CAIRS: A causal artificial intelligence recommendation system for digital mental health

Digital mental health tools have the prospect to enhance and expand access to care for those in need. Some tools provide interventional recommendations to individuals, typically using simple static rule-based systems (e.g., if-else statements) or by incorporating predictive artificial intelligence. However, interventional recommendations require a decision based on the comparison of future outcomes under different interventions, which requires causal considerations. Here we develop CAIRS, a causal artificial intelligence recommendation system that provides personalised interventional recommendations using an individual’s current presentation and the learned dynamics between domains to identify and rank intervention targets that have the greatest impact on future outcomes. Our approach was applied to longitudinal data of multiple mental health and related domains at two timepoints (1 week - 6 months from baseline) collected from a digital mental health tool. In our example, psychological distress was found to be the key influential domain that affected multiple domains (e.g., personal functioning, social connection), and thus was typically the preferred target in complex cases where multiple domains were unhealthy. Our approach is broadly applicable to recommendation contexts where causal considerations are important, and the framework could be incorporated within a live app to enhance digital mental health tools.