Using 151 measures across seven domains, machine learning predicted youth cannabis initiation from individual, family, and neighborhood factors — but while 13 factors predicted first use, only 4 predicted escalation to weekly use, including cannabis outlet density and alcohol access at home.
Prevention program designers, urban planners, parents, and researchers studying substance use trajectories in youth.
What the researchers found
Random Survival Forest predicted first cannabis onset (C-index 0.68) and weekly onset (C-index 0.69). First use was predicted by 13 factors across 6 domains (biobehavior, cognition, family, peer, neighborhood, legal). Weekly use required only 4 factors, with 3 shared with first use: cannabis outlet density, access to alcohol at home, and positive social expectations about alcohol.
Why it matters
The distinction between first use (many factors) and escalation to weekly use (few factors) is crucial for prevention. First use may be hard to prevent because it's influenced by many small factors, but escalation to habitual use may be preventable by targeting just a few key environmental factors.
The numbers in context
450 cannabis-naïve youth. 8-year follow-up. 292 initiated use, 163 reached weekly use. First onset: C-index 0.68, 13 predictive factors across 6 domains. Weekly onset: C-index 0.69, 4 predictive factors. Key factor for first use: lower positive thinking during stress coping. Shared escalation factors: cannabis outlet density, home alcohol access, positive alcohol expectations.
How the study worked
Data-driven analysis of 151 measurements spanning 7 domains from the NCANDA cohort (450 cannabis-naïve youth, ages 12–21, followed 8 years). Random Survival Forest modeled age of first onset (292 transitioned) and weekly use onset (163 transitioned). Factors from individual, microsystem, and exosystem levels assessed.
What this study cannot tell us
NCANDA cohort may not represent all U.S. youth. C-index of ~0.68 indicates moderate prediction. 8-year follow-up may not capture all initiators. Some domains (brain MRI) had limited predictive contribution. Legal environment may have changed since data collection.
How to read the evidence
Longitudinal national cohort with data-driven machine learning across 151 measures and 7 domains, providing robust multi-level prediction of cannabis trajectories.
When this study was published
Published 2025, NCANDA cohort data.
The bigger picture
The finding that cannabis outlet density and home alcohol access drive escalation to weekly use suggests environmental interventions — like zoning regulations and parental alcohol management — may be more effective at preventing habitual cannabis use than individual-level interventions.
Questions still open
- Would reducing cannabis outlet density near schools delay escalation to weekly use? Can these few escalation factors be targeted in precision prevention programs? Do the same factors predict escalation in legalization vs. prohibition states?
Common questions
Can you prevent teens from trying cannabis?
Why does alcohol access at home predict cannabis escalation?
Read the original research
Multi-level patterns predict cannabis use onset among youth.
Developmental cognitive neuroscience, 76, 101639
Citation
Wang, Yixin; Fraser, Robbie; Aguinaldo, Laika; Nguyen-Louie, Tam T; Baker, Fiona C; Tapert, Susan F; Pohl, Kilian M. (2025). Multi-level patterns predict cannabis use onset among youth.. Developmental cognitive neuroscience, 76, 101639. https://doi.org/10.1016/j.dcn.2025.101639
Explore the wider topic
- How to Talk to Your Teenager About Weed (Without Losing Them)
- My Kid's Friends Smoke Weed: A Parent's Guide
- I Smoke Weed But Don't Want My Kids To: Navigating the Hypocrisy
- Quitting Weed as a Parent: Why It Is the Hardest and Most Important Quit
- Quitting Weed as a Teen or Young Adult
- My Kid Is Smoking Weed: A Parent's Guide