Predicting Treatment Outcome from Patient Texts:The Case of Internet-Based Cognitive Behavioural Therapy

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 575–580 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 575 Predicting Treatment Outcome from Patient Texts: The Case of Internet-Based Cognitive Behavioural Therapy Evangelia Gogoulou1 Magnus Boman2 Fehmi Ben Abdesslem1 Nils Isacsson3 Viktor Kaldo3,4 Magnus Sahlgren1 1 Research Institutes of Sweden (RISE) 2School of Electrical Engineering and Computer Science, The Royal Institute of Technology(KTH) & Department of Learning, Informatics, Management and Ethics, Karolinska Institutet 3Department of Clinical Neuroscience, Karolinska Institutet, & Stockholm Health Care Services, Region Stockholm 4Faculty of Health and Life Sciences, Linnaeus University {evangelia.gogoulou, fehmi.ben.abdesslem, magnus.sahlgren}@ri.se {mab@kth.se} {nils.isacsson, viktor.kaldo}@ki.se Abstract We investigate the feasibility of applying stan- dard text categorisation methods to patient text in order to predict treatment outcome in Internet-based cognitive behavioural therapy. The data set is unique in its detail and size for regular care for depression, social anxiety, and panic disorder. Our results indicate that there is a signal in the depression data, albeit a weak one. We also perform terminological and sen- timent analysis, which confirm those results. 1 Introduction The Internet Psychiatry Clinic in Stockholm of- fers Internet-based Cognitive Behavioral Therapy (ICBT) for depression, social anxiety, and panic dis- order with documented significant treatment effects (Titov, 2018). The treatment is delivered on a se- cure online platform that enables patients to submit answers to weekly questionnaires of primary symp- toms, depression, and suicidal ideation, as well as interactive worksheets. Because patients normally are not in contact with the psychologists away from keyboard during treatment, the infrastructure plat- form logs a large amount of data for each patient, yielding a unique data set. Because Sweden was the first country to introduce this kind of ICBT, the clinical experience is considerable (Hedman et al., 2012), and experimental research of the kind reported in this paper rests on well-established pro- cedures for sensemaking of machine learning at the clinic (Boman and Sanches, 2015). The overarch- ing problem is how to be able to offer accelerated care, via identification of those predicted not to succeed with treatment, or at least not with enough decrease in symptoms, and giving these people more attention, as early as possible. We here investigate the possibility of predicting outcome based only on patient texts, in various points in treatment. We formulate this as a uni- modal binary text categorisation problem, where the categories are simplified to either success or failure. In this initial feasibility study, we evaluate the performance of a number of text categorisation methods for predicting treatment outcome. Even if our methodology is mostly exploratory, we also relate terminological and sentiment analysis of pa- tients’ text to the treatment outcome. 2 Related work NLP has previously been used for predicting the outcome of psychiatric treatment in a number of encouraging studies. Stylometric analysis on pa- tients’ texts from the data set under study indicates variation through time in the sentiment sign (posi- tive or negative) in patient text messages (Boman et …

Content truncated to 500 words