Diabetes + diabetes complication codelist revisions

[finersarah]

Revision process:

@mmah-sh has collated all previous codelists for specification of diabetes type using EHR data, including CALIBER, OpenSAFELY, and others to try and refine our diabetes codelists. We will use this for further clinical curation, as follows:

1. Clinical revision of diabetes type - important due to the number of diabetes codes that do not specify type. This refers to both diabetes diagnosis and complication codes:
   • I have allocated all codes to type 1 or type 2 diabetes if type 1 or type 2 specified in the clinical code
   • If a high probability of diabetes type can be inferred from code description (e.g. young onset diabetes with ketoacidosis) I have allocated to that type
   • If diabetes type not specified but clinical code indicates non-insulin-treated diabetes, I have allocated to type 2 diabetes

This leaves a large list which includes:

• unspecific codes - which do not define diabetes type, and which may describe a diabetes diagnosis and/or diabetes complication, e.g. "diabetes" or "insulin-dependent diabetes". If these are diagnosis only, it is possible that more contemporary coding will have reallocated them to a specific diabetes type. If they are a complication code, that they are preceded by a specific diagnosis code. The aim is to reduce the size of this list and allocate as many of these to type 1 or type 2 lists. If this cannot be done with clinical codes, we may need to review insulin prescribing data.
• rare diabetes types - e.g. cystic fibrosis related, maternally inherited diabetes

Next steps: @mmah-sh to further curate the unspecified diabetes types by doing the following: (a) Calculate the frequency of use of all codes in the unspecified/rare list (b) For those in the unspecified/rare list that are marked "needs revision", calculate the frequency of which an unspecific disease only code is superseded (i.e. a later date) by a specific disease only or complication code (c) For those in the unspecified/rare list that are marked "needs revision", calculate the frequency of which an unspecific complication code is preceded (i.e. an earlier date) by a specific disease only code

2. Clinical revision of diabetes diagnosis vs complication lists
   • we still do not know if the codes which describe both a diabetes diagnosis and complication are also accompanied by a diagnosis only code. Currently, we duplicate codes across multiple lists which will overestimate prevalence.

• NB our complications lists describe long-term complications of diabetes only. We do not have separate conditions lists for acute complications (e.g. DKA, hyposmolar coma, poor control, Diabetes Code List_SF.xlsx or hypoglycaemia) so these are not denoted as complications.

Next steps: @mmah-sh to calculate the following for all complication codes: (a) Calculate the frequency of use of all complication codes (b) Calculate the frequency where a complication code and diagnosis only code co-occur (at any point in the EHR).

[Fabiola-Eto]

As discussed with @finersarah the Diabetes codelist (CPRD_Diabetes.csv, n= 390) will be split into three codelists:

• Type1 Diabetes (CPRD_Type1_Diabetes.csv, n= 82)
• Type 2 Diabetes (CPRD_Type1_Diabetes.csv, n= 116)
• Unspecified or Rare Diabetes (CPRD_Unspecified_Rare_Diabetes.csv, n= 192)

The codelists were updated and uploaded to the branch codesets-included-conditions.

[Fabiola-Eto]

NB in CPRD GOLD data there are some patients with diabetes (n = 242,237) that received different combinations of diabetes diagnosis over time:

• 764 (0.3%) has both T1D and T2D
• 11,200 (4.6%) has both T1D and Unspecified/Rare Diabetes
• 105,271 (43.5%) has both T2D and Unspecified/Rare Diabetes
• 2,254 (0.9%) has T1D, T2D and Unspecified/Rare Diabetes

In agreement with @finersarah and @Miriam-S-git we decided that:

• If a patient with Unspecified/Rare Diabetes codes has ever been recorded with either T1D or T2D, then the specific diagnosis code will prevail over the unspecified one in our analysis.
• Patients who received both T1D and T2D codes over time (n = 3,018, 1.2%) will be classified with an Unspecified/Rare Diabetes code in our analysis.