“Depression is like the economy”
Depression can behave in the same way as the economy, according to doctor and researcher Suzanne van Bronswijk. An approach based on econometric modelling can therefore help in deciding between treatment options.
Anyone who has suffered from depression will be familiar with the sometimes laborious quest for an effective treatment. “It’s hard to predict which treatment will work best for a patient,” Van Bronswijk says. “During our training as psychiatrists or psychologists we learn which treatment we can use for which type of patient. But the considerations at the level of the individual patient are actually much more complex. Treatments that have proven effective often work well for a third of patients, to some extent for another third and not at all for the remaining third.” This means a treatment may have to be discontinued after a few weeks and a new one started. The first statistical models to predict treatment outcomes were therefore developed as early as the 1950s. But these models remain stuck in the research phase, with insufficient evidence on whether they predict the treatment outcomes only for the group under investigation or for other patients as well.
“A complex combination of factors determines whether a person is likely to become depressed, remain depressed or respond to treatment,” Van Bronswijk says. When she read the book Dit kan niet waar zijn by journalist Joris Luyendijk, a light bulb went on. “The book is about the banking crisis of 2008. Luyendijk explains how financial products are becoming increasingly complex. They were invented by ‘quants’—quantitative analysts—who work with complex mathematical models and use them to weigh up the risks of investments. It struck me that these models could be relevant for predicting the response (risks) to treatments for depression (investments).”
Suzanne van Bronswijk, is a psychiatrist at the MUMC+ and assistant professor of Clinical Psychology at the Faculty of Psychology and Neuroscience, Maastricht University. She studied medicine at UM and trained as a psychiatrist at the MUMC+ and various mental health institutions in Limburg. In 2019 she obtained her PhD cum laude at UM for her dissertation entitled ‘Personalized treatment strategies for depression.’
She reached out to the Department of Quantitative Economics at Maastricht University with the idea of joining forces. It turned out to be a good move. “They have more knowledge of building these kinds of complex models, we have our clinical expertise in treating depression. Together we were able to set up a great interdisciplinary partnership.”
The collaboration with the Quantitative Economics department, in particular with econometrician Nalan Bastürk, led to the development of a project that started last January and will run for seven years. The aim is to be able to make predictions, for the first time, for a diverse group of patients with depression. To this end, the researchers will set up a model based on Bayesian statistics. “The model will draw on data from a large number of previous studies and input from experts in the field, patients and family members.”
The story behind the prediction
Van Bronswijk and her colleagues aim to make a handy computer program for use in the consultation room. During an intake interview with the patient, the psychologist or psychiatrist can enter additional information that the model uses to make its calculations. Practitioner and patient then receive a clear answer that can help them choose a treatment. “Soon the program will also be able to tell the story behind its recommendation,” Van Bronswijk says. “So it will say, ‘This treatment is probably best due to the severity of the depression, certain past characteristics, and the current home situation. But we’ll keep measuring and predicting, as other factors may increase in importance.’ Because something can happen suddenly in the patient’s life that completely changes the prediction. If you don’t have a model, it takes time to realise that.”
The model will draw on data from a large number of previous studies and input from experts in the field, patients and family members.
The most appropriate treatment depends on not only the initial situation and the occurrence of sudden events, but also the patient’s goals. “Typically they want to reduce the symptoms of depression, but they may have other goals too—returning to work, finding meaning in their lives.” The model will therefore predict different treatments for different goals, such as quality of life. “It may be that one treatment scores slightly better on reducing the severity of the depression, while another scores slightly better on quality of life. The therapist and patient can decide together what to give more weight to.” Van Bronswijk hopes that the computer program will soon support the joint decision-making process of therapists, patients and their loved ones. During the project, she will test it at four different mental health institutions.
Some resistance to the implementation of the program is to be expected. Psychiatrists and psychologists may fear that the computer will usurp their role and make decisions for them. Van Bronswijk would like to offer reassurance. “What we’re trying to do is build a bridge between researchers and clinicians. It’s extremely difficult to predict the outcome of a treatment yourself, no matter how much expertise you have as a psychologist or psychiatrist. A model like this can be a useful tool. We’re not saying the computer can do everything better—in fact, we want to involve practitioners in the process and draw on their expertise to build the model.”
It’s extremely difficult to predict the outcome of a treatment yourself, no matter how much expertise you have as a psychologist or psychiatrist. A model like this can be a useful tool.