Machine Learning Detects Clinically Trivial Personalization in School Mindfulness for Teens

Machine learning models applied to data from the My Resilience in Adolescence trial detected only negligible differential effects of school-based mindfulness training on depression prevention - an effect size of d ≈ 0.07–0.08 - and identified a subgroup the authors called “statistically detectable but clinically trivial.” JAMA Psychiatry lists the secondary analysis as online ahead of print with DOI 10.1001/jamapsychiatry.2025.4638, published on February 18, 2026, and the trial is registered under ISRCTN86619085.

The original MYRIAD trial was a cluster randomized clinical trial run from October 2016 to July 2018 that enrolled students aged 11 to 13 years at baseline in broadly representative secondary schools across England, Scotland, Wales, and Northern Ireland, involving “over 8,000 students.” Interventions compared included SBMT - “teaching core mindfulness skills through psychoeducation, class discussion, and practices” - versus standard social-emotional learning described as “teaching as usual.”

For the JAMA Psychiatry secondary analysis, investigators set out to predict individualized adolescent response to SBMT with machine learning. “School-level nested cross-validation was used to train and evaluate machine learning models for predicting individualized benefit from SBMT,” and the authors report that “data analysis was performed from April 2023 to October 2025.” The analysis targeted depression prevention as the outcome domain and framed the work as an exercise in precision prevention within universal school programs.

Results were explicit about magnitude and meaning: the machine learning models revealed only “negligible differential effects (d ≈ 0.07–0.08)” and the identified subgroup, while detectable by the algorithms, did not translate into clinically meaningful benefit. The paper concludes, “This study found that machine learning identified a subgroup with statistically detectable but clinically trivial differential intervention response.”

The authors place the finding in broader practical terms: “These findings highlight the substantial challenges in achieving clinically useful personalization in universal school-based prevention programs.” That wording underscores limits to tailoring SBMT at scale for depression prevention in early adolescence, suggesting that current ML approaches applied to universal classroom delivery may not yet produce actionable individual-level recommendations.

Key methodological and reporting gaps remain in the supplied materials: the secondary analysis summary does not list author names, exact total sample counts by arm, the specific machine learning algorithms used, the predictor variables or the precise definition of the subgroup, nor p-values or confidence intervals. For readers tracking implementation and policy, those missing technical details will matter when deciding whether to alter school MH programming or invest in precision prevention tools.