We might think the health sector is one of the most innovative, since it is a scientific field that, we would assume, evolves hand in hand with the technology sector. This is partially true, but we must also remember that health professionals are often burdened with bureaucratic tasks that can hold them back from trying to innovate in their day-to-day work.
In this tight space between welcoming innovation and being suffocated by the routine, time-consuming, and most mechanical parts of healthcare work is where physiotherapist Delfina Arnoldi decided to try and develop her own modest improvement for professionals.
While finishing her Master’s Degree in Artificial Intelligence Applications in the Health Sector, Arnoldi, who holds dual Argentinian and Spanish nationality, came up with the idea of creating a personalized AI bot as her Master’s Final Thesis. Her hope was that this work could later be applied in the real world, not only integrated into her own daily life, but also into her colleagues’. At the time, she was working in a private hospital in Madrid, within the subacute rehabilitation unit.
In an interview with Enterprisesandmore.com, Arnoldi explains in detail why she wanted to invest in this project and how it could improve healthcare workers’ efficiency to free up more time for the doctor–patient relationship. Specifically, the customized GPT’s goal was “to optimize and update report writing within the service.”
Were you able to publish this personalized GPT somewhere?
No, it’s not possible. The thing is, when you apply artificial intelligence within hospitals, for example, I used the hospital’s own reports, because I customized it for the writing style used in that specific hospital, to be able to do it in a more efficient way, and so that the reports that were drafted were practically tailor-made for that service. The difference compared with a tool that could be used on a larger scale is that you lose the ability to have that specific language used in that service.
For example, in this case, they used certain scales, which, among all the physiotherapy scales, were the ones they preferred. It was also customized based on the needs and characteristics of this service. What a personalized GPT allows you to do is adapt it to that service.
What I actually did was implement these new technologies, or tools that are already available in the market, because developing an AI solution [from scratch] takes years and a lot of money. The idea is to be able to integrate tools that already exist in the market and customize them, so you can move forward and integrate them much faster.
Are you currently using this tool?
Yes, I’m using it today. In fact, recently a new tool (not mine) came out within a nursing service that also used another GPT for other healthcare professionals, so that each of them can build their own personalized GPT. These kinds of tools are becoming much more common now.
Are you using OpenAI?
I decided to use OpenAI because of the ease it gives professionals in terms of the agility they need to integrate tools. That’s a problem we face today. Many tools are created, but they can’t be implemented because they are too slow for healthcare professionals, for example.
OpenAI, in this case, I chose to use it because it allowed me to use audio, identify language very quickly, structure previous formats into JSON, for example, and integrate them very fast, it generated the Word document directly, so physiotherapists could just download the file. It was really the tool in the market that best fit the service’s needs at that time.
Have you considered turning this into a business one day?
No, for now my idea is more about having a role as a technical coordinator, supporting certain professionals or services that want to integrate AI advances into their professions or services. Serving as a guide or adviser.
Creating a company is very complex in this sector, it’s a huge world and requires a lot of money. My idea is to connect users with solutions.
What motivated you the most to work on this idea?
I’m a fanatic for efficiency. I like optimizing work, and I think technology helps us do things in a simpler way. Above all, because for me bureaucracy was the heaviest part of my profession.
In fact, healthcare professionals spend almost 40% of our time on bureaucratic activities. Personally, I found reports very burdensome. For each patient that comes in, we have to do the initial report, the one-month report, the two-month report… Then, at least four or five scales, load that information, pour it into content, draft a report with those scales, and set goals for each patient.
It’s exhausting. What you really want is to be with the patient, giving them attention and working to help them recover as soon as possible, putting your energy there. You also have to be very careful with data anonymization, because to use GPT in healthcare requires very specific adjustments so as not to violate any data protection laws. Once that is done, physiotherapy services really could operate much more efficiently. And professionals could focus on what they know how to do, physiotherapy, being hands-on with the patient and working on their recovery.
This large amount of bureaucracy you mention, do you think it’s something particular to Spain, or is it general?
No, no, it’s worldwide. It’s been studied. We even saw it during the master’s program. Forty percent of healthcare professionals dedicate that much time to bureaucracy. The health sector involves a lot of data recording. That includes medical histories, follow-up notes, hospital bureaucracy, all of that is required work.
