# Identifying risk profile for adolescent e-cigarette use: A sex-stratified machine learning analysis

## TL;DR

This study uses machine learning to find different risk factors for e-cigarette use among adolescent males and females.

## Contribution

The study identifies sex-specific risk profiles for adolescent e-cigarette use using machine learning.

## Key findings

- Cannabis use was the strongest predictor of e-cigarette use across both sexes.
- Mental health symptoms predicted e-cigarette use only among female adolescents.
- Machine learning models showed moderate performance in predicting e-cigarette use.

## Abstract

Recent studies show that young females now report higher e-cigarette use than males, reversing prior trends. While sex differences in use are documented, little is known about underlying risk profiles. This study applied a machine learning (ML) approach to identify and compare predictors of adolescent e-cigarette use by sex.

We analyzed cross-sectional data from 1829 9th graders in Southern California (M=14.6 years; 54.7% female) surveyed in 2024. Gradient Boosting Machine, an ML algorithm well-suited for binary classification tasks, was employed to develop past 30-day e-cigarette use prediction models by sex. We additionally fitted a model that combined both females and males to assess overall risk factors. Sixty-eight self-reported variables across conceptual domains were included, and the top 10 predictors per model were identified using scaled importance scores.

Overall, 3.6% (n = 66; 3.7% females, 3.5% males) reported past 30-day e-cigarette use. In the female model, depression and post-traumatic stress disorders emerged as leading predictors, but not for males. Top risk factors in the male model included beliefs about and susceptibility to e-cigarette and cannabis use. In the combined model, the strongest predictors were primarily cannabis use and peer e-cigarette use. Model performance was moderate, with area under the receiver operating characteristic curve values of 0.86–0.88 and area under the precision-recall curve values of 0.19–0.54.

The findings of this study underscore the importance of considering sex differences when identifying risk profiles associated with e-cigarette use and developing targeted prevention and intervention programs for adolescents.

•Machine learning identified sex-specific risk profiles for youth e-cigarette use.•Cannabis use was the strongest predictor of e-cigarette use across sexes.•Mental health symptoms predicted e-cigarette use only among female adolescents.•Prevention integrating mental health and e-cigarette use is needed for female youth.

Machine learning identified sex-specific risk profiles for youth e-cigarette use.

Cannabis use was the strongest predictor of e-cigarette use across sexes.

Mental health symptoms predicted e-cigarette use only among female adolescents.

Prevention integrating mental health and e-cigarette use is needed for female youth.