Create prompts for non medics to ask patients questions based on their initial digital consultation

ChatGPT
1

Hi,

I am a GP (family doctor) in the UK.

I would like to use chatGPT to assist with econsultation (electronic consultations with our patients).

Our patients put in ‘free text’ ‘healthcare requests’. I would like to use chatGPT to prompt non-doctors, non medics, reception staff to ask the patient questions based on their free text ‘healthcare requests’.

I have created 2 fictional examples to demonstrate:

Fungal Toenail Consultation

Create 6 questions to ask the patient based on their Econsultation. Make the first 3 questions answerable by yes or no and make the last three questions answerable by a short sentence. Design the questions to get more useful and medically detailed consultation information from the patient that will help a doctor consult with the patient: their e consultations was = I think I have a fungal toenail infection

Sure, here are 6 questions to ask the patient based on their e-consultation regarding a possible fungal toenail infection:

Have you noticed any changes in the appearance of your toenail(s)?

Have you experienced any discomfort or pain in the affected area?

Have you had a fungal toenail infection before?

When did you first notice the symptoms of the infection?

Have you tried any over-the-counter treatments for the infection?

Do you have any underlying medical conditions or take any medications that may affect the infection or its treatment?

Create 6 questions to ask the patient based on their Econsultation. Make the first 3 questions answerable by yes or no and make the last three questions answerable by a short sentence. Design the questions to get more useful and medically detailed consultation information from the patient that will help a doctor consult with the patient. Do not ask them questions relating to their previous medical history or their regular medications. Their e consultations was = I think I have a fungal toenail infection.

Here are six questions to ask the patient based on their e-consultation regarding a possible fungal toenail infection:

Have you noticed any foul odor coming from your affected toenail(s)?

Have you had any recent injury or trauma to the affected area?

Have you been wearing tight-fitting shoes or socks frequently?

How long have you had the infection, and have you noticed any changes in the toenail(s)?

Are you experiencing any itching, burning, or discomfort in the affected area?

Have you had any previous fungal infections, and if so, how were they treated?

Create 6 questions to ask the patient based on their Econsultation. Make the first 3 questions answerable by yes or no and make the last three questions answerable by a short sentence. Design the questions to get more useful and medically detailed consultation information from the patient that will help a doctor consult with the patient. Do not ask them questions relating to their previous medical history or their regular medications. Their e consultations was= I feel fed up and anxious.

Here are six questions to ask the patient based on their e-consultation regarding feeling fed up and anxious:

Have you been feeling anxious or depressed for an extended period of time?

Have you experienced any recent major life changes or stressful events?

Have you noticed any changes in your appetite or sleep patterns?

Are you experiencing any physical symptoms, such as fatigue or headaches, in addition to your emotional symptoms?

Have you tried any self-help strategies, such as exercise or meditation, to alleviate your symptoms?

Have you ever sought professional help for anxiety or depression, and if so, what type of treatment did you receive?

However the data governance rules in the UK are fairly strict. Is there someone that could help me regarding said governance or this project.

Dr Bennett.

Sent from Outlook for iOS

2
3

Hi Dr Bennett … “this (your problem question)” looks familiar to me :slight_smile: The answer is YES it can be done - even before chatGPT … and even then in real-time. Somewhere in my “box of tricks” I have the script it also was multi-lingual and VERY accurate - going into ALL the areas you mention. If you’re interested I could dig it out, OK would need a couple of days work BUT is actually something I think you (and may be other GP’s) might find very useful - an as far as governance rules in the UK, when I built it I did ask the NHS and they seemed OK about it …

Oh, I forgot to add/say it is also accessible via text and speach input.

If you have any “direct” questions you can message/email me via my openAI profile. Regards

4

FYI @paulbennett12

OpenAI has a direct statement in its content use policy, as follows:

(prohibited) Telling someone that they have or do not have a certain health condition, or providing instructions on how to cure or treat a health condition.

image
image960×2079 166 KB

5

Hi Ruby_coder.

