Skip to content

Input Output Specification

jreps edited this page Sep 9, 2024 · 1 revision

Description:

The Characterization R package performs various descriptive analyses.  The current analyses includes:

  •        Time-to-event that looks at timing of when patients in a target cohort experience an outcome (i.e., is the outcome recorded often before or more often after the target index, how close to target index does the outcome occur, etc.).
  •        Dechallenge-rechallenge looks at how often the drug appears to be stopped after an outcome occurs while exposed and how often the drug is restarted after this and whether the outcome comes back when the drug restarts.
  •        Aggregate covariate: Database comparison looks at a cohort’s aggregate features across databases.  Cohorts included are: target, outcome or patients in target with the outcome during time-at-risk. 
  •        Aggregate covariate: Cohort comparison looks at two cohort’s aggregate features to let user find differences or similarities.  Cohorts included are: target, outcome or patients in target with the outcome during time-at-risk. 
  •       Aggregate covariate: Risk factor analysis looks at the cases (patients in target with the outcome during time-at-risk) and non-cases (patients in target without the outcome during time-at-risk) to find differences.  This lets users see features that are associated to having the outcome during the time-at-risk (risk factors). *       Aggregate covariate: Case series analysis looks at the cases (patients in target with the outcome during time-at-risk) at different time periods (before target start, between target start and outcome start and after outcome end).  This lets users see what cases looked like before the exposure, during the exposure until the outcome and what happens after the outcome.   #Inputs:

Time-to-event Settings:

  •        targetIds: a vector of cohort ids corresponding to the target populations of interest (e.g., drug user cohorts).
  •        outcomeIds: a vector of cohort ids corresponding to the outcomes of interest (e.g., hypertension).

The time to event analysis only requires the target and outcome cohort ids as input.  It then finds the time in days between the target index dates and the outcome dates for any patients in both a target and outcome cohort.  Finally, each case were a patient is in both the target cohort and outcome cohort is categorized as being ‘first occurrence of the outcome’ or ‘subsequent occurrence of the outcome’, and categorized as to whether the outcome occurs ‘before the first target start date’, ‘during the first target start and end’, ‘between two target periods’, ‘during a subsequent target start and end’ or ‘after the last target end’.

Dechallenge- rechallenge Settings:

  •        targetIds: a vector of cohort ids corresponding to the target populations of interest (e.g., drug user cohorts).
  •        outcomeIds: a vector of cohort ids corresponding to the outcomes of interest (e.g., hypertension).
  •        dechallengeStopInterval: The number of days after the outcome occurs in which the drug must stop being prescribed/dispensed for a dechallenge to have occurred.  For example, if the value is 30, then the outcome of interest must occur while a patient is exposed to the drug and the patient must stop being recorded as having the drug within 30 days after the outcome date.
  •        dechallengeEvaluationWindow: The number of days …

Aggregate covariate Settings:

  •        targetIds: a vector of cohort ids corresponding to the target populations of interest (e.g., drug user cohorts).
  •        outcomeIds: a vector of cohort ids corresponding to the outcomes of interest (e.g., hypertension).
  •        minPriorObservation: The number of days a patient must be observed in the database prior to target and outcome dates for the dates to be included in the analysis.  This is so that there is sufficient data recorded for the patient prior to analysis index.
  •        outcomeWashoutDays: The minimum time between outcome dates for the outcome date to be included.  If an outcome occurs within less than outcomeWashoutDays after a previous outcome, that outcome date is excluded from the analysis.  This is due to that date being considered a continuation of the prior outcome rather than a new occurrence.
  •        riskWindowStart: The number of days to add to the startAnchor that represents the start of the time-at-risk.
  •        startAnchor:  Whether the time-at-risk start is relative to the target cohort start date or end date.
  •        riskWindowEnd: The number of days to add to the endAnchor that represents the end of the time-at-risk.
  •        endAnchor: Whether the time-at-risk end is relative to the target cohort start date or end date.
  •       covariateSettings: The features to extract for the cohorts for the risk factor, database comparison and cohort comparison analyses.
  •        caseCovariateSettings: The features to extract for the cohorts for the case series analyses.
  •        casePreTargetDuration: A period of time in days before the target date to include in the case series analysis.  Before covariates are constructed between casePreTargetDuration days before the target start date until 1 day before the target start date.
  •        casePostOutcomeDuration: A period of time in days from the outcome date to include in the case series analysis.  After covariates are constructed between the outcome start date until casePostOutcomeDuration days after the outcome start date.
  •        extractNonCaseCovariates: If true the aggregate covariates will be generated for the target and outcome cohorts in addition to the case (patients with the outcome during the time-at-risk period after the target index) cohorts.

Outputs:

Target characterizations:

  •        Database Comparison: This lets you compare target cohort features across databases.  Targets are restricted to first occurrence with user specified minimum prior observation in the database.
  •        Cohort Comparison: This lets you compare two target cohort features within the same database.  Targets are restricted to first occurrence with user specified minimum prior observation in the database.  

Target with outcome characterizations:

  •        Risk Factor Analysis binary: This lets you compare binary features between non-cases (patients in the target cohort who do not have the outcome in the outcome washout period prior to index and do not have the outcome during the time-at-risk relative to index) and cases (patients in the target cohort who have the outcome during the time-at-risk relative to index) in the same database.
  •        Risk Factor Analysis continuous: This lets you compare continuous features between patients in the target cohort and cases (patients in the target cohort who have the outcome during the time-at-risk relative to index) in the same database.
  •       Case-series: This lets you compare binary/continuous features for the cases (patients in the target cohort who have the outcome during the time-at-risk relative to index) in the same database at three different time periods.  The first period is prior to target index (between casePreTargetDays days before index to index), the second period is between target index to outcome index (between the target start date until when the patient has the outcome) and the third time period is after outcome index (from outcome index until casePostOutcomeDays days after).
  •       Time-to-event:  This shows the distribution of when the outcome occurs relative to target index for the cases. 
  •        Dechallenge-rechallenge:  This tells you how often the outcome occurred shortly before the target end date (i.e., if the target is a drug exposure, how often the drug was discontinued after the outcome started).  This is considered a dechallenge. For patients with a ‘dechallenge’ we then look at whether the drug was later restarted (a rechallenge) and if so, whether the outcome reoccurred shortly after the rechallenge (if so it is a rechallenge failure) otherwise it is a rechallenge success. If you see many rechallenge failures then it suggests a relationship between the target and outcome.