In PatientProfiles: Identify Characteristics of Patients in the OMOP Common Data Model

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  message = FALSE,
  warning = FALSE
)

Introduction

The event functions identify which event occurs first or last within a time window. They can search either a cohort table or a concept set:

| Event source | Return the date | Return days relative to the index date | |:--|:--|:--| | Cohort table | addCohortEventDate() | addCohortEventDays() | | Concept set | addConceptEventDate() | addConceptEventDays() |

Each function adds two columns for every window:

• an event column containing the name of the selected event; and
• a date or days column containing when that event occurred.

Unlike the intersection functions, the event functions also return the applicable boundary when no target event occurs in the requested window. This makes it possible to distinguish an observed event, an explicit censoring boundary, and the end of observation.

We will use the GiBleed dataset to illustrate this functionality:

library(PatientProfiles)
library(dplyr)
library(omock)
library(CohortConstructor)
library(CodelistGenerator)

cdm <- mockCdmFromDataset(datasetName = "GiBleed", source = "duckdb")

Let's create some simple cohorts:

codelist <- getDrugIngredientCodes(
  cdm = cdm,
  name = "acetaminophen",
  nameStyle = "{concept_name}"
)
cdm$my_cohort <- conceptCohort(
  cdm = cdm,
  conceptSet = codelist,
  name = "my_cohort"
) |>
  requireIsFirstEntry()
codelist <- list(osteoarthritis = 80180L, diverticular_disease = 4266809L)
cdm$conditions <- conceptCohort(
  cdm = cdm,
  conceptSet = codelist,
  name = "conditions"
)

Finding a cohort event

addCohortEventDays() searches the cohorts in targetCohortTable. In this example we find the first event on or after each record's cohort start date and up to 365 days later.

cohort_event_days <- cdm$my_cohort |>
  addCohortEventDays(
    targetCohortTable = "conditions",
    indexDate = "cohort_start_date",
    order = "first",
    window = list(next_year = c(0, 365))
  )

cohort_event_days |>
  glimpse()

The days_next_year value is relative to cohort_start_date: positive values are after the index date, zero is the index date, and negative values are before it.

We can obtain a quick summary of the result:

cohort_event_days |>
  group_by(event_next_year) |>
  summarise(
    n = n(),
    median_days = median(days_next_year)
  ) |>
  collect()

• censor means that no event was found before the window or censoring boundary;
• end_of_observation means that no event was found before the relevant observation period boundary; and
• any other value is the name of the event that was found.

To find the nearest previous event, use a window ending on the index date and order = "last". Here, "last" selects the latest event in the window. Because the complete window is in the past, this is the event nearest the index date.

cohort_event_date <- cdm$my_cohort |>
  addCohortEventDate(
    targetCohortTable = "conditions",
    indexDate = "cohort_start_date",
    order = "last",
    window = list(previous_year = c(-365, 0))
  )

cohort_event_date |>
  glimpse()

Use targetCohortId to restrict the search to particular cohorts. targetDate determines which date in the target cohort table represents the event and defaults to cohort_start_date.

Finding a concept event

The concept event functions search records associated with one or more concept sets across the OMOP clinical tables. The names of the concept sets become the event names.

concept_event_days <- cdm$my_cohort |>
  addConceptEventDays(
    conceptSet = codelist,
    order = "first",
    window = list(next_event = c(0, Inf))
  )

concept_event_days |>
  glimpse()

For concept events, targetDate can be either event_start_date, the default, or event_end_date. The concept set is validated against the vocabulary in the CDM before the clinical records are searched.

First, last, and censoring

The window defines the period to search, while order determines which event within that period is returned:

• order = "first" selects the earliest event in the window;
• order = "last" selects the latest event in the window;
• a window such as c(-365, 0) searches before indexDate, while c(0, 365) searches after it; and
• both window endpoints are included.

The search is always restricted to the observation period containing the index date. For a forward search, a column supplied through censorDate can shorten the available follow-up. If no event occurs before the applicable boundary, the event column contains:

• "end_of_observation" when the observation period boundary is reached; or
• "censor" when censorDate or the requested window boundary is reached.

The accompanying date or days value is the applicable boundary.

For a forward search, the boundary is the earliest of the upper window boundary, censorDate, and the observation period end. For a backward search, the boundary is the latest of the lower window boundary and the observation period start.

An event on the boundary is returned as the event itself. It is not combined with "censor" or "end_of_observation". If the index date is outside observation, both added values are missing.

Events occurring on the same date

Several cohorts or concept sets can occur on the selected date. The multipleEvents argument controls how these ties are represented:

• NULL, the default, returns the first matching event in the original order;
• TRUE returns all matching event names in alphabetical order, separated by "; "; and
• a character vector defines a priority order and returns the first matching name. Event names omitted from the vector follow in alphabetical order.

Use NULL to follow the original order of the concept set:

cdm$my_cohort |>
  addConceptEventDays(
    conceptSet = codelist,
    order = "first",
    window = list("next" = c(0, Inf)),
    multipleEvents = NULL
  ) |>
  group_by(event_next) |>
  summarise(n = n()) |>
  collect()

Use TRUE to combine all events that occur on the selected date:

cdm$my_cohort |>
  addConceptEventDays(
    conceptSet = codelist,
    order = "first",
    window = list("next" = c(0, Inf)),
    multipleEvents = TRUE
  ) |>
  group_by(event_next) |>
  summarise(n = n()) |>
  collect()

Alternatively, provide an explicit priority order. Here, osteoarthritis is selected before diverticular_disease when they occur on the same date:

cdm$my_cohort |>
  addConceptEventDays(
    conceptSet = codelist,
    order = "first",
    window = list("next" = c(0, Inf)),
    multipleEvents = c("osteoarthritis", "diverticular_disease")
  ) |>
  group_by(event_next) |>
  summarise(n = n()) |>
  collect()

In this example, 390 cohort records have both osteoarthritis and diverticular disease on the selected date. The three calls above show how multipleEvents can resolve these ties using the original order, a combined label, or an explicit priority.

Multiple windows and column names

More than one window can be supplied in a named list. The default nameStyle = "{value}_{window_name}" uses {value} for event, date, or days, and {window_name} for the name of the window.

events_in_two_directions <- cdm$my_cohort |>
  addCohortEventDays(
    targetCohortTable = "conditions",
    order = "first",
    window = list(
      before = c(-365, 0),
      after = c(0, 365)
    )
  )

events_in_two_directions |>
  glimpse()

The same value of order applies to every supplied window. If different windows require different values of order, use separate calls.

nameStyle must contain {value} so that the event and its date or days have different names. It must also produce a unique name for every output column.

PatientProfiles documentation built on Aug. 22, 2026, 1:07 a.m.