Best way to get date of last clinical encounter?
I would like to do survival analysis with my cohort, but it requires me to know the date of each individual's last clinical encounter so I can calculate follow-up time. I noticed there is a date lost to follow-up column, but only 3 out of over 6,000 in my cohort have values in this column. I was wondering if anyone has suggestions on which columns I should use instead? There are a lot of date-related columns and while ideally I would check them all and use the latest date listed, I don’t know that that is realistic.
I have considered checking for death dates, latest GP record, most recent hospital admission/discharge date, and last UK Biobank contact date, and taking the latest date from all of these columns. Does this sound reasonable, or am I missing anything that would be good to check as well?
Thank you for your help!
Comments
10 comments
I have the same doubt
Hi Courtney and Shengan,
Field 191 date lost to follow up will only ever be relevant to a few participants, as it does not include those participants for whom UKB has received notification of death from the Death Registry. For more details see Resource 481516 . At present, field 191 is also out of date, as it was last updated May 2017, see Version Date on Showcase page for field 191 .
We are planning to update fields 190 and 191 in future but currently aren't able to provide a timeline for this. Once a release date has been determined it will be published on the future timelines page of the website. 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:
Thank you for using the forum.
Thank you for your timely reply!
Thank you, that information is very helpful!
Dear Rachael and Community team,
I have also noticed Data-Field 20143 called “Date of last personal contact with UK Biobank”. If the diagnostic criteria of endpoint is based on ICD-code, for participants without Death data, Hospital inpatient data, or Cancer data, is it reasonable to use Data-Field 20143 as the alternative censoring date?
Thank you for your help.
Hi Shengan, I can't really help with that question. I am not an epidemiologist, and UKB does not advise researchers on that sort of detail. However, please note that 20143 does not tell us about the health of the participant. If the last contact in 20143 is based on the participant filling in an online questionnaire such as the Sleep questionnaire, then we do not know what diseases the participant might have been suffering from at the time. We know they were still alive at that point, but we already know that from the absence of any record in the Death Register.
Thank you for your advice!
Dear Rachael, sorry for bothering you again.
I encountered a new problem. For Cancer record origin (Data-Field 40021), many of the records were from “National Cancer Intelligence Network”. However, regarding Cancer data, Data providers and dates of data availability doesn't include the censoring date for “National Cancer Intelligence Network”, and I failed to find relevant explanations in other documents. Could I possibly trouble you for some advice?
Hi Shengan,
that is a good question!
According to Wikipedia , “the NCIN is a UK organisation that attempts to work closely with cancer services in England, Scotland, Wales and Northern Ireland but the majority of the reports are at an England level only”. It also says “The National Cancer Intelligence Network (NCIN), was set up in 2008 to drive improvements in care standards and clinical outcomes. NCIN is now part of Public Health England, following the Health and Social Care Act 2012.”
Field 40021 Cancer record origin mentions 12520 records with origin NCIN. The associated dates for those records are in Field 40005 Date of cancer diagnosis
I have had a look at the record dates for the NCIN records. I haven't checked them all, but at a quick glance, it seems that the latest is from 2012.
This would fit with the idea that UKB received historical cancer data up to 2012 from NCIN, and any more recent records from other sources. Thus, the NCIN records don't need a censoring date because the NCIN data is complete.
Hi Rachael, Thank you for your valuable and timely suggestions!!
Please sign in to leave a comment.