Personalized stress optimization intervention to reduce adolescents' anxiety: A randomized controlled trial leveraging machine learning

Anxiety symptoms are among the most prevalent mental health disorders in adolescents, highlighting the need for scalable and accessible interventions. As anxiety often co-occurs with perceived stress during adolescence, stress interventions may offer a promising approach to reducing anxiety. Previous stress interventions have largely focused on the view that stress is harmful, aiming to manage and mitigate its negative effects. Stress optimization presents a novel intervention perspective, suggesting that stress can also lead to positive outcomes. However, it remains unclear whether stress optimization can effectively reduce anxiety symptoms in adolescents. We developed a single-session stress optimization intervention and investigated the conditions under which it was most effective. A large-scale randomized controlled trial was conducted (N = 1779, aged 12-18 years), with participants reporting their perceived stress, stress mindset, and anxiety over a two-month follow-up period. Machine learning is a promising approach for assessing personalized intervention effects. Conservative Bayesian causal forest analysis was employed to detect both treatment and heterogeneous intervention effects. The findings revealed that the intervention effectively reduced anxiety symptoms in the school context over a two-month follow-up (0.87 posterior probability). Furthermore, adolescents with higher anxiety and perceived stress at baseline experienced the most significant reductions in anxiety outcomes (standard deviations of -0.18 and -0.11 respectively). The single-session stress optimization intervention demonstrated potential for cost-effective scaling.