Population Modeling
Population modeling has a major influence on the regulatory framework of tobacco products.
Population modeling is a key component in meeting regulatory requirements for traditional tobacco and next-generation tobacco products, such as oral and electronic nicotine-delivery system products. Population modeling is a complex process that requires vigorous fundamental procedures for safeguarding clean data, proper computing platforms, suitable resources and effective communication.
Saul Shiffman, senior scientific advisor of behavioral science, study design and analysis at Pinney Associates, said population health is at the center of the U.S. Food and Drug Administration’s (FDA) framework for regulating tobacco products.
“We want to understand what the impact is of the entire population. And that includes people who perhaps are intended to use the product, such as the adult smokers, and people who are not intended to use the products, so those former smokers or perhaps youth,” said Shiffman. “So, there’s a balance of benefits and harms, and the purpose of modeling is to integrate all of those to look at the net effect.”
Although there are many ways that researchers can implement population models, the basic principles are simple, according to Schiffman. The model defines the transitions or flows between the stages of tobacco use (going from being a never smoker to a smoker and going from being a smoker to being an abuser of cigarettes). “We start with a model that’s referred to as the base case—that is, what does the world look like now before a policy change or product introduction?” he explains. “And then, we contrast that to [a] counterfactual case, which is basically what do we expect to happen once this policy change or product introduction is implemented?”
Everyone models. They just do it very informally, said Shiffman. “If you [or] someone has ever thought to themselves, ‘How much does the harm that a new product might do to youth compare to the benefit to adult smokers?’ you’re doing modeling,” he said. “You’re just doing it nonquantitatively, intuitively. [What experts] have done is to do it very systematically and quantitatively.”
Ray Niaura, professor of social and behavioral science for the College of Global Public Health at New York University, said that conventional statistical analyses of data gathered is all about the past. It’s analyzing what happened, and by definition, the results are the results—it’s over and done with. However, population modeling is very different.
“It’s very hard to look at what happened in the past and project that into the future without a formal toolbox and framework to do that, and that’s what modeling represents,” he said. “It’s really the tools that allow us to go from the past to the future, the multiple futures. Any other statistical analysis just does not permit it. So, that’s why population modeling is a great set of tools.”
Population models are based on complex algorithms, according to Ryan Black, senior director of psychometrics, analytics and methods at Juul Labs. However, at its core, it’s quite straightforward.
“It really is. It’s posing the question whether or not—in the tobacco regulatory research—more people are going to move down the continuum of harm versus move up, and it does a fair assessment, taking into account both beneficial pathways—that is, cigarette smokers switching exclusively to noncombustible products as well as initiation, as well as relapse,” he said. “But a model is only [as] good as its inputs.”
Benjamin Apelberg, director of the Division of Population Health Science at the Center for Tobacco Products in the Office of Science at the FDA, said that Black touched on one of the more challenging questions faced by tobacco companies: trying to understand how new products, novel products, are going to behave once they’re on the market.
“Ultimately, it tends to be a kind of triangulation of different types of information that can be informative in a premarket setting. Sometimes we see actual use studies, so over a short period of time trying to understand how consumers respond to a product. We’ve seen experimental studies, consumer perception studies trying to understand at least the marketing and the positioning of a product.”
Schiffman said that while modeling is a powerful tool, it isn’t perfect. He said an example of the fallacies can be seen in most surveys pertaining to youth use. “I’m thinking here not about publications, but what’s in the press often neglects the fact that when they’re talking about use, they’re talking about any use in the past 30 days,” he said. “Someone who had a puff on a friend’s cigarette at a party is counted as a user, and that obviously leads to confusion because that has no health impact, and yet it’s what we count as use.”
According to David Levy, professor of oncology at Georgetown University, models become useful in not what they tell researchers but in that they suggest what is important. “We live in a world that can be simplified … what models can do is get us started thinking about what factors are important,” he said. “They ultimately tell us what pathways are important.”
