NHS reform or die

Oddly they are quite pleasant and don’t each have two heads or anything.

They are lovely at our practice. And the 8am daily queue outside the practice trying to get past the 8am phone game are always pleasant to them too. I think they have an impossible job to be honest. When I last asked in person for a non-urgent appointment, I was told there were none available. For at least four weeks and even then the calendar was closed as they booked only up to three weeks in advance. Kafka would be impressed. Blood taking, when available is prioritised for those who can’t easily travel to the nearby hospital. I’m sure matters are better elsewhere. At least I hope they are, because they could not be worse.

I would hope so! And hence we have projects like the UK Biobank to direct that learning by linking genome data to healthcare records. YOUR healthcare records (if you are one of the 500,000 who signed up). That’s a big enough sample set for any inference.

So the training data is about 1% of the real population.  And do you think there are any inherent bias’s in that sample? e.g. how many non-english speakers agreed to give their data? under 16s? homeless people?  and they are just the most obvious examples of how admin engagement with “the system” can get stuck without considering if there are health record issue.

 wouldn’t call those simple issues. Name matching is a nightmare. Tying it all together to a single identity, including of course the possibility to correct that identity in all those different systems… Simple it isn’t.

Simple may be the wrong word – but it doesnt need (or necessarily helped by) AI.

Surely AI would rely on DOB and unique NHS number too?

If you’ve got robust fields like that – no need for AI, its just data processing then.

I’d rather rely on proper systems than human checking. Humans make mistakes too.

Well I can tell you about a pharmacy that had mother and daughter both hand in prescriptions for insulin the same week (they are both T1D) and when 16 yr old daughter went to collect hers, the “check” they asked was the address so failed to spot they had given the wrong type of insulin to the patient.  Fortunately the 16 yr old was able to point out the error and it was corrected (I bet it didn’t even get logged as a near miss).   Your challenge is to define “proper systems” – that means designed not left to “learn”.  I’ve absolutely no concerns about proper engineering being involved in medical software, but black boxes which seem to give the right answer almost all the time are hard to robustly validate and prone to inherent systemic biases.

BTW “AI” seems to be a recent catch-all for data science and analytics.

That’s just lazy and I expect that on this IT forum for people with bicycles that we can be better than that!  There are really useful applications of AI in medical data analytics etc, I just don’t think that linking 100 databases/systems is one of them.

As for taking blood, I don’t think I can recall the last time a  GP took my blood. If blood is requested, we must go to the hospital to have a blood draw, unless the phlebotomist is my is visiting the practice. And if they are, guess what? They’re booked up on advances, so off to the next town to have your blood taken.

Ours has in-house phlebotomists and they are far better at getting blood than the GP ever was.  The “system” requires you to come back another day to get that done (I believe if it is urgent the GP will stab you themselves), which seems like an unnecessary step.  It’s actually one of the times when it might make sense for your own GP to do a phone/video consultation – quickly determine blood is needed then get the blood draw booked in and have a face to face with the results.  BUT I still don’t understand how private GP “apps” deal with this (or other) remarkably common diagnostic steps?  The convenience of a GP app is lost if rather than go a mile to my GP I now need to drive 30 minutes to a hospital for blood to be taken.  I have the app, everyone in my company has the app, but I don’t know anyone who has ever used it for anything other than exactly the sort of thing TiRed mentioned – rubber stamping a diagnosis or referral you’ve already done yourself.

You know chat GPT doesn’t know how many r’s in strawberry and doesn’t “learn” when corrected.

ChatGPT is a generative AI based on a large language model connected to a chat interface. That’s just one thing that a small part of what comes under AI has been used for.

That’s like basing an opinion on food on your experience of fried chicken.

Sorry but Ai is mostly a branch of statistics, and statistics is a branch of Analytics. I’ve run my own random forests and SVM’s. And I’ve run plenty of dimensional reduction analyses (PCA’s), linear and non-linear mixed effects models too on big and small data. I don’t, however do IT. Our statisticians can’t understand how the AI guys managed to steal their ball. I concur that AI need not be about merging databases, but one can use fuzzy logic to inform on such merges rather than SQL Upper(Name) 😉