 1, startAnchor = "cohort start",
  riskWindowEnd = 365, endAnchor = "cohort start",
  outcomeWashoutDays = 9999,
  minPriorObservation = 365,
  covariateSettings = exampleCovariateSettings 
)

Next we need to use the exampleRiskFactorSettings as the settings to computeRiskFactorAnalyses, 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 covariates with mean values below 0.01, in addition we can specify minSMD to exclude covariates that are not sufficiently associated to having the outcome and we must specify the outputFolder where the csv results will be written to.

One key input when running risk factors analysis is the mode. There are currently three supported modes:

Efficient - in this mode the non-cases are the target population
CohortIncidence - in this mode the non-cases are the target population without the outcome and with >=1 day of time-at-risk
PatientLevelPrediction - in this mode the non-cases are the target population without the outcome and without the outcome during the outcome washout days prior to index.

runCharacterizationAnalyses(
  connectionDetails = connectionDetails,
  cdmDatabaseSchema = "main",
  targetDatabaseSchema = "main",
  targetTable = "cohort",
  outcomeDatabaseSchema = "main",
  outcomeTable = "cohort", 
  outputDatabaseSchema = 'main', 
  outputTable = 'example_char_cohort',
  characterizationSettings = createCharacterizationSettings(
    riskFactorSettings = exampleRiskFactorSettings
  ),
  databaseId = "Eunomia",
  minSMD = 0.1, # only keep moderate to strongly associated covariates
  minCharacterizationMean = 0.01,
  outputDirectory = file.path(tempdir(), "example_char", "results"), 
  executionPath = file.path(tempdir(), "example_char", "execution"),
  minCellCount = 10,
  incremental = FALSE,
  nTargetJobs = 1,
  threads = 1, 
  mode = 'CohortIncidence' # can also pick 'Efficient' and 'PatientLevelPrediction'
)

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

Case Series Covariates

To run an ‘Case Series Covariate’ analysis you need to create a setting object using createCaseSeriesSettings. 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)
a limitToFirstInNDays that removes target exposures that occur within this number of days of a prior exposure. Use 99999 to restrict to first target exposure.
a minPriorObservation that specifies the minimum number of days in the database a person needs to have at target index to be included.
the case covariate settings using Characterization::createDuringCovariateSettings or by creating your own custom feature extraction code.
the time-at-risk settings
riskWindowStart
startAnchor
riskWindowEnd
endAnchor
the casePreTargetDuration which is the time before target index to characterize
the casePostOutcomeDuration which is the time after outcome index to characterize

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 conditions and visit counts:

exampleCaseCovariateSettings <- Characterization::createDuringCovariateSettings(
  useConditionOccurrenceDuring = TRUE, 
  useVisitCountDuring = TRUE
  )

We also need to specify two 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).

exampleCaseSeriesSettings <- createCaseSeriesSettings(
  targetIds = exampleTargetIds,
  outcomeIds = exampleOutcomeIds,
  limitToFirstInNDays = 99999, # limit to first target index
  riskWindowStart = 1, startAnchor = "cohort start",
  riskWindowEnd = 365, endAnchor = "cohort start",
  outcomeWashoutDays = 9999,
  minPriorObservation = 365,
  caseCovariateSettings = exampleCaseCovariateSettings,
  casePreTargetDuration = 90,
  casePostOutcomeDuration = 90
)

Next we need to use the exampleCaseSeriesSettings as the settings to computeCaseSeriesAnalyses, 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",
  
  outputDatabaseSchema = "main",
  outputTable = 'example_char_cohort',
  characterizationSettings = createCharacterizationSettings(
    caseSeriesSettings = exampleCaseSeriesSettings
  ),
  databaseId = "Eunomia",
  minCharacterizationMean = 0.01,
  minCovariateCount = 2, 
  outputDirectory = file.path(tempdir(), "example_char", "results"), 
  executionPath = file.path(tempdir(), "example_char", "execution"),
  minCellCount = 10,
  incremental = FALSE,
  nTargetJobs = 1,
  threads = 1
)

You can then see the results in the location file.path(tempdir(), '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",
  outcomeFolder = file.path(tempdir(), "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",
  outcomeFolder = file.path(tempdir(), "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",
  outcomefolder = file.path(tempdir(), "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 = FALSE,
  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
)

Using Characterization Package

Introduction

Setup

Examples

Target Baseline Covariates

Risk Factor Covariates

Case Series Covariates

Dechallenge Rechallenge

Time to Event

Run Multiple