Trends in cigarette consumption across the United States, with projections to 2035

Prior linear projections suggested US cigarette smoking could be eliminated by 2035 if previous national trends continued, but such targets may be unrealistic without extraordinary resources or comprehensive bans. Behavioral change often follows an S-shaped trajectory rather than linear decline, and national averages can mask wide state-level variation (e.g., 2020 BRFSS adult smoking prevalence ranged from 7.9% in Utah to 23.8% in West Virginia). Many jurisdictions have adopted ambitious targets (e.g., Healthy People 2030: ≤5% smoking prevalence; Canada: ≤5% any tobacco use by 2035; Australia, Finland, Ireland, New Zealand with various ≤10% or ≤5% goals). Few analyses have quantified state-by-state chances of achieving such targets. This study aimed to project state-level per capita cigarette consumption using 70 years of sales data, quantify inequity across states using the Gini coefficient, forecast to 2035, calibrate a ppc target approximating 5% prevalence via BRFSS, estimate each state’s probability of achieving this target by 2030 and 2035, and identify realistic (50% chance) and aggressive (25% chance) state-specific targets.

Data sources: Annual state-specific per capita cigarette consumption (ppc; tax-paid packs per capita) for the 50 US states were obtained from the Tax Burden on Tobacco reports for 1950–2020 (N=3,550). Tax-paid sales are derived from state excise tax stamp purchases or wholesaler filings, with population denominators from US Census Bureau estimates as of July 1 for each fiscal year. Mapping ppc to a 5% prevalence target: Using 2011–2020 BRFSS adult current smoking prevalence by state (≥100 lifetime cigarettes and current every day/some days smoker), a linear mixed-effects model regressed log(ppc) on log(prevalence) with a state random intercept (lme4 in R). The model explained 97% of variance in ppc (conditional R2=0.97). Predicted ppc corresponding to 5% prevalence yielded state-specific values with a population-average ≈13 ppc (SD 3.8), used as the idealized target. Summarizing historical trends: For each state, linear regression summarized annual percent change in ppc since 1980; distributions across states were summarized by mean and SD. ARIMA model selection and evaluation: For each state, ARIMA(p,d,q) models on log(ppc) were fit using the Hyndman–Khandakar algorithm (auto.arima in R forecast package), allowing differencing determined by unit root tests, p and q by AICc minimization, and optional drift. Forecast accuracy for a 15-year horizon was assessed by training up to 2005 and comparing 2006–2020 forecasts by mean absolute error (MAE) against exponential smoothing (ETS) models. Given higher accuracy when training on post-1980 data, final models were trained on a 25-year window (1996–2020), which outperformed alternatives on average. Final model specifications are in the Supplement (Table D). Forecasting 2021–2035: For each state, forecasts used 1,000 simulated future paths via residual resampling to allow uncertainty to grow over time; sensitivity with 5,000 resamples showed stable estimates. Medians and 95% prediction intervals (2.5th–97.5th percentiles) summarize forecasts. The proportion of simulated ppc ≤13 quantified the probability of achieving the ideal target by year. The 25th and 50th percentiles of simulated ppc defined aggressive and realistic targets, respectively, for 2030 and 2035. Inequity across states: Annual Gini coefficients summarized dispersion of ppc across states (0=complete equity, 1=maximal inequity). Post-1984 trends in Gini were estimated by linear regression. Forecast-year Gini distributions (n=1,000 per year) were derived from the simulated state ppc forecasts, summarized by medians and 95% PIs.

- From 1980 onward, average state ppc declined 3.3% per year (SD 1.1%), with wide inter-state variability. - Case examples: In 1980, ppc was high in California (120.2), Michigan (140.7), and Missouri (142.1). Decline rates: California −5.2%/yr (95% CI: −5.4, −5.0), Michigan −3.5%/yr (−3.7, −3.3), Missouri −1.6%/yr (−1.7, −1.4). By 2020, ppc: California 15.1, Michigan 39.9, Missouri 72.1. - Inequity: Gini reached 0.09 in 1984, then increased 2.8%/yr (95% CI: 2.5%, 3.1%) through 2020. Projected Gini rises from 0.24 (2020) to 0.31 (2030; 95% PI: 0.29, 0.34) and 0.35 (2035; 95% PI: 0.32, 0.39), a 48.1% increase from 2020 (95% PI: 35.3%, 64.2%). - State target attainment: Eight states have ≥50% probability of reaching ≤13 ppc by 2030 (CA, NY, UT, WA, MA, CT, AZ, NM), with four more by 2035 (IL, NJ, MN, MD). Twenty-two states have essentially no chance (<1%) and seven states have <5% chance to reach ≤13 ppc by 2035. - 2035 forecast examples: California 6.8 ppc (95% PI: 4.6, 9.0); Michigan 25.3 (17.6, 35.4); Missouri 50.8 (40.9, 65.0). California’s level projected to be 3.7× lower than Michigan’s and 7.4× lower than Missouri’s by 2035. - Example targets: California realistic <9 ppc (2030) and <7 (2035); aggressive 8 (2030) and 6 (2035). Michigan realistic 29 (2030), 25 (2035); aggressive 27 (2030), 22 (2035). Missouri realistic 57 (2030), 51 (2035); aggressive 53 (2030), 47 (2035).

State-level cigarette consumption has declined everywhere but at markedly different rates since the 1980s, producing growing inequity across states that is projected to continue through 2035. Only a minority of states are on track to reach consumption levels approximating an idealized ≤5% smoking prevalence by 2035. Nonetheless, all states can reduce consumption further. States with comprehensive tobacco control programs (e.g., California, New York, Washington) are positioned to meet ideal targets, reflecting sustained implementation of best practices (WHO MPOWER and US CDC recommendations), including higher excise taxes (e.g., by 2022: NY $4.35/pack, WA $3.02, CA $2.87 vs US state average $1.91) and comprehensive smoke-free workplace laws. Additional policy innovations (e.g., menthol/flavor bans in Massachusetts) can further reduce consumption even in strong-program states. National and local actions complement state efforts, as seen with local smoke-free ordinances and Tobacco 21 policies later adopted federally. The findings underscore the importance of expanding comprehensive tobacco control strategies tailored to lagging states to close gaps and improve the likelihood of achieving more aggressive targets.

Using the longest continuous state-level cigarette sales series available, this study applied state-specific ARIMA forecasting to project per capita cigarette consumption through 2035, quantified the growing inequity in consumption across states, and estimated each state’s probability of achieving a consumption threshold approximating ≤5% adult smoking prevalence. While ideal national targets are unlikely for most states by the next decade, every state can make progress; the study provides realistic (50% chance) and aggressive (25% chance) targets to guide policy. Continued and expanded comprehensive tobacco control—tax increases, smoke-free policies, cessation support, marketing restrictions, and local initiatives—are needed to reduce consumption and inequity. Future research should integrate policy projections, address illicit trade and demographic dynamics, and compare alternative forecasting approaches and data sources.

- The mapping from per capita consumption to prevalence (≈13 ppc as proxy for 5% smoking prevalence) assumes the 2011–2020 relationship remains stable through 2035. - Forecast uncertainty reflects only residual error; uncertainty in autocorrelation parameters and model order is not fully propagated, potentially yielding narrow prediction intervals. - Sales-based ppc can be biased by illicit trade and cross-border purchasing, which may differ by state. - Population structure changes may not be fully captured by aggregate trends. - Models do not explicitly incorporate future policy changes (e.g., tax increases, flavor bans) or co-integrated processes of initiation and cessation; alternative models and survey-based validations could refine projections.