Censoring date of Hospital Inpatient Data?
I'm conducting a survival analysis using hospital inpatient data, with depression as the outcome. To calculate follow-up time, I need to determine the appropriate censoring date.
The UK Biobank resource (https://biobank.ndph.ox.ac.uk/showcase/exinfo.cgi?src=Data_providers_and_dates) lists the HES data censoring date as 31 October 2022, but I found a diagnosis record dated 09 November 2022 in UKB-RAP.
Is there a recommended or confirmed endpoint to use for follow-up in survival analyses?
Thank you!
Comments
3 comments
Hi,
Have you upgraded to the new v19 data release? In my dataset from the previous release, 2022-10-31 is indeed the last date. Having upgraded, I do see participants who have epistart and epiend dates higher than October, to a max of 2022-11-09, as you indicated. However, the number of participants is in the single digits as far as I can tell.
I haven't looked at the other tables in the new release for most recent censoring date, but it is possible to get newer dates for tens of thousands of participants from participant and covid19_result tables even with the older release (mostly from Population characteristics > Ongoing characteristics > Date of last personal contact with UK Biobank and Assessment centre > Recruitment > Reception > Date of attending assessment centre).
Best of luck!
Hello Po-Wen Ku,
Thank you for posting your query on the UK Biobank Community Forum (and thank you, Mehmet Altan Orhon for your comment above).
The Data providers and dates of data availability page gives the current suggested censoring dates for each type of linked health data. Censoring dates are estimated as “the last day of the month for which the number of records is greater than 90% of the mean of the number of records for the previous three months, except where the data for that month is known to be incomplete in which case the censoring date is the last day of the previous month.”
The censoring dates are not applied by UK Biobank to the data made available to researchers which will always contain the latest data regardless of censoring dates, and may include incomplete data after the dates shown on the page. The suggested dates are intended for guidance only. Researchers should censor outcomes based on their own research protocol.
In the meantime, based on feedback from our Epidemiology team, we would recommend using the earliest of the following to determine clinical outcome and censoring dates in analysis:
Kind regards,
Aditi
Hello Aditi S and Mehmet Altan Orhon,
Thank you for your suggestion. I will define the study endpoint based on my study design. I also appreciate the detailed information provided on the Data providers and dates of data availability page, as well as the reminder about the censoring dates for analysis.
Best regards
Please sign in to leave a comment.