Novel algorithms can predict psychosis before it strikes
Society’s quest for ever more wealth, comfort and growth is driving citizens into a corner. Affective and non-affective psychoses have never been so widespread, to the point where they’ve become the most expensive brain-related disorder in Europe. In affective disorders, the pattern is often the same: a sedentary, sunlight-deficient, sleep-deprived lifestyle starting from a young age, combined with an increased use of drugs and growing emotional neglect. “The cost for society is overwhelming,” says Prof. Dr Nikolaos Koutsouleris from Ludwig-Maximilians University of Munich. “Younger populations are strongly affected, and the resulting disability over their lifetime – due to the frequently relapsing course of these disorders – causes very high direct and indirect costs in 50 % of cases.” In Europe alone, psychotic and affective disorders amount to a burden of EUR 207 billion every year. One might wonder how to rein in the rising tide of mental illness. According to Prof. Dr Koutsouleris, we’re looking at three main breaches in current countermeasures. The first is that, in most EU countries, preventive psychiatry is still in its infancy, with no suitable mental healthcare infrastructures in place. The second reason lies in how early recognition strategies are derived from group-level statistical analyses, making it very difficult to reliably identify individuals at risk. Finally, early intervention procedures (mainly psychotherapy) are also derived from group-level clinical trials which have not been tailored to produce treatment recommendations for individual patients. People at high risk are very difficult to recruit for these clinical trials. The PRONIA (Personalised Prognostic Tools for Early Psychosis Management) project was built around this need for more representative studies and personalisation tools. “In PRONIA, we aimed to address the second shortcoming, that is, the need for tools that allow for a more accurate and representative measurement of risk in the single patient. We also tried to operationalise poor outcomes more broadly by including the likes of functional impairment in our prediction target, as well as to include more objective data in our prognostic tools such as neuroimaging, neurocognitive data and genetic or proteomic information.” PRONIA’s prognosis tools are tailored to high-risk populations, where this risk has already been established by a clinician. They complement the ‘gut feeling’ that currently rules patient prognosis with a quantification of the actual risk. “In the future, this could lead to a stratified preventive approach and a more rational allocation of therapeutic resources. The main innovation resides in how we trained machine learning algorithms to predict outcomes at single-subject level, by feeding them with sequentially-acquired multi-modal prognostic data,” Prof. Dr Koutsouleris explains. “In a sense, this mimics prognostic chains in real-world clinical settings. We add computer-aided support to these workflows to enhance medical decision-making at critical junctures in the process.” Concretely, clinical experts will be able to use a tool provided with quantitative risk estimates – risk scores – across different domains, such as risk for disease transition or risk for functional impairment at six-month, one-year or two-year follow-up points. Such an approach could facilitate a more flexible, broader and more accurate quantification of risk in each patient, although it does not resolve infrastructural challenges. The PRONIA consortium is in the process of drafting a business plan for a company that will test the prototype telepsychiatric decision support system in real-world clinical environments across different EU countries. “Obviously, when moving from bench to bedside many challenges will have to be addressed, including certification, patient safety and ethical considerations,” Prof. Dr Koutsouleris concludes.
Keywords
PRONIA, psychosis, mental health, algorithm, prediction, diagnosis
