November 2022
CAIM Young Research Award winner Charlotte Kern and her colleague Verena Schöning use different approaches such as data mining, machine learning, and modelling and simulation studies in pharmacometrics to predict the effects of medication on diseases and thus inform clinical decision making. As a team they work together with medical doctors at the University Clinic of General Internal Medicine, in the Department of Clinical Pharmacology and Toxicology at the Inselspital, Bern University Hospital. During the COVID-19 pandemic, their research directly contributed to improved patient care.
Verena, what is your research about?
In our research, we essentially model the way that drugs enter the human body and the effect they have on certain pathogens there. This so-called pharmacometric modeling tells clinicians about the dynamics of medication in the body to help make decisions. Prompted by the COVID-19 pandemic, Charlotte and I first created a model of how the virus reproduces within the human body, or more exactly how infected cells produce the virus and release it into the body. We also modeled the plasma concentration profiles of different drugs including most of the World Health Organization’s SOLIDARITY trial and substances discussed by the general public. Finally, we combined the models and could thus predict the effects of these drugs on virus exposure in our simulated patients. We could directly see which treatment was the best candidate to be used in therapy.
Based on machine learning models such as multilayer perceptron and random forests we subsequently developed a “COVID-19 severity assessment score” to triage SARS-CoV-2 positive patients. Via data mining and statistics, we were able to use all available patient data (lab results, demographics, etc.) to identify risk factors to predict whether someone will eventually require intensive care or is at risk of dying in hospital. This worked with > 80% accuracy.
Charlotte, what motivates you about your work?
With our research in pharmacometric modeling and simulation, we analyse pharmacokinetics data and build models, taking into account the variability between patients and also random error. We look at the time course of drug exposure in the body and the drug’s effects, and in the end, we aim to help guide clinicians to choose the right treatment and dose for their patients.
In the context of the pandemic, our efforts gained incredible momentum: we had to switch projects because of delays in clinical trials abroad and understand what was coming at us. There was so much contradictory information available everywhere, and we wanted to try to help clinicians get their bearings. In early 2020, we used viral load data from untreated patients in Singapore, fitted it to our viral kinetics model and used all preclinical and clinical data as it became available to constantly screen for potentially effective drugs and treatment schedules. It’s extremely motivating to work in real-time with real-life data and know that your work has an impact! I felt I was part of something much bigger than us, that we were all up against – something we did not quite grasp the impact of yet.
It’s extremely motivating to work in real-time with real-life data and know that your work has an impact!
What are the fields of development in your line of research?
Verena: Definitely, Artificial Intelligence! Personally, I am really interested in disease progression and how AI could work for disease modeling.
I read one particularly interesting study which brought me to this idea about strong adverse reactions to chemotherapy. By changing the schedule and dose you can still have the same efficacy but with less side effects. This kind of information is helpful and might have a positive impact on people’s lives!
Then there are drugs, like certain antibiotics, with very narrow efficacy: if you take too little, they don’t help and could foster resistance. And too much quickly becomes toxic. These drugs are very difficult to dose as so many factors play into the correct dosage. The same goes for cyclosporine, which is applied after an organ transplant. Even after weeks of treatment, less than a quarter of children get the correct dose! With machine learning we could inform clinician’s dosing decisions. And we can leverage the big data available to us to help people.
Do you consider diversity in research important in your field? Why?
Charlotte: In our small group we are actually very well represented! Felix is my PhD thesis advisor, I consider Verena as my mentor, and within my group there is great support and encouragement for my development. However, when I look outside of science and I tell someone that I am working in computer science, they straight out don’t believe me! The mentality is evolving slowly that women can do this too. After all, some of the first programmers were women! (Ada Lovelace, Grace Hopper, ...).
Verena: There is still this prejudice that women are not good at coding. And vice versa, some women are afraid to start for no good reason. I think we are overall moving in the right direction. I started coding when I was 33. Therefore, it’s not something you need to learn when you are very young. You can start later and still make very good progress. Also, I have a pretty good work-life balance. As a mother, it is helpful that I do not always need to be in the lab. You can get organized better around your family duties than with other jobs which might be less flexible. It’s a great choice doing tech!
