Main Results
Table 1 shows the combined NHIS sample distribution. The total sample consists of nearly one million observations (N = 870,652) and is majority NH White, majority female, majority never smoking/e-cigarette using, and is approximately evenly distributed by age category.
Table 1
Combined Sample characteristics
Demographic | Percent of sample % (n) |
Total | N = 870,652 |
Age | 18–34 | 31.8 (252,410) |
35–54 | 36.9 (309,068) |
55+ | 31.3 (309,174) |
Race/Ethnicity | Hispanic | 12.4 (128,597) |
NH White | 70.7 (576,919) |
NH Black | 11.5 (119,210) |
NH Other | 5.4 (44,869) |
Sex | Female | 51.9 (489,143) |
Male | 48.1 (381,506) |
Cigarette Smoking Status | Current | 20.6 (180,511) |
Former | 22.4 (196,163) |
Never | 57.0 (486,181) |
E-Cigarette Use Status | Current | 3.5 (6,013) |
Former | 11.2 (19,931) |
Never | 85.3 (158,470) |
Root mean square errors of all models were consistent and small relative to the y-axis scale, ranging from 0.518 to 1.115, at least one order of magnitude smaller than cigarette smoking prevalence (see Additional File 1, Supplementary Table 1).
Figure 1 shows the results of counterfactual trend modelling among all adults. Smoking prevalence declined steadily from 1990–2010. This decline apparently accelerated in the post-2010 period, where actual smoking prevalence was as much as approximately 3.4 ± 0.5 (SE) percentage points lower than projected. This smoking discrepancy coincided with a rise in e-cigarette use prevalence to approximately 4.5 ± 0.2 percent of adults in 2019. The correlation between smoking discrepancy and e-cigarette use prevalence from 2010–2019 was high and statistically significant (Pearson r = 0.803, p = 0.005).
Figure 2 shows the results by age group. Smoking prevalence declined steadily from 1990–2010 among 18–34 and 35–54 year olds, while smoking prevalence was more stable among those aged 55+. The discrepancy between projected and actual smoking prevalence was most pronounced among 18–34 year olds, with discrepancies up to 8.0 ± 0.9 percentage points. This age cohort also had the highest e-cigarette use prevalence, with approximately 8.2 ± 0.4% of 18–34 year olds being current e-cigarette users in 2019. Smoking discrepancies were approximately half as pronounced among 35–54 year olds as they were among 18–34 year olds, but were still substantial (up to 3.5 ± 0.7 percentage points). E-cigarette use prevalence among 35–54 year olds was approximately 4.6 ± 0.3% in 2019, which is also about half the prevalence among 18–34 year olds. Smoking discrepancies were not apparent among 55 + year olds, with most actual smoking prevalence estimates from 2010–2019 falling within the 95% confidence limits of the counterfactual model, as seen in Fig. 2. This age cohort also had the lowest e-cigarette use prevalence at 1.4 ± 0.1% in 2019. Correlation between smoking discrepancy and e-cigarette use was higher among 18–34 year olds (r = 0.869, p = 0.001), followed by 35–54 year olds (r = 0.614, p = 0.06), and those age 55+ (r = 0.115, p = 0.8).
Figure 3 shows the results by sex cohort. Similar to the age cohort results, smoking discrepancies were most pronounced for the cohort with the highest e-cigarette use prevalence. Among males, smoking discrepancies up to 4.2 ± 0.6 percentage points were observed, while among females, smoking discrepancies up to 2.5 ± 0.6 percentage points were observed. E-cigarette use prevalence meanwhile was approximately 5.5 ± 0.3% among males and 3.5 ± 0.2% among females in 2019. Correlation between smoking discrepancy and e-cigarette use was stronger among males (r = 0.869, p = 0.001) than among females (r = 0.634, p = 0.05).
Finally, Fig. 4 shows the modelling results by race/ethnicity cohort. From 1990–2019, smoking prevalence declined consistently among all three race/ethnicity cohorts. Smoking prevalence discrepancies up to 4.2 ± 0.6 percentage points were observed among the NH White cohort, whereas discrepancies were less apparent among the NH Black and Hispanic cohorts (up to 1.9 ± 1.2 and 2.0 ± 0.8 percentage points respectively). E-cigarette use prevalence in 2019 was highest among NH White individuals (5.1 ± 0.2%) compared to NH Black (3.4 ± 0.4%) and Hispanic (2.8 ± 0.3%) individuals. Finally, correlation between e-cigarette use prevalence and cigarette smoking discrepancy was greatest for the NH White cohort (r = 0.804, p = 0.005) followed by the NH Black (r = 0.676, p = 0.03) and Hispanic (r = 0.570, p = 0.09) cohorts.
