AI in healthcare. Here’s what’s happening now.
Update on developments in artificial intelligence in healthcare.
Hey there,
My uncle Reinder is 60 years old. His parents, my grandparents, quickly noticed something was off.
At three, he still wasn’t talking. Reinder had temper tantrums and would break things.
What was the diagnosis? Reinder has an intellectual disability, likely due to prolonged oxygen deprivation at birth.
At four, he moved out and has since lived in facilities for people with disabilities.
As I often lecture on AI in healthcare, I began wondering how care for Reinder looks today, what technologies are in use, and what matters most to him.
I answer that last question at the end of this newsletter. But first, here are two concrete examples of how AI now is being used at ZGT and UMCG hospitals (both in the Netherlands).
Happy reading,
Peter
P.S. Even if you don’t work in healthcare, this newsletter has valuable insights and practical tips that can be applied beyond healthcare.
P.P.S. If you prefer watching, I made a video on the latest AI trends in healthcare.
For English subtitles: select the icon of a sprocket (settings), and then subtitles.
AI in healthcare
Artificial intelligence in healthcare: What are the latest developments and best practices? Practical tips to advance AI in healthcare.
These are two standout projects:
Emergency Department at Ziekenhuisgroep Twente (ZGT) in Almelo.
Composing Emails at UMCG in Groningen.
1. Emergency Department
Annually, approximately 30,000 patients are treated in the emergency department of Ziekenhuisgroep Twente (ZGT) in Almelo—about 86 patients daily.
Around 40% of these patients are admitted to the hospital. Before admission, it’s essential for pharmacy assistants to conduct medication verification.
Starting in February 2024, pharmacy assistants will use an algorithm to identify patients with a high likelihood of admission within 10 minutes of their arrival, allowing these patients to be prioritized for medication verification.
I see this as a prime example of how AI can streamline healthcare processes.
I spoke with data scientist Job Maathuis, who developed the algorithm as his thesis project:
"The data used for predictions includes demographic information, arrival times, triage details, lab results, and vital signs like blood pressure and heart rate."
Success Factors
I asked Job what contributed to the project’s success:
Rapid data processing: updates predictions every 10 minutes.
Broad applicability: integrated with the hospital’s bed management system, enhancing visibility into bed availability later in the care process.
Alignment with an ongoing improvement project in the emergency department.
Engagement of various disciplines in development, including pharmacy, emergency department, bed management, and data teams.
A mix of technical factors (1 & 2) and social components (3 & 4) contributed to success.
2. Composing Emails
Several hospitals use large language models (LLM’s), similar to ChatGPT, to answer patients’ emails.
The UMCG in Groningen leads this innovation. I discussed it with Bart Scheerder, a business developer in data science and AI.
Here’s how it works:
Patients submit questions to their doctors via an online platform.
The tool analyzes the question, supplements it with patient-specific information from the Electronic Patient Record (EPR), excluding sensitive data.
The tool creates a detailed prompt based on the question and relevant EPR information.
A language model, based on ChatGPT, generates a draft response.
The doctor reviews and edits the response before sending it to the patient.
Best Practices
I asked Bart why this project is successful. Key factors include:
Privacy: Patient data isn’t stored outside UMCG, and the language model has no direct access to the EPR. Generated responses are deleted after use.
Time: The language model’s prep work significantly reduces the time doctors spend on patient questions, freeing time for other tasks.
Quality: Patients perceive responses generated by the language model as more empathetic and of higher quality compared to those fully written by doctors.
Training: Doctors can only use AI after completing e-learning.
Like with my conversation with Job at ZGT, I noticed that UMCG prioritizes technical prerequisites, patient and provider benefits, and attention to the social aspects of these changes.
To succeed with AI in healthcare, all these elements are necessary: technological foundations, clear benefits, and consideration of the human side of change.
Deep Dive on AI & Work
Articles, books, podcasts, videos, documentaries, and more on this theme.
1. READ / These previous newsletters I wrote about artificial intelligence in healthcare:
2. READ / I often emphasize in lectures that AI can also lead to new diagnostic methods.
For example, sound: this article details Google’s work analyzing sounds like sniffing, coughing, and breathing to assess the likelihood of conditions like tuberculosis.
3. WATCH / During a talk at the annual congress of the Federation of Medical Technology Companies, I discussed potential tech trends following AI. One possibility is nanotechnology.
In 1966, the film Fantastic Voyage depicted a creative exploration of this concept, with tiny doctors navigating the bloodstream in submarines searching for blockages.
🙏 Thank you for reading
This newsletter is free, but not cheap to make.
You can help me in a number of ways: forward this newsletter to someone who likes it, subscribe to my YouTube-channel, or hire me to speak.
Interested to hire me? Feel free to fill this form!