Assessing population health effects of tobacco product use for regulatory compliance: a comparison of available methods
The US FDA encourages use of statistical models to project effects of a new tobacco product on population health, for Modified Risk Tobacco Product Applications (MRTPAs) and Pre-Market Tobacco Applications (PMTAs). We review different methods chosen for recently developed models. “Cohort models” follow a single cohort through the death of each member, while full (“cross-sectional”) population models follow a population of mixed ages and tobacco exposures over a specified period, including births as well as deaths. Both types either calculate expected counts of members of each product use category, or use summary statistics from Monte Carlo simulation of tobacco use histories of many individuals. Output metrics include the difference in cumulative deaths between a reference scenario and a scenario where a new product is added, or may use measures such as smoking-attributable deaths or life years, perhaps quality-adjusted. Each choice has advantages; we give examples of each model type and review pros and cons.
Full cross-sectional population models have been criticized as overly complex, subject to bias, and limited to inadequately short time periods, unlike cohort models. However, we demonstrate that a full population model does not need not to have any of these limitations, and can even reduce to a cohort model if birth rates are set to zero (and initial ages are set to a single age if desired). Thus, additional complexity becomes the user’s choice. In return for the birth rate requirement, full population models project annual prevalence of use of each product and mortality. This allows validation to include running over historical years to check evolution of the total population, smokers, and former smokers, all with age and sex breakdowns.