finersarah
commented
3 years ago The diabetes codelist is problematic for the following reasons:
- It merges all types of diabetes, despite significant differences in their aetiology and pathogenesis. However, it is very challenging to separate the different types of diabetes out using clinical codes due to (a) historic codes such as "insulin-dependent diabetes" that do not discriminate, and (b) misclassification of diabetes type which is a common problem and likely ethnicity-patterned. There are some studies that have developed algorithms to ascertain T1D and T2D from electronic health records with good accuracy (e.g. https://onlinelibrary.wiley.com/doi/full/10.1002/pds.4377?casa_token=QV5HohKVmP4AAAAA%3ALeNVSZceYEAwUh9Fh2YMvzVfHm4ePTRbT0KkJE1dWXb24ytn_SpRYYtPDDFizCdzvMKG3DKTtdDAgQ) but these are not validated in multi-ethnic populations. NB, separate work is taking place led by S. Finer to estimate rates of misclassification of T1D and T2D in EHRs in people of south Asian origin, using genetic risk scores, and this will be able to inform approach to diabetes classification in this study at a later date.
- It contains multiple codes that relate to both diabetes diagnosis and complications. These will be frequently duplicated across codelists for diabetic neuropathy, diabetic eye disease and peripheral vascular disease. As such, we may over-estimate rates of complications.
- There is variable association between diabetes diagnosis and complications in the coding architecture, and therefore it is difficult to generate a structure for inclusion/exclusion of these codes.
- There are temporal changes in the specificity of these codes, e.g. the less specific codes (e.g. 'brittle diabetes') are more likely to be used historically, compared to more specific codes (e.g. 'type 2 diabetes with mononeuropathy') which are more likely to be used more recently
- There are rare types of diabetes, e.g. cystic-fibrosis related diabetes, that are likely to be accurately coded, which are included in this codset and which have significantly different aetiology. However they have some common features with regards pathogenesis.
Plan
- We include all diabetes codes (excluding pregnancy-associated codes) for data extraction now
- We look for specific patterns in the use of 'diagnosis + complication' codes - i.e. do they always follow a 'diagnosis-only' code, and therefore can they be placed only in the complications code sets
- We consider using an existing algorithm to separate diabetes types and, at the least, use this in a sensitivity analysis for clusters that included diabetes.
- We await the outcome of Sarah's study on misclassification rates of diabetes in south Asians
Fabiola-Eto
@f-eto please could you check the attached codelist against the existing diabetes codelist and let me know of any differences? Many thanks
Please note the remission/resolved codes for your files
Diabetes_SF.xlsx