Toward precision mindfulness: Predicting mindfulness outcomes after Mindfulness-Based Stress Reduction using interpretable machine learning.

BACKGROUND

Mindfulness-based interventions are increasingly used to target transdiagnostic processes implicated in emotional distress, yet little is known about whether post-treatment mindfulness outcomes can be predicted from clinically interpretable participant profiles. This secondary analysis examined whether machine-learning models could predict post-intervention mindfulness following a Mindfulness-Based Stress Reduction (MBSR) program.

METHODS

The sample comprised 378 adults who completed an 8-week, face-to-face MBSR program. Candidate predictors included sociodemographic, health, meditation-history, baseline psychological, and practice-related variables. Outcomes were post-intervention Five Facet Mindfulness Questionnaire total and facet scores. Four regression approaches (i.e., XGBoost, least-squares support vector machines, LASSO, and Elastic Net) were compared using 10 × 10-fold cross-validation. Cohort-level generalizability was further evaluated using nested leave-one-intake-out validation.

RESULTS

Post-intervention mindfulness was predicted with moderate accuracy. LASSO matched or outperformed more complex algorithms while yielding the most interpretable solution. A four-predictor model (i.e., baseline mindfulness, hedonic well-being, frequency of informal meditation practice during the program, and current meditation practice before the program) achieved stable performance for total mindfulness scores (cross-validated R2≈.50), with similar performance under leave-one-intake-out validation. Baseline mindfulness was the dominant predictor across total and facet-level models. Informal practice emerged as a recurring predictor, particularly for Observing, Describing, and Non-Reactivity.

CONCLUSIONS

These findings suggest that post-MBSR mindfulness outcomes are partly forecastable from a small set of psychologically meaningful and practice-related variables. Interpretable prognostic models may support stratified delivery of mindfulness-based interventions, although external validation and outcomes beyond self-reported mindfulness are needed before clinical implementation.