Increasing patient engagement with the help of Artificial Intelligence and Mixed Methods

A year into my coding journey I came across a quote that went something like this -- 'We cannot improve what we do not measure'. The quote was referring to people putting in large amounts of works to make applications more 'efficient', but if you aren't collecting data on how well your application is performing -- you're really just wasting your time. This person posited that without collecting the necessary data, no one is going to believe you when you say 'I made the app faster', 'I made the user experience better', or 'Everyone likes that cat background better'.

I began to think about this quote in a different way, however. I would like for the applications that I am building to improve people's lives. More specifically, I would like them to improve patient outcomes, assist in patient care, make nurses' lives a little easier, and so much more. But how can I possibly know that they will do this without any sort of measurement?

Enter from stage left, our old friend from nursing school whom we never thought we'd see again: Research.

Searching through Research

I began to read nursing research (and other medical research) in my spare time inspiration, education, guidance.. really you can get quite a lot out of reading research in a field of interest to you. In doing so, I encountered something called a Question Prompt List. Question Prompt Lists, or QPLs, are a simple list of questions given to a patient before a meeting with a healthcare provider that is meant to assist the patient in asking questions that the might not otherwise ask.

Asking questions in healthcare is extremely important, but not everyone knows what to ask or feels as though they have the courage to ask a question they might have on their mind. QPLs increase the amount of questions people ask, which in turn increase patient engagement, understanding, satisfaction and because of all this: improved outcomes.

Okay, sure, that's wonderful. But if QPLs help so much, why had I never seen them the 5 years I spent at the bedside charting on the computer while a Doctor consented a patient for a procedure? When asked if they had any questions, time and time again, I heard them say, "I don't even know what to ask..".

The truth is if I had done this research earlier I could have created this QPL for my own unit. But could I do it for every procedure? Could I consider all the edge cases? Would I have time to do that alone? I don't think so. But my friend chatGPT might be able to.

The Application

One of my previous projects was an application that used an LLM, specifically Google's GEMINI AI, to scan medical paperwork for difficult words and convert them into simplified flashcards. Having already used an AI chatbot to process words on paper, it seems a logical next step create an application that leverages AI to process spoken words.

In this next application, I plan to create a simple application that does 3 simple things:

1. Takes a recording of an informed consent as an input.
2. Processes the words of said recording, identifying the particular procedure and additional context.
3. Produces and displays a list of potential questions as an output.

The application technically creates a Question Prompt List tailor-made for any procedure/surgery/medication that a patient may encounter. But will it be any good? Will it help in the same way that a true QPL, made through the input of number health care providors and patients familiar with the procedure in question?

Well, there's at least one way to find out.

The Experiment

The plan is to create an MVP of the application using React, Node, and the gpt-4o API. The application will run on an Ipad on wheels that can brought into patient's rooms before they are consented for a procedure or surgery.

The patient will be informed about the study and be asked to sign a consent form approved by the IRB. They will then complete a brief pre-consent assessment measuring the patient's anxiety, knowledge about the procedure, and additional information. The start button will be pushed on the application and the consenting health care provider will begin the informed consent. When the informed consent is complete, the application-- which is tentatively called iQPL-- will process the audio and produce a list of potential questions.

The patient can ask a question provided in the iQPL, ask one of their own, or ask no questions at all. After completion of the consent, the patient will then be asked to fill out a similar brief survey.

Additional surveys will be collected from patients being consented for procedures without the use of the iQPL application. These will serve as our control group.

In the end, we will be comparing the experimental group and the control group for differences in amount of questions asked, self reported knowledge about the procedure after consent, difference in pre-procedure anxiety before and after consent, among other items.

We could also consider a qualitative approach in which we do short interviews with patients to obtain more personalized and in depth answers on their experiences using iQPL during the consent process. The quick turnover from patient consent to procedure start time may make detailed interviews challenging.

Why go through all this trouble?

The primary reason is proving dynamically generated QPLs is not only possible, but beneficial. Hospitals are not going to take my word for it. I need data. What better way to collect data than through a beautifully conducted experimental mixed methods study?

My goal is to make patient's lives better. I've been at the bedside for long enough. I've seen the panicked look of hundreds of people as a physician leaves the bedside and they didn't quite get all of their questions answered. I want a future where every question gets answered, even the one's they didn't know they had.

This is an idea still in its early stages, but I am having a lot of fun with it so far.

Until next time,

Ian