A Machine Learning Approach to Identify Predictors of Frequent Vaping and Vulnerable Californian Youth Subgroups

Key Risk Factors Associated With Electronic Nicotine Delivery Systems Use Among Adolescents

Importance

The prevalence of electronic nicotine delivery systems (ENDS) use among US youths has increased significantly during the past decade. Identifying key factors highly associated with ENDS use is essential in monitoring and preventing this harmful behavior among youths.

Objective

To identify the most important risk factors in wave 4.5 (ie, December 2017 to December 2018) of the Population Assessment of Tobacco and Health Study (PATH) data that are associated with ENDS use in wave 5 (ie, December 2018 to November 2019) among adolescents who were tobacco-naive at baseline.

Design, Setting, and Participants

This prognostic study examined data from waves 4.5 and 5 of the PATH youth data set using machine learning techniques. The PATH study is a nationally representative longitudinal cohort study of tobacco use and health in the United States among individuals aged 12 years and older. The data analysis was carried out between January and April 2023.

Main Outcomes and Measures

Wave 5 current ENDS use status of wave 4.5 adolescents who were tobacco-naive.

Results

The analyzed data set comprised 7943 individuals who were tobacco-naive in wave 4.5. Among this group, 332 participants (4.2%) indicated their present use of ENDS in wave 5, 5047 (63.5%) were aged 12 to 14 years, 4066 (51.2%) were male, and 2455 (30.9%) were Hispanic. The most important risk factors of ENDS use in wave 5 among adolescents who were tobacco-naive in wave 4.5 were the likelihood of using ENDS if offered by a best friend (mean SHAP value, 0.184), the number of best friends using e-cigarettes (mean SHAP value, 0.167), household tobacco usage (mean SHAP value, 0.161), curiosity about ENDS use (mean SHAP value, 0.088), future intention to use ENDS (mean SHAP value, 0.068), youth's total average weekly earnings (mean SHAP value, 0.060), and perceptions of tobacco product safety (mean SHAP value, 0.026).

Conclusions and Relevance

The findings of this study suggest that family and friends play an important role in ENDS use among adolescents. The top-ranking factors associated with ENDS use in this study are areas for further exploration, given the increasing prevalence of ENDS use among youths in recent years. Additionally, these findings highlight the important role of families and schools in shaping adolescents' tobacco-related knowledge, which can protect them from using ENDS.

Are the Relevant Risk Factors Being Adequately Captured in Empirical Studies of Smoking Initiation? A Machine Learning Analysis Based on the Population Assessment of Tobacco and Health Study

INTRODUCTION

Cigarette smoking continues to pose a threat to public health. Identifying individual risk factors for smoking initiation is essential to further mitigate this epidemic. To the best of our knowledge, no study today has used machine learning (ML) techniques to automatically uncover informative predictors of smoking onset among adults using the Population Assessment of Tobacco and Health (PATH) study.

AIMS AND METHODS

In this work, we employed random forest paired with Recursive Feature Elimination to identify relevant PATH variables that predict smoking initiation among adults who have never smoked at baseline between two consecutive PATH waves. We included all potentially informative baseline variables in wave 1 (wave 4) to predict past 30-day smoking status in wave 2 (wave 5). Using the first and most recent pairs of PATH waves was found sufficient to identify the key risk factors of smoking initiation and test their robustness over time. The eXtreme Gradient Boosting method was employed to test the quality of these selected variables.

RESULTS

As a result, classification models suggested about 60 informative PATH variables among many candidate variables in each baseline wave. With these selected predictors, the resulting models have a high discriminatory power with the area under the specificity-sensitivity curves of around 80%. We examined the chosen variables and discovered important features. Across the considered waves, two factors, (1) BMI, and (2) dental and oral health status, robustly appeared as important predictors of smoking initiation, besides other well-established predictors.

CONCLUSIONS

Our work demonstrates that ML methods are useful to predict smoking initiation with high accuracy, identifying novel smoking initiation predictors, and to enhance our understanding of tobacco use behaviors.

