Cumulative psychosocial and health disparities in US adolescent cigarette smoking, 2002-2019

Abstract

Background

Understanding disparities in adolescent cigarette smoking is important for effective prevention.

Methods

We investigated disparities in adolescent smoking based on cumulative reported psychosocial and health risk among respondents ages 12-17 years in the US National Survey of Drug Use and Health from 2002 to 2019. Multivariable regression estimated associations of cumulative risk, survey years, and their interaction predicting past-month and daily smoking. Eleven psychosocial and health variables associated with youth smoking formed composite measures of cumulative risk, categorizing risk as low (0-2), moderate (3-4), or high (≥5). The main outcomes were weighted past-month and daily smoking by cumulative risk and time, examining prevalence and proportional change across years.

Results

Among 244 519 adolescents, greater cumulative risk predicted higher smoking prevalence across all outcomes. Compared with the low-risk category, past-month smoking odds (adjusted odds ratio) were 9.14 (95% confidence interval [CI] = 8.58 to 9.72) and 46.15 (95% CI = 43.38 to 49.10) times greater in the moderate-risk and high-risk categories. For daily smoking, odds were 14.11 (95% CI = 11.92 to 16.70) and 97.32 (95% CI = 83.06 to 114.03) times greater among the moderate-risk and high-risk categories. Regarding proportional change, the low-risk category exhibited the steepest decline (−85.1%) in past-month smoking from 2002-2003 to 2018-2019, followed by the moderate-risk (−79.2%) and high-risk (−65.7%) categories. Daily smoking declined more steeply among the low-risk (−96.5%) and moderate-risk (−90.5%) than high-risk category (−86.4%).

Conclusions

Cumulative risk is a robust predictor of adolescent smoking. Although record-setting reductions in adolescent smoking extend across risk categories, disparities favoring youth with fewer risks are evident throughout. Recognizing cumulative risk can inform the development of more targeted and effective prevention efforts.

Introduction

Cigarette smoking remains a leading cause of preventable death in the United States, responsible for an estimated 480 000 annual deaths.¹ Approximately 88% of adults who smoke initiated smoking as adolescents.² Much effort has been directed toward reducing cigarette smoking among adolescents to great effect. These efforts are associated with record-low prevalence rates, a notable public health achievement.^3,4 Compared with 1997, daily cigarette use among eighth, tenth, and twelfth grade students has decreased from 9.0%, 18.0%, and 24.6% to 0.8%, 1.2%, and 3.1% in 2020, respectively.⁴

A critical question is whether declines in smoking prevalence are observed across all populations or whether at-risk groups disproportionately account for them. Despite this striking decline, much is still unknown about disparities in adolescent cigarette smoking, including how and in what ways the accumulation of various psychosocial- and health-related risk factors predict cigarette use. Among adults, smoking risk increases in an orderly, summative manner by the total number of risks experienced (ie, cumulative disadvantage), and populations with higher cumulative risk have slower rates of decline in smoking prevalence.^5-7 Whether similar trends are observed in adolescent smoking is unknown. A better understanding of the relationship between disparities and adolescent cigarette smoking risk can inform targeted prevention and cessation interventions to eliminate the cigarette-smoking epidemic.^8-10

We examined trends in disparities in adolescent cigarette smoking and whether a cumulative-risk model that effectively characterizes adult smoking risk extends to adolescents who participated in the National Survey on Drug Use and Health (NSDUH) between 2002 and 2019.¹¹ We examined associations between cumulative risk and past-month and daily cigarette smoking prevalence across time, with daily smoking indicative of the onset of chronic smoking and associated adverse health outcomes.^12,13

Methods

Participants

This sample included a nationally representative group of adolescents aged 12-17 years participating in the NSDUH from 2002 to 2019 (see Table 1). The NSDUH is an annual, cross-sectional population-based survey of noninstitutionalized residents from all 50 US states and the District of Columbia assessing tobacco, alcohol, drug use, mental health, and other health variables. Survey data were collected via face-to-face household interviews. All youth respondents and parent participants provided verbal informed consent before participation. We did not include data before 2002, when the survey’s title was changed, sampling was improved, and respondents began receiving monetary compensation for survey completion or beyond 2019 because of COVID-19–related survey changes (ie, switching to remote data collection) precluding comparisons with prior years.¹¹ The National Center for Health Statistics ethics review board approved all data collection. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology reporting guidelines for cross-sectional studies.

Table 1.

