Chatbot Intervention for Anxiety and Depressive Symptoms in Young Adults

This clinical trial compares the effects of using an artificial intelligence based therapy chatbot and a self-help book to lower anxiety and depressive symptoms. The therapy chatbot (named Fido) is an application delivering support for mental health via a conversation-like interface in a highly inflected language (Polish). The team will recruit young people (aged 18-35) who currently struggle with anxiety and/or depression. The participants will use the chatbot or self-help materials for two weeks. Then, the researchers will check if using the chatbot or the book improved several aspects of mental health and the quality of life. Moreover, they will compare the groups for any differences in outcomes.

The primary objective of this randomized controlled trial is to evaluate the effectiveness of a chatbot-delivered cognitive-behavioural intervention on anxiety and depressive symptoms (and their correlates, such as quality of life or feelings of loneliness) in a subclinical sample of about 80 participants. The intervention results are compared to an active control intervention with popular self-help materials. Secondary objectives include the assessment and analysis of the human-chatbot therapeutic bond and general user experience. The study uses a 2 x 3 mixed factorial design with two intervention arms (Fido chatbot vs self-help book) and three time points (T1: just prior to the intervention, T2: immediately after the intervention, T3: at a follow-up one month after the previous measurement). The primary intervention lasts two weeks. All primary and secondary outcomes are assessed using self-administered online questionnaires in Polish. Each primary outcome measure is analyzed using a 2 (arms) x 2 (time points) ANOVA models with T1-T2 comparisons to test for the intervention effect and T1-T3 comparisons to check the stability of the intervention effect. Additional analyzes utilize standard NHST calculations and procedures (such as t-tests and correlation coefficients) alongside the conventional ⍺=.05 level. Similarly to the previous research in this area, no blinding is used. Participants are informed about different conditions, but the research hypotheses are masked.