LC–MS Validates ML Models for Vaping Nicotine

Liquid chromatography-mass spectrometry (LC–MS) plasma nicotine data trains machine learning (ML) models for vaping exposure.

Researchers based at Florida International University (Miami, Florida) set out to build machine learning (ML) models that could predict how much nicotine an e-cigarette emits, as well as how much nicotine winds up boosting a vaper's bloodstream. To make these predictions, the models looked at puffing patterns (how hard, how long, how often someone vapes), details about the person and their device, and how much e-liquid got used up in the process. To gather the data, they worked with 259 measurements from vapers aged 21 to 35. Each person vaped normally while their puffing behavior was recorded, then a "puffing robot" replayed those exact puffing patterns in the lab so the team could measure precisely how much nicotine came out. Plasma nicotine levels of the participants were determined via liquid chromatography–mass spectrometry (LC–MS). A paper based on this work was published in Scientific Reports .1

Why Is It So Hard to Measure Nicotine Exposure from Vaping Accurately and at Scale?

Vaping has exploded in popularity around the world, changing the way people use nicotine and raising real concerns about how much they're being exposed to, how addictive it is, and how it should be regulated. The newest generation of pod-style vapes deliver nicotine so efficiently that they can match or even surpass what you'd get from a regular cigarette. Since nicotine is the main ingredient that keeps people hooked and craving more, getting an accurate read on how much nicotine someone's absorbing is key to figuring out how risky these products are, understanding addiction, and shaping smart regulations.2-5

Traditionally, figuring out how much nicotine someone gets from vaping has meant relying on laboratory work. One method involves recreating a person's exact puffing pattern using a "smoking robot," then testing the vapor it produces for nicotine. Another involves drawing blood to measure nicotine levels directly in the bloodstream, which shows how much actually got absorbed into the body. Both methods are accurate, but they're also expensive and slow: they need specialized equipment, trained staff, and a proper lab setting, which makes them hard to use for big studies or everyday monitoring. To get around these hurdles, researchers have started turning to statistical and machine-learning approaches instead.6,7

In the minds of the researchers, the challenge is that these biological markers can be thrown off by all sorts of unrelated factors, like exercise or stress, and they do not account for what is driving nicotine exposure in the first place: how someone vapes and what device they are using. Past research shows these factors, puffing style, nicotine strength, and the device itself, all play a big role. Yet no model out there pulls all these pieces together to reliably predict nicotine exposure at scale. A model that did would, according to the researchers, better reflect real-world exposure and could be a genuinely useful tool for regulators and public health tracking.1

What Model and Inputs Best Predict Nicotine Emissions and Blood Nicotine Boost from Vaping Behavior?

The researchers used six supervised regression models (ordinary least squares, lasso, random forest, XGBoost, support vector regression, and neural networks) to see which one could best guess nicotine emissions and the nicotine boost in someone's blood, just from their vaping habits. To test how well each tool worked, they split the data so most of it trained the models while a smaller chunk checked how accurate the predictions were, and they repeated this process 200 times to make sure the results were trustworthy. They then judged accuracy using three standard measures of how close the predictions came to reality.1

Of the six models used, XGBoost, came out on top every time, giving the most accurate and consistent predictions for both nicotine emissions and the nicotine boost in the blood. When predicting how much nicotine came out of a device, XGBoost worked best when it factored in either how many puffs someone took or how much e-liquid they used up, combined with how long each puff lasted and details about the device itself, like its nicotine strength. This combo explained roughly 75–78% of the variation in nicotine emissions.1

Then, taking those emission predictions and adding in personal details like age, sex, height, and weight, the model got even more useful at predicting the actual nicotine boost in someone's bloodstream, explaining about 61% of the variation when based on puff count, and about 70% when based on e-liquid consumption.1

“This study,” writes the authors of the paper,1 “shows that ML methods can effectively estimate nicotine emissions and plasma nicotine boost exposure directly from puffing behavior and device features. These models can support product regulation, assess addiction risk, and facilitate population-level surveillance.”

The researchers believe that their study's biggest strength is that it traces the whole nicotine journey, from how someone vapes, to what the device emits, to how much hits their bloodstream, using real laboratory measurements rather than guesses. It also showed the relationship is not a simple straight line like previously assumed. Having participants abstain for 12 hours beforehand (confirmed by blood tests) gave everyone a clean baseline, so results reflected the device and user, not leftover nicotine from earlier use.The team believes that future research should test these models on larger, more diverse groups and a wider range of devices and e-liquids.1

LC–MS Validates ML Models for Vaping Nicotine

Comments