Thanks. I probably didn’t explain properly. It isn’t to use in that way. It is to generate questions for non medics to ask. The doctor would have responsibility for checking both that the questions were reasonable and would be responsible for assessing the answers.

There would be no information imparted, no instructions or treatment. I’m afraid contrary to hype A.I. is a very long way away from managing undifferentiated illness. My job is safe!

The patients econsult (even if names/ages etc were removed) could reveal patient identity as there econsults can be so specific. Therefore the econsult needs to be kept confidential. Ideally on a local network only. Is this doable?

6

Hi, could you give me any more details on this?

7

Hi Paul, apologies, I’m on my mobile, so please excuse any typos.

Basically “eons” ago, well feels like it, I built a “simple to use” but not simple in programming “patient facing/orientated” pre consultation “screening” system

Concept was very like your “requirements/suggestions/thoughts” in delivering the doctor good info prior to consultation so time was saved etc. The ability for translation eg UK to Fr, UK to ES - and visa versa meant it could also be used as say “patient owned” so it would work on holidays etc.

OK that was pre GTP days, but it was still totally real time, now with GTP could obviously become “more chat orientated” but a lot of look ups from say Hopkins, Drug Bank, Harvard Medical and the NHS alike would actually still be valid without GTP.

Regards
Russell

8

Hi Paul, something I forgot, it was also designed so patients could/would access the system via simple QR code posters in practice waiting rooms etc.

9

I think this is the hardest part because GPT-3 isn’t compliant to these medical privacy firewalls. If you went with embeddings with canned answers back without GPT-3 in the loop, you could possibly make it compliant (just have offline data and use your own complaint servers).

BTW, you would have to spin your own embeddings behind the firewall, can’t use the OpenAI ones for the incoming queries.

But if it’s like searching WebMD, then I don’t know. I am not a lawyer!

10

Hello Dr Bennett, I totally understand your question as I am also a medical doctor from India with special interest in AI. I can develop the tool for you. In fact, I developed something similar recently as a assistant to doctors to do various tasks e.g. Differential Diagnosis, Normal Range, Investigations etc. Please take a look and let me know what you think about this. AI Doctor

11

My experience, in general and personally, “I really dislike chatbots” in most applications; so if I had a doctor who asked me to chat with a chatbot, I would simply find another doctor.

Furthermore it is easy to tell when people on the other end of the call and are relying on chatbots to help them provide service, which is almost as annoying - it’s like a robocall with a talking human who you cannot get out of their “chatbot” role, haha.

On the other hand, I live on the seacoast in Thailand where when you go to a gas station people still greet you, smile, wash your windows, and pump your gas and every grocery store has people at checkout, not robots and touchscreens.

The last time my wife and I were back the US, we laughed so much at the fact everything has become so impersonal and automated; and we could not wait to get back home to Thailand.

Anyway, having rambled on about that, @Russell , good luck with your chatbot.

If you don’t mind me saying so, I’m sticking to talking to lovely nurses and consulting with real humans who don’t use chatbots.

:slight_smile:

1 Like
12

@ruby_coder, thanks for your interesting thoughts, TBH I’m of the same opinion (health wise) regarding “chat bots”, though I do end up using them a lot in my work talking to real people in terms of customer support, however, my “scribing” was in really in answer to the thread and not “as such” a chat bot project of mine.

What I developed a few years ago wasn’t a “bot scenario” more an interactive user interface - like you I’m not in the UK, I’m in Greece so sometimes there are language barriers (my Greek is not that good, and certainly tourists may not have much of any Greek) which means that barrier may/could be a disadvantage in medical diagnosis/treatment. Hence the “system” was about the user could use the system to “describe” medical problems that could be/are automatically translated.

I get the impression this thread is “along the same lines” in terms of getting/establishing the right info but without the need for translation - though that could be useful anyhow.

Regardless thanks for taking the time to reply and contribute - have a good time over there, hope all is well and above all, stay safe & healthy

1 Like