What’s something you learned during the development?
For example, what I found was, first, that the language each professional and each service uses has to be respected. Every professional has their own methodology and way of structuring assessments and work. That’s both personal to the professional and tied to the institution. It’s something that has to be preserved.
That’s why I think personalization is what makes the difference. Because if you’ve been working for 20 years and are used to speaking in a certain language, and you bring me a tool that speaks a different language, it’s going to be very hard for me to implement. Respect for the professional and the institution, and the language they use in their daily work, the tools must adapt to the service, not the service to the tool.
In the health sector, do you think there is speed in implementing new technologies?
I think it’s mixed. Some people like to implement technology very quickly. In fact, in our profession robotics is used a lot in recovery from injuries, for example, robots for gait assistance or balance recovery. We use a lot of technology.
But it’s true that some people resist its introduction because they see it as a threat to their job. I think it’s complementary, not a replacement. Work is enriched when technology is incorporated. It broadens the possibilities.
Just the other day I attended a meeting with top voices in AI in healthcare, and they were talking about the problems they are facing to integrate AI solutions in the sector.
There were many brilliant ideas, but when it came to implementation in the field, they were very hard to apply. Sometimes it’s because the solution isn’t fully ready, maybe they come to market too green and test professionals too much, and then get discarded because they’re not ready yet. And since this looks like a race to see who comes out faster, solutions don’t end up being fully prepared. Then, when you’re in direct work, with all the workload you have, if the solution isn’t agile, you’re going to discard it quickly. That’s one of the problems.
Sometimes it’s due to professionals’ resistance, and sometimes it’s also the industry, which releases products that aren’t fully ready. Another point they raised was that often clinicians are not involved in product design. So the products may be brilliant in theory, but in practice they lack field clinical knowledge.
What has been the biggest obstacle you encountered in developing this tool?
The work of preparing the data, structuring previous reports, processing the data to generate that flow into a final report that simulated clinical language, and getting a GPT almost similar to a clinician’s work in drafting, that took time. That was the hardest part.
After that, I was lucky because the team was actually very open. That hospital does a lot of work with robotics and is a very innovation-friendly team. So I didn’t face resistance from professionals.
Form created by the personalized GPT. Picture by Delfina Arnoldi.
And how did you solve that challenge?
Well, many hours of work and adjusting workflows. It had to be as efficient as possible, so that when it generated the report it was practically ready to print. And that the GPT wouldn’t invent things, putting in many safeguards to limit its creativity, so it stuck to what it had to do, because it hallucinates and takes on roles it shouldn’t. It had to have a very constrained structure, with scales passed in completely, no steps skipped, it had to just do its job and do it well.
What impact would you like to leave as a professional in the health sector?
What I’d like is for innovation to be integrated with consistency and practicality. Because often, as I said before, solutions are brilliant in theory but not practical in application. The biggest challenge we have is to find that balance between innovation and practicality. That’s it.
Looking ahead 5 or 10 years, do you think there’s another technology apart from AI that will impact the sector?
I think right now we have a lot to do with AI, but I’m also curious to see what happens with quantum computing, which is coming very fast. We’re living through a major revolution, and everything being developed behind the scenes that we don’t see yet, right now, many people are already using AI to revolutionize their work, and we’ll see the results in time.
People are starting to lose their fear of it and are more willing to integrate it. In a few years we’ll see the results of AI integration, and not too far down the line, those of quantum computing as well.
Arnoldi called her work “Pilot Case: AI-based Clinical Assistant for Physiotherapy in Subacute Neurorehabilitation”, and explained its functioning in the following way in a LinkedIn post:
“We began with Botpress to structure the conversational flow and then migrated to OpenAI technology, which provided greater agility and adaptability for daily clinical use. We also developed an internal manual for responsible use, with clear guidelines to ensure safe and ethical integration in the hospital environment. The assistant guides the collection of: qualitative assessment, standardized quantitative assessments (FAC, Tinetti, Ashworth, TIS, among others), and individualized physiotherapy goals.”
“With this data, the system automatically drafts a clinical report adapted to the structure and language the team usually employs in the hospital, significantly reducing the administrative burden and focusing time on direct care,” she detailed.
