Conversational agents for mental health and well-being: a systematic assessment of scope, characteristics, behaviour change techniques and quality (Preprint)

BACKGROUND Mental disorders cause substantial health-related burden worldwide. Mobile health (mHealth) interventions are increasingly used for the promotion of mental health and well-being, as they have the potential to improve access to treatment and reduce associated costs. Behaviour change is an important feature of interventions aimed at improving mental health and well-being. There is a need to discern the active components that can promote behaviour change in such interventions and ultimately improve users’ mental health. OBJECTIVE We systematically identified mental health Conversational Agents (CAs) currently available in app stores, assessed the behaviour change approaches used, and described their main features, technical aspects, and quality in terms of engagement, functionality, aesthetics, and information using a validated scale. METHODS The search, selection, and assessment of apps were adapted from systematic review methodology and included a search, two rounds of selection, and evaluation following predefined criteria. We conducted a systematic search for apps from Apple’s App Store and Google Play using 42matters. Apps with CAs that were in English, uploaded or updated from January 2020, and provided interventions aimed at improving mental health and well-being, and assessment or management of mental disorders, were tested by at least two reviewers. The Behaviour Change Technique (BCT) Taxonomy v1, a comprehensive list of 93 BCTs, was used to identify the specific behaviour change components in apps. RESULTS We found 18 app-based mental health CAs. Most CAs had fewer than 1,000 user ratings on both app stores (n=12, 66.7%) and targeted several conditions such as stress, anxiety, and depression (n=13, 72.2%). All CAs addressed more than one mental disorder. Most CAs employed cognitive behaviour therapy. Half of the CAs identified were rule-based (i.e., only offered pre-determined answers) and the other half were AI-enhanced (i.e., included open-ended questions). CAs used 48 different BCTs and included an average of 15 BCTs (SD= 8.77; range= 4–30). The most common BCTs were 3.3 “Social support (emotional)”, 4.1 “Instructions for how to perform a behaviour”, 11.2 “Reduce negative emotions”, and 6.1 “Demonstration of the behaviour”. CONCLUSIONS Mental health CAs are mostly transdiagnostic and targeted various mental health issues such as stress, anxiety, and depression, reflective of a broad intervention focus. They also have a limited number of downloads or user ratings and employ a similar limited choice of BCTs. The most common BCTs identified serve to promote self-management of mental disorders with little therapeutic elements. Future research should assess the role of AI in promoting behaviour change within CAs. It should also aim to determine the choice of BCTs in evidence-based psychotherapies to enable systematic, consistent, and transparent development and evaluation of effective digital mental health interventions.