Predicting Outcome in Exposure-Based Cognitive Therapy for Depression Using Exploratory Graph Analysis on Treatment Narratives

Abstract

Text mining is an approach to studying written or spoken material that is exploratory and data-driven and that seeks to uncover patterns or associations between text factors identified in the data. Text mining of clinical data, particularly written narratives obtained during the course of treatment, has the potential to greatly increase our understanding of how people respond to treatment and why. This method of text analysis might also allow for the coding of much larger datasets than more labor-intensive human coding systems. The current study used a novel analytic approach to study text data, Exploratory Graph Analysis (EGA; Golino & Demetriou, 2017; Golino & Epskamp, 2017; Golino et al., 2018). EGA was used to identify topics in clinical text from a large sample of narratives written over the course of treatment for depression. In a sample of 82 adults with depression receiving Exposure-Based Cognitive Therapy (EBCT) at two sites, we identified latent text factors in narratives written each week before treatment. Using the Pennebaker (1997) expressive writing paradigm, clients were asked to write for 10 minutes about “their deepest thoughts and feelings related to their depression.” Baseline depression symptoms and the EGA text factors were examined as predictors of treatment response, defined as a 50% or more reduction in depressive symptoms. Neither baseline depression nor EGA text factors alone were much better than chance at predicting outcome, but a combined model including both variables was highly accurate (88.10%) and much better than chance (kappa=.71) at predicting treatment response status. Although there is much work to do to improve the application of EGA to clinical trial data, these findings suggest promise for incorporating text mining to make full use of available data and to improve understanding and prediction of treatment.