Did you know that anxiety and depression are not just about feeling sad or being nervous? In addition to emotions, they are shaped by patterns of biased cognition: ways that our brain perceives, interprets, remembers, and anticipates events. These hidden mechanisms shape the way we see the world and ourselves, often making everyday situations feel more threatening or hopeless than they genuinely are.

Our research focused on examining one of the greatest challenges in mental health: how to accurately diagnose anxiety and depression, which are known to have a vast overlap and heterogeneity of symptoms. While both conditions are usually classified as separate disorders, people with anxiety and depression commonly present shared symptoms, blurring the boundaries between them and forming a spectrum of severity. Our study, aimed to investigate whether underlying cognitive mechanisms, such as attentional biases, memory distortions, and interpretive biases, can offer a more accurate and objective means of assessing symptom severity. By moving beyond the labels of conditions, our goal is to pave the way for mechanism-based diagnosis and, ultimately, more effective interventions.

To explore this, we tested 225 participants with clinical and subclinical levels of symptoms on a comprehensive battery of computerised tasks that measure six types of cognitive biases: selective and spatial attention, expectancy, interpretation, memory, and cognitive control. Their performance was analysed using machine learning models, specifically Random Forest regressors, which allowed us to predict anxiety and depression symptom severity with high accuracy and to identify which cognitive reactions have the most significant impact on severity. Notably, interpretation and expectancy biases turned out to be the strongest predictors, which shows how much the way we expect the future or interpret daily events can shape the severity of symptoms.

This project is the result of a close collaboration between clinicians, data and computer scientists, and neuroscientists at the University of Haifa, led by

• Dr Thalia Richter – Clinical Psychologist, Postdoctoral Researcher, School of Psychological Sciences, University of Haifa (Currently: Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany). Developed and validated the cognitive test battery, led data collection, conceptualised the research questions,  provided clinical counselling, guided the methodology, and was the lead author of the manuscript.
• Prof. Hagit Hel-Or – Department of Computer Science, Head of the computational human Behavior Lab and a member of the Artificial Intelligence Research Center (AIRC), University of Haifa. Led the computational aspects of the project, supervised advanced data analysis and machine learning modelling.
• Prof. Hadas Okon-Singer – School of Psychological Sciences, the Integrated Brain and Behavior Research Center, and member of the Artificial Intelligence Research Center (AIRC), University of Haifa. Supervised the study design, writing, and analysis, provided resources, and guided the broader integration of psychological and cognitive neuroscience perspectives.

Our findings support a novel diagnostic framework: instead of relying mainly on self-reports of symptoms, clinicians may be able to incorporate cognitive performance measures to fine-tune the identification of disorder severity and type, by integrating AI tools for analysis. The study also contributes to the ongoing debate between disorder-specific and transdiagnostic approaches for investigating mental health, highlighting the advantages of the transdiagnostic approach and evidence for both unique anxiety-related biases and shared anxiety-depression mechanisms. This dimensional, mechanism-based approach may help increase precision in psychiatric evaluations and lead to more effective interventions.

The key product of this research is our peer-reviewed article, published in the Journal of Affective Disorders (5-year IF: 5.6, Q1 Psychiatry, Q1 Clinical Neurology):

📄 Richter, T., Stahi, S., Mirovsky, G., Hel-Or, H., & Okon-Singer, H. (2024). Disorder-specific versus transdiagnostic cognitive mechanisms in anxiety and depression: Machine-learning-based prediction of symptom severity. Journal of Affective Disorders, 354, 473–482.
 Read the article here