IMPLICATIONS

Understanding individual risk factors for smoking initiation is essential to prevent smoking initiation. With this methodology, a set of the most informative predictors of smoking onset in the PATH data were identified. Besides reconfirming well-known risk factors, the findings suggested additional predictors of smoking initiation that have been overlooked in previous work. More studies that focus on the newly discovered factors (BMI and dental and oral health status,) are needed to confirm their predictive power against the onset of smoking as well as determine the underlying mechanisms.

Table 2 Fallacy in Descriptive Epidemiology: Bringing Machine Learning to the Table

There is a lack of rigorous methodological development for descriptive epidemiology, where the goal is to describe and identify the most important associations with an outcome given a large set of potential predictors. This has often led to the Table 2 fallacy, where one presents the coefficient estimates for all covariates from a single multivariable regression model, which are often uninterpretable in a descriptive analysis. We argue that machine learning (ML) is a potential solution to this problem. We illustrate the power of ML with an example analysis identifying the most important predictors of alcohol abuse among sexual minority youth. The framework we propose for this analysis is as follows: (1) Identify a few ML methods for the analysis, (2) optimize the parameters using the whole data with a nested cross-validation approach, (3) rank the variables using variable importance scores, (4) present partial dependence plots (PDP) to illustrate the association between the important variables and the outcome, (5) and identify the strength of the interaction terms using the PDPs. We discuss the potential strengths and weaknesses of using ML methods for descriptive analysis and future directions for research. R codes to reproduce these analyses are provided, which we invite other researchers to use.

A Machine Learning Approach Reveals Distinct Predictors of Vaping Dependence for Adolescent Daily and Non-Daily Vapers in the COVID-19 Era

Since 2016, there has been a substantial rise in e-cigarette (vaping) dependence among young people. In this prospective cohort study, we aimed to identify the different predictors of vaping dependence over 3 months among adolescents who were baseline daily and non-daily vapers. We recruited ever-vaping Canadian residents aged 16-25 years on social media platforms and asked them to complete a baseline survey in November 2020. A validated vaping dependence score (0-23) summing up their responses to nine questions was calculated at the 3-month follow-up survey. Separate lasso regression models were developed to identify predictors of higher 3-month vaping dependence score among baseline daily and non-daily vapers. Of the 1172 participants, 643 (54.9%) were daily vapers with a mean age of 19.6 ± 2.6 years and 76.4% (n = 895) of them being female. The two models achieved adequate predictive performance. Place of last vape purchase, number of days a pod lasts, and the frequency of nicotine-containing vaping were the most important predictors for dependence among daily vapers, while race, sexual orientation and reporting treatment for heart disease were the most important predictors in non-daily vapers. These findings have implications for vaping control policies that target adolescents at different stages of vape use.

Correlates of wanting to seek help for mental health and substance use concerns by sexual and gender minority young adults during the COVID-19 pandemic: A machine learning analysis

The COVID-19 pandemic has worsened the mental health and substance use challenges among many people who are Two Spirit, lesbian, gay, bisexual, transgender, queer, questioning, and intersex (2SLGBTQI+). We aimed to identify the important correlates and their effects on the predicted likelihood of wanting to seek help among 2SLGBTQI+ young adults for mental health or substance use concerns during the pandemic. A cross-sectional survey was conducted in 2020-2021 among 2SLGBTQI+ young adults aged 16-29 living in two Canadian provinces (Ontario and Quebec). Among 1414 participants, 77% (n = 1089) wanted to seek help for their mental health or substance use concerns during the pandemic, out of these, 69.8% (n = 760) reported delay in accessing care. We built a random forest (RF) model to predict the status of wanting to seek help, which achieved moderately high performance with an area under the receiver operating characteristic curve (AUC) of 0.85. The top 10 correlates of wanting to seek help were worsening mental health, age, stigma and discrimination, and adverse childhood experiences. The interactions of adequate housing with certain sexual orientations, gender identities and mental health challenges were found to increase the likelihood of wanting to seek help. We built another RF model for predicting risk of delay in accessing care among participants who wanted to seek help (n = 1089). The model identified a similar set of top 10 correlates of delay in accessing care but lacked adequate performance (AUC 0.61). These findings can direct future research and targeted prevention measures to reduce health disparities for 2SLGBTQI+ young adults.