ebmdatalab / open-nhs-hospital-use-data

For analysis of https://opendata.nhsbsa.net/dataset/secondary-care-medicines-data
0 stars 0 forks source link

Categorising dmd drugs into categories #12

Open HelenCEBM opened 2 years ago

HelenCEBM commented 2 years ago

Use case: some of our topics have many dm+d codes and so it will be helpful to present them in groups of similar drugs.

Options:

  1. BNF chemical/paragraph
  2. Ingredients
  3. Whole products

Details:

  1. For products with BNF codes we can extract BNF chemicals to link to DM+D products. Drawbacks:

    • some products in DM+D do not have BNF codes so may be missing a category using this method.
    • In the BNF each drug can only have one parent so products with multiple ingredients can only have one category (this is not always a drawback but sometimes we may want to count by ingredient instead.
  2. We can use DM+D to extract all the active ingredients for each product and link these to a category Drawbacks:

    • some drugs may have 3 or 4 active ingredients so this can produce a large number of categories and care must be taken not to double count
    • Difficult to find a mapping for all ingredients to categories
  3. Each drug can be classified using SNOMED hierarchies. Drawbacks:

    • some drugs may be mapped to 2 different categories so care must be taken to handle these appropriately and not double count
    • Difficult to find a mapping for all products to categories
Jongmassey commented 2 years ago

re: 3

HelenCEBM commented 2 years ago

As an example, for antibacterials I've used (for now at least) BNF paragraphs (but with some regrouping of the smaller ones) .

I found a paragraph for each ingredient in products identified by BNF code, and then mapped ingredients back to paragraphs for additional products identified in wider DMD search which don't have BNF codes. https://github.com/ebmdatalab/open-nhs-hospital-use-data/blob/80169a55f4c9162c101f862a835f5a52aeae6092/notebooks/antibacterials.ipynb

There is of course a duplication issue with some multi-ingredient products mapping to multiple paragraphs but this was rare.

Jongmassey commented 2 years ago

The WHO AWARE classification system has now been imported into bigquery and matched at a VMP level and is available at ebmdatalab.jonm.who_aware_vmp

I'd be interested to see what proportion of antibacterial drugs you have identified @HelenCEBM which are missing from this classification (as processed and matched by me)

HelenCEBM commented 2 years ago

Thanks Jon - there are some VMPs with 2 different categorys - is this because they have 2 ingredients with different classes? If so can I use the "highest" category?

sebbacon commented 2 years ago

Just seagulling to say that compound medications have come up many times before in the pretimes, and I don't think we ever satisfactorily addressed them. Is this a question to involve pharmacists in? I recall my intuition often being wrong in this regard

Jongmassey commented 2 years ago

@HelenCEBM thanks for the heads up - some products are classified differently according to their route of administration (despite having the same ingredients). Nonetheless I see there are 8 products which are multiply classified despite having no route of admin specified. I'll investigate

Jongmassey commented 2 years ago

The 8 9 I identified are all combo products that are not listed combinations by the WHO so yeah I think the higher category is the best shout - thoughts @richiecroker @brianmackenna @orlamac ? I can put that logic in the view if you want?

<html xmlns:v="urn:schemas-microsoft-com:vml" xmlns:o="urn:schemas-microsoft-com:office:office" xmlns:x="urn:schemas-microsoft-com:office:excel" xmlns="http://www.w3.org/TR/REC-html40">

nm -- Generic Maxitrol eye ointment Generic Maxitrol eye drops Generic Otosporin ear drops Polymyxin B 10,000units/ml / Trimethoprim 1mg/ml eye drops Generic Gregoderm ointment Generic Neosporin eye drops Generic Copal G+C cement 40g sachets Generic Deteclo 300mg tablets Polymyxin B 10,000units/g / Trimethoprim 5mg/g eye ointment

Jongmassey commented 2 years ago

Source for AWARE classification: https://www.who.int/news/item/01-10-2019-who-releases-the-2019-aware-classification-antibiotics WHO-EMP-IAU-2019.11-eng.xlsx

Jongmassey commented 2 years ago

Although for those eye drops... the EML website only lists Polymyxin B under the injection and oral route there's nothing about topical applications. Whereas the Excel spreadsheet doesn't give a route for this compound (so I assumed it applies to all routes of administration).

HelenCEBM commented 2 years ago

Would be good to fix in your logic if possible, once we decide on right approach.

HelenCEBM commented 2 years ago

I have a lot of VMPs without a WHO classification but they all contain ingredients not listed in the WHO spreadsheet. Although two of them (Clindamycin hydrochloride and Gentamicin sulfate) are listed by WHO but without the salts.

Ingredients not in WHO list but found in DMD antibiotics: Aminosalicylic acid Bedaquiline fumarate Capreomycin sulfate Clindamycin hydrochloride Clofazimine Colistimethate sodium Cycloserine Dapsone Delamanid Demeclocycline hydrochloride Ethambutol hydrochloride Ethionamide Fidaxomicin Gentamicin sulfate Isoniazid Methenamine hippurate Metronidazole Metronidazole benzoate Minocycline hydrochloride Nalidixic acid Nitazoxanide Pyrazinamide Sodium fusidate Sulfadiazine Sulfamethoxypyridazine Sulfapyridine Taurolidine Tinidazole Vancomycin hydrochloride

Jongmassey commented 2 years ago

Please let me know if you find any other issues or you think my conclusions on any of these points are wrong.

HelenCEBM commented 2 years ago

Thanks @Jongmassey , sorry just getting back to this now

Jongmassey commented 2 years ago

Have you now updated for 2021 list ? Should we include anything from either 2019 or 21 in case some things have been removed on the latest list but still appear in our older data?

I'll take a look at this now. WHO-HMP-HPS-EML-2021.04-eng.xlsx

As for the unusual routes of delivery (dental, ocular etc.) this will be a mismatch with the specified routes of delivery in the WHO data.

Jongmassey commented 2 years ago

Summary of data quality issues found in 2021 dataset here: https://github.com/ebmdatalab/open-nhs-hospital-use-data/blob/who_warn_update/notebooks/antibiotics/who_warn.ipynb

Some VMPs which previously had classes now do not, largely by virtue of the ingredient (Neomycin and Polymixin B primarily) featuring in the 2019 data without route of administration but in the 2021 data listed separately for IV and Oral routes. The "missing" VMPs are topical preparations and thus do not match the routes of administration specified.

HOWEVER for Neomycin and Polymixin B the class and AWARE category are the same regardless of the route of administration (albeit with different ATC codes) so perhaps the logic of the SQL view that does the matching to VMP could be altered to take account of this. There are certainly cases where different routes of administration have different AWARE categories and thus the question of how to categorise drugs of the same ingredient but differing route of admin arises (or indeed where the VMP has multiple routes in the droute table, some matching the AWARE dataset, some not).

2021 data available in ebmdatalab.jonm.who_aware_vmp_2021, 2019 data remains untouched