Can a Computerized Task Treat Depression?
Gal Rabinovich#, Reut Shani and Hadas Okon-Singer - Department of Psychology
Tomer Sidi - DSRC
Artificial Intelligence Machine Learning
Supervision Grant 2021
Depression is an extremely prevalent psychiatric disorder, yet it is still undertreated, so psychologists and researchers keep looking for additional treatment interventions. One promising intervention that made critical acclaim in the past 10 years is Cognitive Bias Modification (CBM). CBM is a computerized task designed to alter cognitive mechanisms associated with clinical symptoms or disorders. Cognitive Bias Modification for Interpretation (CBM-I) targets Interpretation Bias, the tendency of individuals with higher levels of depression to interpret ambiguous scenarios in a negative manner compared to healthy individuals. CBM-I trains participants to positively interpret ambiguous situations, such that over time they might acquire healthier interpretation mechanism. Nevertheless, mixed results doubt the efficacy of CBM, and meta-analyses suggested that significant effects may have been found due to outliers and publication bias. So, should we neglect CBM as an intervention tool and as a field of research? We believe not. Mixed results could indicate that some people benefit from such interventions while others not. In this study, we were able to identify individuals more susceptible to CBM-I by inspecting their intra-training performance trajectories using advanced data-driven methods.
45 participants, after screening bad performance subjects, went through a single session of CBM-I, and performed several questionnaires and tasks before and after the CBM-I interventions. Such measurements allowed us to assess participants’ levels of subclinical depression, prior depression diagnosis, changes in mood and in positive interpretations ratio. Using unsupervised machine learning of K-Means Clustering algorithm, we divided our sample to three different subgroups based on their trajectories in CBM-I task.
Before we jump into our results, this kind of challenge illuminates how certain fields of research may benefit from interdisciplinary cooperation. And as you finish reading this article, we might be able to convey that it is particularly true when using advanced analysis techniques in the research of cognitive interventions. We assembled a unique team and joined the forces of Psychology and Data Science: from the School of Psychological Sciences, Gal Rabinovich (B.A. student) and Reut Shani (PhD student), supervised by Prof. Hadas Okon-Singer, a neurocognitive psychologist with unique expertise developing sensitive cognitive measures and cognitive trainings; and Dr. Tomer Sidi from the DSRC.
Three clusters were identified, pointing at various performance trajectories in CBM-I: Consistent High Accuracy (HA, n=10, green), Improving Accuracy (IA, n=13, red), Consistent Low Accuracy (LA, n=22, blue). While intuitively we may be drawn to focus on individuals who performed better like HA or IA, it is the inaccurate group who intrigued us the most. Our findings suggest that individuals who consistently struggle favoring positive interpretations along CBM-I session were more likely to be diagnosed with depression in the past, and were more likely to benefit from CBM-I in terms of mood, such that they became happier, less sad, and less afraid following a CBM-I intervention.
Previous studies in our lab have focused on individual and training differences as possible causes for these mixed results in CBM. However, the influence of performance and learning success during the training itself has rarely been investigated so far. Our research highlights the advantage of advanced analysis techniques over classic statistical models in the analysis of cognitive interventions such as CBM-I, as commonly used statistical methods are limited in revealing this kind of results. Currently, further research is needed before we will be able to use CBM-I as a therapeutic tool, targeting individuals who gain more than others using such cognitive intervention. Yet, using machine learning for intra-training analysis may just be what we need to better understand the nature of CBM-I.
This project relies on a long-lasting successful collaboration between The Cognition-Emotion Interaction (CEI) Lab and the DSRC, and was presented in two conferences taken place in The University of Haifa (click here to view poster).