Study: To the Combat Youth Vaping Epidemic, AI Can Help Determine Cessation Strategies That Will Work Best

While potential health risks are widely known, people find it hard to stop vaping, and the younger they are, the harder it is.

These factors led University at Buffalo cancer researchers to launch a study on why young people vape, and the best ways to get them to stop. They conducted an online survey of 119 people who vape, with three quarters of them aged 21-26.

The results were published May 5 in PLOS Digital Health. Supriya D. Mahajan, PhD, MPH, associate professor of medicine in the Jacobs School of Medicine and Biomedical Sciences at UB, is senior corresponding author. 

“As cancer researchers in the divisions of Hematology/Oncology and Allergy, Immunology, and Rheumatology at UB, we see the direct clinical consequences of nicotine dependence in our patients,” says Satheeshkumar Poolakkad Sankaran, DDS, first author of the study and research scientist in the Division of Hematology/Oncology in the Department of Medicine. “We wanted to understand not only who is vaping but also who is successfully quitting—and then translate those insights into better cessation support for our cancer patients and into broader social determinants of health research.”

To figure that out, they applied AI techniques, such as machine learning, to the question of why some strategies designed to get youth to stop vaping work with some individuals and not with others. Their findings are applicable not just to vaping but to many other public health issues as well.

They tested five different computer models to predict who would successfully quit vaping.

“The simplest and most reliable ones were like a smart checklist that automatically figured out which life factors mattered most in deciding whether or not to stop vaping,” says Poolakkad Sankaran.

By contrast, he says, the more advanced “explainable AI” tools act like a translator for the computer’s decisions. “For example, the model called Accumulated Local Effects (ALE) shows how each factor—for example, being under age 21—changes the odds of quitting across the whole group, almost like a graph of ‘what-if’ scenarios,” he says.

Another model, Local Interpretable Model-Agnostic Explanations (LIME), zooms in on individual people. “It can look at one specific vaper and say, ‘For this person, social triggers are the biggest barrier—here’s exactly how much they lower their chance of success,’” says Poolakkad Sankaran. 

The researchers found that the earlier someone starts vaping, the harder it is for them to quit later. Poolakkad Sankaran says their models revealed this finding very clearly.

“Starting before age 18, and especially before 15, was one of the strongest predictors of continued use,” he says. “This tells us that prevention must begin early, before the brain’s reward system becomes wired to nicotine. For kids who already started young, the message is hopeful but urgent: The sooner they get help, the better their chances.”

He says programs geared toward getting kids to stop vaping should therefore focus on people under age 21. “These should be age-tailored strategies: short, frequent digital nudges; peer support; and trigger-management tools, because their developing brains make these people especially vulnerable but also especially responsive to timely intervention,” he explains.

Poolakkad Sankaran says that based on their findings, he thinks universities like UB can play a critical role in getting students to stop vaping. “UB is uniquely positioned to translate these exploratory machine learning/explainable AI results into real-world programs that reduce nicotine addiction, lower long-term health care costs and address health disparities affecting Buffalo’s young population,” he says.

University health services and local county health departments can use the insights the researchers gleaned to create personalized text-message campaigns that specifically address the main local triggers. He adds that campus apps or quit lines could flag high-risk students, i.e., those who are younger, are frequent users and are vulnerable to vaping in social situations, and then offer immediate, tailored support.

The researchers plan to integrate the model into existing digital tools, like an expanded version of the ‘This is Quitting’ vape-cessation program, so counselors can receive plain-language explanations of why a particular student is struggling and what intervention is most likely to work.

“Because the study was done locally with Western New York participants, the findings already reflect the realities our students and young adults face,” says Poolakkad Sankaran.

He adds that the research demonstrates how machine learning and predictive analytics can help public health teams move from one-size-fits-all programs to precision interventions by identifying who is most at risk and what will actually help them before they drop out of treatment.

“Our study pushes the field forward by showing that explainable AI (XAI) can make these powerful tools transparent and trustworthy for clinicians and policymakers,” he says. “Instead of a black-box prediction, we deliver actionable, human-understandable explanations that can be directly built into digital health apps and community programs.”

Raw data for the survey were collected by Mikail Ebanks, a co-author, member of Mahajan’s lab in UB’s Clinical and Translational Research Center and a master’s candidate in the School of Public Health and Health Professions.

Additional co-authors, who also have volunteered with UB Hematology/Oncology's outreach initiative, Neighborhood Network of Integrated Care (NICE), include Ian Lango, a master’s candidate; Swarnali Zafor, an MD candidate; Rahul Kumar Das, PhD, postdoctoral associate; Kit Wai Cheung, a UB neuroscience graduate; and Roberto Pili, MD, associate dean for cancer research and integrative oncology, all of the  Jacobs School.