Machine learning models predicted cannabis use disorder transitions with 74% accuracy using demographics, wearables, and social factors

What the researchers found

For cannabis users, both elastic net and random forest models achieved AUC of about 0.74 (no significant difference). Demographic variables were the strongest predictors across both models. Social determinants of health, particularly income, contributed substantially. Wearable-derived metrics (activity and sleep data) provided incremental value in linear models but limited independent contribution in random forests.

Why it matters

Predicting who will progress from cannabis use to a use disorder could enable targeted prevention. This study shows that readily available demographic and social data already provide moderate predictive power, with wearable technology adding incremental value.

The numbers in context

Cannabis cohort AUC: EN = 0.740, RF = 0.741 (DeLong p = 0.764); stimulant cohort AUC: RF = 0.732, EN = 0.698 (DeLong p = 0.219); demographics strongest predictors; income most important SDoH variable

How the study worked

Data from the All of Us Research Program, a nationwide cohort integrating electronic health records, surveys, wearable data, and social determinants. Individuals with baseline cannabis use were followed for incident SUD diagnoses. Elastic net logistic regression and random forest models were trained and compared using AUC on independent test sets.

What this study cannot tell us

Moderate predictive accuracy limits clinical utility. All of Us cohort may not be fully representative. Electronic health record diagnoses may undercount SUD. Wearable data had limited contribution, possibly due to data quality or relevance.

How to read the evidence

Moderate: large diverse cohort with multimodal data and appropriate ML methodology, but moderate predictive accuracy and observational design.

When this study was published

2026 publication using the All of Us Research Program cohort.

The bigger picture

This represents a step toward precision prevention in substance use. While 74% accuracy is moderate, combining easily collected demographic data with emerging wearable technology could eventually enable proactive clinical intervention.

Questions still open

• Would longer follow-up periods improve prediction accuracy? Could genetic data or neuroimaging biomarkers significantly boost performance? How should moderate-accuracy predictions be used ethically in clinical settings?

Common questions

Read the original research

Comparing random forest and elastic net models to predict substance use disorder transitions in participants with cannabis and stimulant use: Evidence from the All of Us cohort.

Drug and alcohol dependence, 278, 113012

Citation

Zamora, Gabriel; Gunawan, Tommy; Zhao, Qingyu; Meruelo, Alejandro D. (2026). Comparing random forest and elastic net models to predict substance use disorder transitions in participants with cannabis and stimulant use: Evidence from the All of Us cohort.. Drug and alcohol dependence, 278, 113012. https://doi.org/10.1016/j.drugalcdep.2025.113012

Explore the wider topic

• Physical vs Psychological Cannabis Dependence: What Science Actually Shows
• The Gap Between Cannabis Perception and Science: Why What We Believe Outpaces What We Know
• Cannabis Use Disorder: Am I Addicted? Self-Assessment Guide
• Cross-Addiction After Quitting Weed: When One Habit Replaces Another
• Is Weed Addictive? What the Research Actually Shows
• Is Cannabis Addictive? What the DSM-5 and Brain Science Say
• Quitting Weed and Alcohol Together: What to Know About Dual Cessation
• Is Rehab Necessary for Weed? An Honest Assessment
• Signs You May Have Cannabis Use Disorder: An Honest Self-Assessment
• Weed Vape Pen Addiction: Why Vaping Makes It Harder to Quit