Creating observation periods

In the OMOP CDM, the observation period table defines

spans of time during which two conditions are expected to hold: (i) Clinical Events that happened to the Person are recorded in the Event tables, and (ii) absence of records indicate such Events did not occur during this span of time. https://ohdsi.github.io/CommonDataModel/cdm54.html#observation_period

There are many nuances to defining observations periods, and different definitions can have a major impact on studies that rely on observation time (such as incidence and prevalence studies that require identifying a suitable denominator population, survival studies which require follow-up into the future, and so on).

In HERON-UK we have defined two conventions for constructing observation periods, depending on whether source data has a start and end date for individuals or not. In addition to these conventions, which are summarised below, we have also developed the OmopConstructor R package which can be used to manipulate observation periods on a study-by-study basis.

Source data that contains a start and end date for individuals

In some data sources, such as primary care data, enrollment/ registration and exit dates are available for individuals. In this case, we use these dates to set observation_period_start_date and observation_period_end_date, respectively. In the case where individuals have multiple entries into and exits out of the data, we keep only the first entry per person to define the observation period. (Note, in HERON-UK we do not have any insurance claims data where such a rule may be considered too stringent).

Source data that does not include a start and end date for individuals

In many data sources, there is no defined start and exit date for individuals. In such cases, we define observation periods using the following approach:

  • Observation period start date: the date of the first record for a person.
  • Observation period end date: the earliest of the following:
    • Date of death
    • A censoring age (commonly 120 years, though this can vary; e.g., pediatric databases may use 18 years)
    • Other censoring events, such as migration out of the study region.
    • Date of data extraction

This definition results in a single observation period per person. (Note, all data partners in HERON-UK have mortality data - in the absence of death data, additional censoring rules would likely be needed).

This convention is designed to err on the side of including as much possible time for individuals. When running studies, however, the OmopConstructor package can be used to apply study-specific persistance and surveillance windows. See our OHDSI Europe 2025 poster for more information on approaches to and implications of methods for defining observation periods, and the OmopConstructor R package for implementing them on a study-by-study basis.

Defining observation periods when using linked data

Where there are multiple source data that are being linked, we define observation periods as the time in which individuals can be observed in all datasets. For example, if data source A went from from 2014 to 2023 and data source B went from 2018 to 2025, then an OMOP CDM with linked data from A and B would be expected to have observation periods from 2018 to 2023.