Defining a Townsend local area deprivation quintile in ATLAS as provided by the THIN database(?)

lhjohn · September 17, 2020, 7:24am

Hi,

I am replicating dementia prediction models from literature and creating cohorts in ATLAS that will be used as custom covariates in a Cox regression model. The model that I am currently replicating is by Walters (Link) where one of the model’s predictors is social deprivation, defined as the Townsend local area deprivation quintile with a scale from 1=least deprived to 5=most deprived.

The development data of Walters’ model comes from the THIN database in the UK, which I assume has information about the Townsend local area deprivation quintile. Can anyone working with THIN UK data confirm this?

So I started looking for ways to model this scale, but have only been able to define a binary variable from it as follow:

I wonder if anyone has experience with the THIN database, or maybe was involved in mapping it to the OMOP CDM, and can give me a hint on where to find back the Townsend local area deprivation quintile in one of the CDM’s standard vocabularies.

Daniel_Prieto · September 17, 2020, 7:45am

Hi, not that much experience with THIN (renamed to UK IMRD) but in CPRD this is only obtained through data linkage… so my guess is you don’t have access to this? @sseager ?

sseager · September 17, 2020, 9:11am

@Daniel_Prieto @lhjohn we use UK IMRD (previously known as THIN) a lot, but have not undertaken any analysis that required deprivation.
If we were to, then we would also have to link it ourselves

danielmorales · September 17, 2020, 9:20am

THIN used to have it but it became unavailable as the linkage had not been refreshed (this was just before it became UK IMRD). In any case this was practice based deprivation as opposed to patient level. CPRD has/had both.

lhjohn · September 17, 2020, 9:38am

This means one solution would be to link patients/GPs postal codes to population census data, which in my case will likely not be feasible as I want to externally validate on database that may not have this information.

I will look into approximating the social deprivation in another way.

Thanks everyone for helping me out with this!