Stefan Jovinge, a pioneer's journey in cardiac intensive care and AI
From leading institutions in the USA, Research Director Stefan Jovinge shares his insights on how experiences in AI and health data have shaped his vision for the future of Life Science in Sweden.
In this interview, we meet Stefan Jovinge, Research Director at Skåne University Hospital, who has made an impressive journey from the Karolinska Institute to leading institutions in the USA, including Stanford and Harvard. With a unique combination of clinical expertise and advanced skills in artificial intelligence (AI), he shares his insights on how his experiences in the USA have shaped his work in Sweden. He also highlights the challenges and opportunities Sweden faces within the life science industry and provides his perspective on how we can balance innovation with patient safety and privacy.
Can you tell us about your background in Sweden and the USA?
- I was born in Stockholm and moved to Helsingborg during my upbringing. I attended medical school at the Karolinska Institute, completed my doctoral studies, and specialized as a cardiac intensive care physician there as well. I spent two years as a postdoc at UCLA (University of California, Los Angeles). After my time in Los Angeles, I was recruited to Skåne. I co-founded the Stem Cell Center in Lund and became a professor at Lund University. Simultaneously, I was the head of the cardiac intensive care unit at Skåne University Hospital (SUS) in Lund. After working as the Medical Director at the Cardiac Intensive Care Unit in Lund for five years, I was recruited to Michigan State University and Spectrum Health, where I worked as a senior physician and cardiologist in the thoracic intensive care unit and served as Research Director for nearly 10 years. During this time, I also became a professor at the Van Andel Institute in Michigan, as well as at Stanford University and the University of Texas. During my time in the USA, I earned a degree in epidemiology from Harvard University, specializing in secondary analyses. Subsequently, I completed a three-month program in Artificial Intelligence at the Massachusetts Institute of Technology. With these qualifications, I worked on data lakes related to intensive care.
“You don't need to understand the details of AI to benefit from it, but you must understand the results.” - Stefan Jovinge
What are the biggest differences in working with health data in Sweden compared to the USA?
- In the USA, there is a strict yet consistent approach to data lakes. Individual data requires permission to be used. What has allowed AI to advance so far is the process of handling data differently when it is pseudonymized. This enables the creation of datasets where identifiable data is removed, and through keys, an so-called honest broker, who is not part of the projects using the data, ensures that the dataset can continuously be analyzed and updated. Creating these pseudonymized datasets requires permission granted after review and if the data meets the legal requirements for pseudonymization with sufficient security. This is referred to as non-human subject data. These datasets do not require ethical approvals or consent from the patients from whom the data originates. For example, a patient having a blood pressure of 125/75 mmHg requires the patient's permission to be used, but not 125/75 without a connection to the individual. This allows the use of complete datasets, which is currently not possible in Sweden.
How do your experiences from the USA influence your work in Sweden?
- I learned everything I know about large health datasets and advanced analyses of these in the USA. The Americans have solved the legal question, and this inspires us in Europe, where we now have the opportunity to locally adapt the EHDS and AI Act to create the same conditions that North America has, within the EU and Sweden.
You have extensive experience with AI and data lakes from your time in the USA. How do you view the development of these technologies within Swedish healthcare?
- Sweden has begun to understand the seriousness, and the EU is driving Sweden forward. Risks are starting to be identified. The development has finally led to the understanding that all data management involves risks, but we cannot stop all handling of large datasets because there is a risk—everything entails risks. For example, we do not shut down all traffic because 350 people die annually.
How can we balance the need for large datasets for AI development with the protection of patient privacy?
- Here we can learn from North America: they talk about alternative risk. When running, for example, a pilot project for the use of AI, one can calculate the benefits it brings if used on a large scale. For instance, how many breast cancer patients' lives are saved by the AI algorithm. By weighing this alternative risk against the privacy risk, a balance can be found. Our biggest privacy risks lie with our current electronic medical record systems. These data are often unencrypted and without pseudonymization. We happily use these systems today.
You are a member of an advisory group for life science to the Swedish government. What issues do you consider most urgent to strengthen Sweden's position in this field?
- We discuss many issues, but the use of health data is something I have driven. Our quality registers have given us an international reputation, and we still want to be an important player here. If our research and industry are to remain at the forefront, we must be able to make complete datasets available in pseudonymized form for analyses—otherwise, we will lose our position here. Life science is our third-largest export product, and it is extremely important from a national strategic perspective.
How can legislation be adapted to support innovation while protecting patient safety and privacy?
- We need to make complete pseudonymized datasets available that do not require patient consent. AI methodologies have enabled the discovery of more complex relationships than traditional statistics, but this also means that AI can more easily be misled by biased data. So, if you ask whether you can use patients' data, those who approve it will constitute biased data that can mislead AI.
How do you view Sweden's role in the global life science industry, and what steps are needed to promote innovation within the sector?
- Our actors (universities, companies, and academic hospitals) must be able to act as freely as possible, and it is crucial that they become attractive to international partners. It is through international collaborations that we can make a difference and be part of the international front line. Attempts to create "purely Swedish" solutions have been a Swedish theme that risks isolating us and causing us to fall behind.
If you could change one thing within the medical research field in Sweden, what would it be and why?
- Sweden must invest significantly more and not pour money into groups that can distribute it freely. All investments need to be reviewed by international expert panels to ensure that the funds have a chance to become beneficial.
Is there a particular project or research area you are especially enthusiastic about right now?
- Yes, three things.
I believe that the AI initiative with the help of AI Sweden enables very good collaborations and projects.
Projects around “AI on AI,” where AI is used to explain predictions to the user about what drives the prediction, are something we are interested in and I am fascinated by. You don't need to understand the details of AI to benefit from it, but you must understand the results.
The work on common standards for AI is beginning to take shape, and I believe it will be crucial for AI to be used as safely as possible. We hope for no setbacks due to misuse.
Note! The content on this blog reflects my personal opinions and does not represent my employer. As the publisher, I am not responsible for the comments section. Each commenter is responsible for their own posts.