Censoring of algorithmically defined dementia in UKB v20

Katherine Elizabeth Ogilvie Taylor

Hi All,

I'm conducting a survival analysis with time to dementia as my primary outcome. I am planning on using algorithmically defined dementia. 

  • Why does the maximum date of all cause dementia on the RAP not match the maximum date of the UKB showcase?
  • What is the case cut-off/censoring date for algorithmically defined dementia in v20? Is the maximum date of the showcase correct e.g. 2024-11 or is it later up to data entry e.g. 2025-09 or a particular cut-off from one of the datasets used to define dementia?

Thank you!

Comments

1 comment

Date Votes
  • Comment author
    Molly D Data Analyst The helpers that keep the community running smoothly. UKB Community team
    • Edited

    Hi Katherine,
     
    I think the best place to start is by checking your RAP project is up-to-date, as researchers sometimes see discrepancies when older project versions are used. Refreshing your project ensures your counts align with the UKB Showcase. For guidance, see: Why is data missing from my UKB-RAP project.
     
    Regarding the differences between raw linked health data on the RAP and versioned algorithmically defined outcomes (ADOs) shown in the Showcase:

    • The UKB-RAP contains the latest linked health records (hospital inpatient, death registry, primary care), updated according to the data you have access to. Event dates on the RAP can extend beyond the algorithm-defined cut-off used in the versioned ADOs.
    • Algorithmically defined dementia (ADO) is a derived, versioned outcome, generated at a specific point in time using predefined rules and fixed snapshots of the linked datasets. Newer records that appear on the RAP after the algorithm was last run will not be reflected in the ADO fields until a future release.
    • There is no separately published censoring date for algorithmically defined dementia. The effective cut-off is simply the latest event date present in the ADO field itself.

    For data release v20, this corresponds to the maximum date shown in the UKB Showcase for all-cause dementia (≈ November 2024). Dates later than this (e.g., into 2025) that appear on the RAP reflect newer raw linked data, not additional cases included in the v20 algorithm.
    If you are using algorithmically defined dementia as provided in v20, follow-up should therefore be censored at:

    1. The latest ADO dementia date in v20 (≈ Nov 2024), or
    2. Earlier death or withdrawal.

    Using later RAP dates would require constructing a bespoke dementia definition directly from the raw linked datasets and justifying censoring based on data-provider completeness guidance.
     
    Additional resources:

    Kind regards,

    Molly

     

    0

Please sign in to leave a comment.