JMIR Publications

ABSTRACT

Background:

Digital mental health tools promise to enhance the reach and quality of care. Current tools often provide interventional recommendations to individuals, typically using generic rule-based systems or predictive artificial intelligence. However, interventional recommendations require causal considerations to compare future outcomes under different interventions.

Objective:

Here we develop CAIRS, a causal artificial intelligence recommendation system that uses an individual’s current presentation, their preferences, and the learned dynamics between domains to identify and rank interventions.

Methods:

We frame the recommendation problem within a Bayesian decision-theoretic framework, whereby a preference ordering of decisions can be estimated using the expected utility of outcomes under interventions. The causal processes are assumed to follow a structural causal model, where we learn the posterior distribution of causal structures from observational data using a Markov chain Monte Carlo method. Expected utilities under interventions are then estimated using a do-operation, which passes through the effects of changing a variable on the outcomes, while accounting for confounders. We apply our approach to rank domains relating to mental health and wellbeing as intervention targets for adults (N=619) that used the Innowell Fitness app between September 2021 to September 2023 and completed a questionnaire at two timepoints (1 week - 6 months from baseline).

Results:

Psychological distress had the most prominent effects on other domains with paths to personal functioning (ppath=86%), social support (ppath=92%), sleep (ppath=88%), and physical activity (ppath=86%). Optimal interventions weighted in accordance with the probability of the baseline presentation were personal functioning (popt=30%), psychological distress (popt=29%), social support (popt=18%), nutrition (popt=9.6%), substance use (popt=6.7%), sleep (popt=4.5%), and physical activity (popt=2.2%). Psychological distress was typically the optimal intervention target in complex cases where multiple domains were unhealthy at baseline, due to its wide-ranging effects on multiple domains.

Conclusions:

This work illustrates the incorporation of causality and decision-making principles for an automated system to personalise interventions for digital mental health tools.