Sensitivity Test Results
Results from the main analyses were largely robust to the five sensitivity tests described in the Methods, namely, (1) excluding the 2019 point estimates due to NHIS survey changes; (2) excluding the regression-estimated e-cigarette prevalence; (3) using 2009 as an alternative to the Kneedle cut-off year of 2010; (4) using 2011 as an alternative to the Kneedle cut-off year of 2010; and (5) using exponential decay functions instead of linear functions (see Additional File 1, Supplementary Table 2).
For the total sample (all adults), the Pearson correlation between smoking discrepancy and e-cigarette use prevalence were similar to the main result (main result: r = 0.803, range of r values across sensitivity tests (lowest to highest): r = 0.679–0.843; p < 0.05 for 4/5 tests). This was also true for the 18–34 cohort (main result: r = 0.869, test range: r = 0.789–0.889; p < 0.05 for 5/5 tests), the Male cohort (main result: r = 0.869, test range: r = 0.782–0.897; p < 0.05 for 5/5 tests), the NH White cohort (main result: r = 0.804, test range: r = 0.668–0.840; p < 0.05 for 4/5 tests), the NH Black cohort (main result: r = 0.676, test range: r = 0.572–0.742; p < 0.05 for 3/5 tests), and Female cohort (main result: r = 0.634, test range: r = 0.488–0.700; p < 0.05 for 3/5 tests).
For the 55 + cohort, the main correlation result differed more substantially from the sensitivity test results (main result: r = 0.115, test range: r=-0.011–0.452), however correlations were consistently low (below r = 0.5) and non-significant in all analyses for this cohort.
For the 35–54 cohort, while the main correlation result was high (r = 0.614), the correlations across sensitivity tests ranged from low (r = 0.386) to high (r = 0.692). This is also true for the Hispanic cohort for which the sensitivity tests also ranged widely (r = 0.175–0.728), but only reached significance in sensitivity tests and not in the main analysis.
Root mean square errors for non-linear models were the same as those for the linear models to between one and three significant figures (see Additional File 1, Supplementary Table 1), suggesting little difference between the linear and non-linear fits for these data.
Other Considerations
The effect of the FSPTCA and the CDC’s ‘Tips®’ campaign, which represent major, distinct national population interventions, were considered by comparing quantitative estimates for the association between these two interventions and smoking prevalence from the published literature, to the smoking prevalence observed in the present study.
The association between the Tips® campaign and smoking prevalence is quantified in the literature by a CDC study which estimated approximately one million Tips® campaign-associated sustained quits between 2012 and 2018 [41]. This equates to 0.4 percentage point decrease in smoking prevalence (because one million adults represent approximately 0.4% of the US adult population [43]). By comparison, in the present study, a 3.3 ± 0.5 percentage point smoking discrepancy was observed among all adults in 2018. Because the 3.3 ± 0.5 percentage point discrepancy observed in the present study is much greater than the 0.4 percentage point decrease in smoking prevalence associated with Tips®, Tips® does not explain the smoking discrepancy observed.
The association between the FSPTCA and smoking prevalence is quantified in the literature by a study which estimated a 0.6% reduction in US adult smoking prevalence each quarter following implementation of the FSPTCA in June 2009 [42]. Cumulatively, this would result in a 24% reduction in adult smoking prevalence from mid-2009 to mid-2019 (0.6% times 40 quarters). NHIS smoking prevalence among all adults in 2009 was approximately 20.6 ± 0.4% (present study). Applying the 24% reduction associated with the FSPTCA to the 2009 NHIS smoking prevalence provides a predicted adult smoking prevalence of approximately 15.7% in 2019, due to the FSPTCA. By comparison, in the present study the actual NHIS smoking prevalence in 2019 was approximately 14.0% (95% CI: 13.5–14.5%), which is statistically lower than the 15.7% prevalence from the FSPTCA. Because the actual NHIS smoking prevalence of 14.0% is statistically lower than the 15.7% prevalence predicted from FSPTCA effects, FSPTCA effects do not explain the smoking prevalence observed.