After graduating in 2009, Jannick Dorresteijn ended up “more or less by accident” at a symposium on prediction, on the occasion of Harvard professor Paul Ridker being awarded an honorary doctorate at Utrecht University. “The symposium was organised by Frank Visseren and Yolanda van der Graaf. Following on from the symposium, we started thinking about the many large studies carried out in the medical world, so-called randomised control trials, in which thousands of people are monitored for years. Those studies naturally provide a wealth of information. Traditionally, all that data is analysed and condensed into a single number. At the time, we asked ourselves the question: can’t we do more with all this information? Can we use data more effectively to make statements about individuals?”
As a PhD student in vascular medicine, Dorresteijn was then asked to come to Boston. There, he worked with Harvard colleagues on studies such as the JUPITER trial (on the effect of cholesterol-lowering drugs) and the Women’s Health Study (aspirin versus placebo in healthy women). “We looked at whether we could do something sensible with that data for the individual patient, using an algorithm. The average effect of such a trial does not apply to anyone at all. Because the average patient doesn’t exist. There are people within the group who benefit more than average from treatment and those who benefit less. The challenge is to identify the group that benefits more than average. After all, as a doctor in the consultation room, you try to assess that for each patient as well." The tricky thing about cardiovascular prevention is just that you are trying to prevent something that could potentially happen far in the future, Dorresteijn explains. “You need an algorithm to estimate how high that probability is. That is very different from assessing a CT scan, for which algorithms are also used.” The use of algorithms in medicine was not new even before Dorresteijn et al started in Boston: "They were already in limited use, but were really only applicable to healthy middle-aged people. We have gradually found out that it is quite possible to use this type of algorithm for people who already have vascular diseases or diabetes. As well as for people over 70.”
"The challenge is to identify the group that benefits more than average from the treatment. As a doctor in the consultation room, you try to do the same for every patient."
The research questions that Visseren and Dorresteijn posed originated in their own consultation rooms at UMC Utrecht. Indeed, both are not only researchers but also vascular physicians. Increasingly, they felt the need to selectively apply preventive treatment options to the patients who would benefit most. They also felt the need to personally educate patients about the individual effect of medication to enable them to make these decisions about it together. They therefore focused their research on whether computational algorithms could be developed that were applicable in practice for large groups of patients. “The breakthrough came when we started combining existing methodologies to look at data from a different perspective. Previously, an algorithm only allowed you to see into the future for a limited time – usually 5 to 10 years – because that is the time for which large studies are set up. But instead of looking at how long someone was included in the study, we started looking at the age at which someone joined and left the study. Using age as a time scale enables you to look much further ahead, and calculate lifetime risks. In this way, we continued advancing in methodology and gained more and more collaborations with other researchers, which resulted in a huge network and access to a huge amount of data, and eventually in U-Prevent." The researchers were able to prove conclusively that it was theoretically possible to personalise cardiovascular disease risks, and the effects of treatments, based on existing data. After that, a great deal of additional knowledge was collected. By constantly investigating how to increase the reliability of predictions, further developments were made – “until we reached a point where we said: this is SO good, this has to go into clinical practice.”, Dorresteijn says.
"The breakthrough came when we started combining existing methodologies to look at data from a different perspective."
Within the European Society of Cardiology (ESC), experts in the field from across Europe have come together in the Cardiovascular Risk Collaboration (CRC) to jointly develop computational algorithms that can be recommended by European guidelines. “The updated version of the 2003 SCORE table was published last year. The 2003 table was outdated and not properly calibrated for use in Eastern Europe. The new SCORE2, however, can be used across Europe, and predicts not only the risk of death, but also of having a heart attack or stroke. In addition, a similar score was developed for elderly people aged 70-89 years, making it possible to now reliably estimate cardiovascular risk for that group, as well. And for patients who already have vascular disease, the CRC developed the SMART2 risk score. This algorithm was derived from the Utrecht SMART cohort study and is therefore now used across Europe. This is important because, particularly for patients with vascular diseases, there are many more options today than there used to be, in terms of treatment. But who will benefit from what? The SMART2 algorithm calculates exactly that. The ESC guideline that recommends U-Prevent and the algorithms we developed is now the prevailing line, and I’m quite proud of that." Just as for the large number of U-Prevent users: “Processing algorithms into a calculator has been a big step. U-Prevent receives many compliments, and we are still working to continuously improve the tool. Recently, we figured out what to do if you don’t now something about a patient that you must enter in the model. One variable in the SMART2 risk score, for example, is the CRP value: it is not routinely measured. In a case such as this, if you fill in the value that is average for that group, you don’t hurt the estimate much. Research shows that it is still reliable. That too is another step forward.” A further step that has already been realised is that additional information about the patient that is not included by default in the U-Prevent model can still be added to the risk estimate calculated by the algorithm. “That, in turn, is going to bring a great deal of additional convenience and make the algorithm more applicable to larger groups of patients.”
"The ESC guideline that recommends U-Prevent and the algorithms we developed is now the prevailing line, and I’m quite proud of that."
Doctors naturally always approach their patients as individuals, and treatments are therefore already individualised, Dorresteijn wants to stress. “The novelty of our approach is that you can substantiate numerically what it is about the effectiveness of treatment and then have a conversation with each other,making it possible to further personalise the already personalised treatment.” The algorithm is certainly not intended to replace the Dutch CVRM guideline: "The goal of U-Prevent is to support healthcare providers in implementing the guideline. Nor is the guideline a rulebook: the patient always has the final say. The guideline describes what to present to the patient; what to discuss to arrive at the right decision for each person. In two similar cases, the best decision may be different for each.” He believes there are still great opportunities in GP care and the proper use of algorithms: "In a healthy population, how can we determine at an early stage who is at risk of cardiovascular disease and then provide preventive care to those people? A collaboration with researchers at Leiden University is now underway in this area to see if we can use U-Prevent for population health management. In other words, to screen GP patient files for high-risk patients. You can also detect these with blood pressures or sugar levels that the patients measure at home themselves, or by having cholesterol pricked once at the supermarket. If an algorithm picks that up and qualifies someone as ‘high-risk’, the GP himself no longer needs to actively screen for it. That takes a lot of work off a GP’s hands.”
"What’s new about our approach is that you can substantiate numerically what it is about the effectiveness of treatment and then have a conversation with each other."
Jannick Dorresteijn studied medicine (2009) and epidemiology (2011) at Utrecht University. His PhD research focused on personalising cardiovascular disease treatment based on computational models. This research was carried out partly at Harvard University and led to a cum laude PhD at Utrecht University in 2013. He undertook his training (2012-2019) at Diakonessenhuis Utrecht and UMC Utrecht. In 2016, he received a Dekker grant from the Dutch Heart Foundation [Nederlandse Hartstichting] to continue his research. In 2018, this research resulted in the go-live of http://U-Prevent.nl, a website that makes computational medication-on-demand models on preventive medication for cardiovascular disease available for use in the consultation room. From 2019-2020, he worked as an vascular physician at Rijnstate hospital in Arnhem and, since 2020, at UMC Utrecht.
Between the theoretical question of whether clinical data can be used to estimate cardiovascular risks for individual patients, and the online calculation tool U-Prevent as it stands, there is a lot of very high-level research. That research and numerous collaborative projects have led, among other things, to U-Prevent being included in the Cardiovascular Risk Management (CVRM) guideline of the European Society of Cardiology (ESC). Co-founder Jannick Dorresteijn outlines the genesis of U-Prevent.
Interview with Jannick Dorresteijn, physician at UMC Utrecht and co-founder U-Prevent