Development and pilot trial of a chatbot as a digital wellbeing intervention to reduce stress in tertiary students

Abstract

Background: There is an increasing interest and demand for digital mental health interventions, but concerns have been raised about poor adherence rates in web based and smartphone app tools. Conversational agents such as chatbots, which mimic brief verbal interactions, could be a more engaging mode of delivery leading to a greater efficacy. Objectives: The aim of this study was to develop and pilot a custom-made digital wellbeing intervention designed to deliver stress management techniques for young adults attending a tertiary institution. Methods: The first phase of the project consisted of content development for a chatbot to be delivered using the Facebook Messenger platform. This was derived from cognitive behavioural therapy (CBT) and included elements known to increase resilience such as mindfulness, and gratitude. The second phase consisted of an open trial single arm design pilot study to determine the feasibility, engagement and acceptability of the chatbot. This was assessed via usage data, collected passively and via satisfaction ratings and qualitative feedback. Effectiveness was assessed via outcome measures for wellbeing, stress and anxiety (World Health Organisation (Five) Wellbeing Index; Personal Wellbeing Measure (ONS4); Perceived Stress Scale – 10-item version (PSS-10); and Generalized Anxiety Disorder 7-item scale (GAD-7)) Results: 124 participants commenced the chatbot and 110 completed both baseline and post-intervention assessment. 81% were female, and 51% were first year undergraduate students. A majority were either NZ European or Asian (45% and 41% respectively). Mean engagement was 11 days out of the total 21-day program (SD = 7.8). Most found the chatbot easy to use (81%) and glitch-free (78%) but only a third felt it reflected their culture. 63% rated their satisfaction with the chatbot at 7 or more out of 10. Qualitative feedback revealed that convenience and content were the most valued features. Significant effect sizes were found for improvement in wellbeing (0.49, SE =1.9) as measured by the WHO-5, and reduction in stress (0.38, SE= 0.6) as measured by the PSS-10. There was a significant improvement in anxiety in the subgroup of participants who were clinically anxious at baseline (p=0.011), as measured by the GAD-7. Conclusions: Using a chatbot to deliver universal psychological support appears to be feasible, acceptable and led to a reduction in stress and improvement in subjective wellbeing. Future research should explore customising of the content in co-design with end-users, and more robust assessment of effectiveness.