Predicting cannabis use moderation among a sample of digital self-help subscribers: A machine learning study.

Area of Science:

• Digital health interventions
• Machine learning in behavioral science
• Cannabis use disorder research

Background:

• Many individuals seek to reduce cannabis use independently using self-help tools.
• Existing digital tools show small effect sizes for cannabis use reduction.
• Identifying predictors of success is crucial for optimizing cannabis use moderation strategies.

Purpose of the Study:

• To investigate predictors of successful cannabis use reduction using machine learning.
• To evaluate the accuracy of machine learning models in predicting success in a digital self-help context.

Main Methods:

• Analysis of data from a randomized controlled trial comparing digital interventions (ICan vs. educational modules).
• Inclusion of 253 participants aiming to reduce or quit cannabis use.
• Definition of success as ≥50% reduction in weekly cannabis use at 6-month follow-up.
• Application of machine learning models (Random Forest, Logistic Regression) with nested k-fold cross-validation.

Main Results:

• Machine learning models demonstrated low predictive accuracy (AUROC: 0.61 for Random Forest, 0.57 for Logistic Regression).
• Key predictors for success included not identifying as a cannabis user, not using tobacco, high depressive symptoms, high psychological distress, and high baseline cannabis use.
• Overall, the associations between predictors and success were not strong.

Conclusions:

• Machine learning models offer modest accuracy in predicting success for individuals attempting cannabis reduction via digital tools.
• Further research is needed to improve prediction models for cannabis use moderation.