Effects of AI4U training, a machine learning-based, adaptive ecological momentary intervention for personalized mental health promotion in youth: findings from a micro-randomized trial

Abstract

Background: Recent advances in digital mental health have paved the way for more personalized mobile health (mHealth) interventions by leveraging intensive longitudinal ecological momentary assessment (EMA) data. Ecological Momentary Interventions (EMIs) facilitate reach of individuals in a given moment and context, in their daily living environments, which can, in turn, lower barriers to public mental health services. The use of advanced Machine Learning (ML) algorithms, generative Recurrent Neural Networks in particular, may augment effects of EMIs on mental health outcomes by improving the process of tailoring EMI components to individual preferences, needs, and contexts. This study aimed to investigate the feasibility, safety, and effects of a novel ML-based, adaptive EMI – the AI4U training - for personalized mental health promotion in youth. Methods: A within-subject micro-randomized trial (MRT) was conducted to investigate effects of EMI components on time-lagged proximal mental health outcomes in daily life. Participants were repeatedly randomized to one of two conditions up to seven times per day over a 30-day period with equal probability (i.e., up to 210 decision points for each individual): the experimental condition, receiving an EMI component based on a personalized ML-algorithm, or the active control condition, receiving an EMI component randomly. First, the feasibility and safety of delivering the AI4U training for mental health promotion in youth were evaluated. Second, the effects of ML-based vs. random assignment of EMI components on time-lagged proximal outcomes (i.e., positive affect, momentary resilience, negative affect) at tn+1 were examined. Third, uncontrolled effects of the AI4U training on distal outcomes (i.e., psychological distress, resilience, emotion regulation) at post-intervention were investigated. The AI4U training consists of eight EMI components on positive imagery, breathing techniques, and positive refocusing as well as a dedicated web-based dashboard for real-time self-monitoring of mental health outcomes and behaviours.Results: A total of 83 individuals aged 14–25 were identified from the general population and psychological counselling. Of these, 58 met eligibility criteria and provided informed consent, with 57 (98.2%) completing the AI4U training and post-intervention assessment. There was evidence on feasibility of the AI4U training overall, including high ratings of satisfaction, acceptability, and usability. No serious adverse events were observed. Beneficial effects of ML-based vs. randomized assignment of EMI components were evident for momentary resilience when the experimental and active control condition were compared (marginal excursion effect, B = 0.147, 95% confidence interval (CI), 0.004 – 0.290, p-value: 0.044). However, there was no evidence of beneficial effects of ML-based intervention assignment on momentary positive affect (B = 0.012, 95% CI, -0.012-0.128, p-value: 0.839) and negative affect (B = 0.093, 95% CI, -0.020-0.206, p-value: 0.103), respectively. Uncontrolled effects of the AI4U training on distal outcomes indicated reduced psychological distress (Cohen’s d: -0.25) along with improved resilience (Cohen’s d = 0.32) and use of adaptive emotion regulation strategies (Cohen’s d = 0.30).Discussion: This study provides new evidence on feasibility, safety, and beneficial effects of a novel data-driven and ML-based, adaptive EMI for personalized mental health promotion in youth. A confirmatory trial is now warranted.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 6 citations exist upstream.