Date of death variables in ECHILD - HES Mortality and HES APC

[yaszadeh]

Question / Problem statement

Hello!

I'd really appreciate any advice on how best to derive date of death in my ECHILD project.

I have both HES-APC (2010-2022) and HES-Mortality (FILE0184888_civil_reg_deaths) data.

(1) In HES APC: I have generated date of death (YYYY-MM-DD) based on discharge method and destination (as in script 10 of Matt Jay's 'deliveries cohort' pipeline https://github.com/UCL-CHIG/deliveries-cohort).

(2) In HES Mortality: I was expecting to have more information from REG_DATE_OF_DEATH in format YYYY-MM-DD (as listed in the ECHILD data dictionary [ECHILD_DataCatalogue_v3_20251017]: "Date on which the patient died. Where ‘Death Record Used’ = ‘MIX1’, ‘HES1’ or ‘HES2’, the date of death from HES (discharge date or episode end date of death episode) is used. Otherwise, this date is taken from the ONS death record.")

However, REG_DATE_OF_DEATH has only been given to me in YYYY-MM. Is this as expected? I have set the day (DD) to be 01 for everyone, which isn't ideal for identifying erroneous HES activity after death in my cohort.

• Should I have received REG_DATE_OF_DEATH in YYYY-MM-DD format as suggested in the ECHILD data dictionary?
• If yes, I could use Hes-Mort date if present, HES-APC if not
• Otherwise, I'm using HES-APC YYYY-MM-DD if present, and HES-Mort YYYY-MM-01 if not.

Thank you for any insight or advice!
Yasmin

Compliance checklist

• I confirm I have not entered any information from within the ONS SRS (or only information that has been cleared for release by the ONS SRS output checking team).
• I confirm I have asked only one question in the textbox above.
• I confirm I have checked for similar questions amongst existing Q&As.

[zcabh35]

Hello below is a combined response from the ECHILD Team and the Early Careers Research Data Clinic Team:

Should I have received REG_DATE_OF_DEATH in YYYY-MM-DD format as suggested in the ECHILD data dictionary?

• We're viewing the data which the ONS has been shared with you and can see that you should have a variable called "REG_DATE". This should be YYYYMMDD. Apologies, there are many variables with date values in them. Could you kindly confirm whether you see this variable or not?

In terms of using HES APC and ONS death dates, the following has been suggested: "I would use the ONS/civil registration date of death as the primary source where available, and use HES APC death episodes as a supplementary source to identify additional deaths indicated in hospital records, for example using discharge method = 4. So the logic would be: ONS DOD​ takes precedence, and HES-derived DOD​ is only used where ONS DOD​ is missing."

[yaszadeh]

Thank you very much for collecting these responses and for replying so quickly.

I do have "REG_DATE" in YYYYMMDD format. I didn't initially select this variable as the data catalogue description says that this is the "Date on which the death was registered". I expect the registration date will often differ from the actual Date of Death, which should be recorded in "REG_DATE_OF_DEATH" (YYYYMMDD) and "REG_DATE_OF_DEATH_YYYYMM" (YYYYMM) (but those latter two variables are actually equal in my dataset in YYYYMM).

Despite it being just the date of registration, I assume the recommendation would still be to use "REG_DATE" as the primary source where available, and the HES APC death episodes as a supplementary source.

Thank you again,
Yasmin

[zcabh35]

Dear Yasmin

I must heavily apologize, You are right the REG_DATE is the date of registration and is not the actual date of death. After having three people checking to ensure we got this correct, our own CURRENT copy of the civ_reg_deaths data has our REG_DATE_OF_DEATH variable truncated to YYYYMM too. We believe in the past we had this in YYYYMMDD and about to email the SRS support team to see what's going on. We'll get back to you once this is being fixed.

In the mean time, I can say the following, when we compared the Date of Death and Date of Registration in an older copy of our data, the median difference was rather small, although even if your copy, you can see that the YYYYMM diff between the both can be rather large.

For you to power on with your work, specifically coding, may I kindly suggest using the following proxy for now: If the YYYYMM between the REG_DATE and the REG_DATE_OF_DEATH are the same, use the REG_DATE for now. If the REG_DATE and the REG_DATE_OF_DEATH is vastly different based upon YYYY and MM, please bang on "-15" to the end of the REG_DATE. This should allow you to power on with your coding in the mean time.

We will contact the SRS team to try and resolve this.

Again apologies for this!

[zcabh35]

Howdy

I have bad news. It seems that we are not actually allowed access to the exact date of death and month and year is supplied only.
In the up coming week, I'll post a script here to demonstrate how we derived the date of death to reduce the residuals

Apologies for the inconvinience

Best Wishes
Batman

[yaszadeh]

Hello!

Thank you so much for investigating and keeping me updated along with suggestions and solutions. I really appreciate your support and will look out for your code when it's ready.

Best wishes,
Yasmin