rethinkTHC Search
Menu
Study breakdown

Machine learning identified different predictors of sustained cannabis vaping depending on state legalization

Longitudinal CohortModerate evidence
The takeaway

Using PATH Study data, machine learning found that predictors of multi-year cannabis vaping differed by legalization status: cigarette use and bullying in legal states versus heroin use and nicotine vaping in non-legal states.

Substance use epidemiologists, prevention program designers, policy researchers

Predictors of multi-year cannabis vaping differed completely between legalized and non-legalized states

What the researchers found

In legalized states, CART split on cannabis use, cigarette use, bullying, and ethnicity; in non-legal states, split on cannabis use, heroin use, nicotine vaping, and hookah; predictors of sustained vaping differed from initiation predictors.

Why it matters

Understanding who continues vaping cannabis over years is more relevant for intervention than who starts, and different predictor profiles by legalization status suggest policy context shapes sustained use.

The numbers in context

PATH Waves 4-6 (December 2016-November 2021); five-terminal-node CART models for each legalization stratum; prior cannabis use was primary split in both.

How the study worked

Secondary analysis of PATH Study Waves 4-6 (2016-2021); two-stage ML (LASSO + CART) stratified by state recreational cannabis legalization; representative US young adult sample.

What this study cannot tell us

ML identifies associations not causes; legalization status changed during study; self-reported measures; PATH attrition may bias results.

How to read the evidence

Nationally representative longitudinal data with sophisticated ML, but observational design and changing policy landscape complicate interpretation.

When this study was published

Published 2025, data from 2016-2021

The bigger picture

As cannabis vaping becomes the predominant consumption method among young adults, identifying who progresses to sustained use can help target prevention.

Questions still open

  • Why do bullying and cigarette use predict sustained vaping only in legal states? Does the heroin-cannabis vaping connection reflect polysubstance risk? Would post-2021 data differ?

Common questions

What predicts long-term cannabis vaping?
It depends on location. In legal states, cigarette use, bullying, and ethnicity were key. In non-legal states, heroin use, nicotine vaping, and hookah use predicted sustained vaping.
Is starting cannabis vaping the same as continuing it?
No. Predictors of multi-year cannabis vaping differed from those previously identified for vaping initiation.

Read the original research

Identifying predictors of multi-year cannabis vaping in U.S. Young adults using machine learning.

Addictive behaviors, 160, 108167

Citation

Choe, Siyoung; Agley, Jon; Elam, Kit; Bidulescu, Aurelian; Seo, Dong-Chul. (2025). Identifying predictors of multi-year cannabis vaping in U.S. Young adults using machine learning.. Addictive behaviors, 160, 108167. https://doi.org/10.1016/j.addbeh.2024.108167

Explore the wider topic