Sample descriptive statistics and association of covariates with smoking status

Covariate	Prevalence, No. (%)	Past-month smoking		Daily smoking
		Smoking prevalence, %a	Difference, OR (95% CI)	Smoking prevalence, %a	Difference, OR (95% CI)
Gender
Female	120 746 (49.38)	7.50	1.01 (.97 to .1.05)	1.82	1.05 (.97 to 1.13)
Male	123 773 (50.62)	7.44	1 (Referent)	1.74	1 (Referent)
Race and ethnicity
Alaska Native or Native American	3303 (1.35)	12.30	1.38 (1.17 to 1.62)	2.91	1.16 (.83 to 1.61)
Asian	8154 (3.33)	2.60	.26 (.22 to .31)	0.30	.12 (.07 to .19)
Black	31 260 (12.78)	4.30	.44 (.41 to .47)	0.55	.21 (.18 to .26)
Hispanic	42 053 (17.20)	5.24	.54 (.51 to .58)	0.71	.27 (.24 to .32)
Native Hawaiian or Pacific Islander	994 (0.41)	5.36	.56 (.38 to .82)	0.55	.21 (.10 to .47)
More than 1 race	10 844 (4.43)	7.89	.84 (.75 to .94)	1.75	.69 (.55 to .85)
White	147 911 (60.49)	9.25	1 (Referent)	2.53	1 (Referent)
Age, y
12	35 031 (14.33)	0.67	1 (Referent)	0.04	1 (Referent)
13	39 821 (16.29)	2.16	1.45 (1.38 to 1.53)	0.15	1.82 (1.67 to 1.98)
14	41 358 (16.91)	4.29	2.36 (2.23 to 2.49)	0.55	3.76 (3.37 to 4.20)
15	42 807 (17.51)	7.71	4.40 (4.11 to 4.70)	1.43	9.85 (8.39 to 11.57)
16	43 369 (17.74)	11.93	8.93 (8.12 to 9.82)	2.92	36.21 (27.02 to 48.53)
17	42 133 (17.23)	16.45	29.10 (24.79 to 34.15)	5.18	132.66 (71.45 to 246.30)
Family income
<$20 000	38 407 (15.71)	9.14	1.64 (1.55 to 1.74)	2.59	2.44 (2.18 to 2.72)
$20 000-$49 999	75 845 (31.02)	8.73	1.56 (1.49 to 1.64)	2.28	2.14 (1.94 to 2.36)
$50 000-$74 999	44 884 (18.36)	7.58	1.34 (1.27 to 1.42)	1.76	1.64 (1.46 to 1.84)
≥$75 000	85 383 (34.92)	5.77	1 (Referent)	1.08	1 (Referent)
Residence
Rural	52 678 (21.54)	10.41	1.56 (1.49 to 1.63)	2.84	1.81 (1.67 to 1.96)
Urban	191 841 (78.46)	6.93	1 (Referent)	1.59	1 (Referent)

a

Weighted to the US population. OR = odds ratio; CI = confidence interval. 

Measures

Cigarette smoking status

The primary outcomes were past-month and daily cigarette smoking status. Participants who responded yes to smoking in the past month were classified under “past-month cigarette smoking.” Participants who reported past-month smoking were further classified by daily or nondaily use; participants who responded yes to daily smoking were classified under “daily cigarette smoking.”

Past-month and daily cigarette smoking were each examined for prevalence and proportional rate of change over time, totaling 4 outcomes. All data were aggregated into 2-year bins to reduce variability in the estimates. Prevalence change was quantified as the change over time (2002-2003 to 2018-2019) by risk category. Proportional change was quantified by comparing smoking prevalence with the 2002-2003 prevalence (baseline). This calculation subtracted the current smoking prevalence from baseline, divided by the baseline prevalence. The proportional change range is from -1 to 1, with negative scores indicating the degree of decrease from 2002 to 2003.

Risk factors

Eleven psychosocial and health variables formed an index composite measure of cumulative risk. Each variable was selected a priori as a general risk factor based on associations with smoking in the extant literature.^14-21 From the available variables in the NSDUH dataset, 13 candidate variables were selected. Two variables (social support, depression) were excluded because of a weak association with smoking (r < .10). Each of the 11 risk variables was coded into a binary yes-no variable (see Table 2). All risk factors with 2 or more questions were averaged to form a composite score.

Table 2.

Prevalence of risk and past-month and daily smoking status stratified by cumulative risk category and each risk aggregated across survey years, 2002-2019

	Prevalence No. (%)	Past-month smoking		Daily smoking
		Smoking prevalence, %a	Difference, AOR (95% CI)b	Smoking prevalence, %a	Difference, AOR (95% CI)b
Cumulative risk index
Low, 0-2	161 971 (67.81)	1.44	1 (Referent)	0.15	1 (Referent)
Moderate, 3-4	54 965 (21.86)	11.28	9.14 (8.58 to 9.72)c	1.93	14.11 (11.92 to 16.70)c
High, ≥5	27 583 (10.33)	38.97	46.15 (43.38 to 49.10)c	12.21	97.32 (83.06 to 114.03)c
Past-month alcohol binge drinking
No	223 568 (91.43)	4.35	1 (Referent)	0.89	1 (Referent)
Yes	20 951 (8.57)	43.68	4.69 (4.41 to 4.98)c	12.12	3.16 (2.86 to 3.48)c
Past-month marijuana use
No	225 301 (92.14)	4.33	1 (Referent)	0.87	1 (Referent)
Yes	19 218 (7.86)	47.02	4.55 (4.27 to 4.85)c	13.27	3.03 (2.73 to 3.35)c
Perceived indifference toward peer substance use
No	181 038 (74.04)	2.66	1 (Referent)	0.42	1 (Referent)
Yes	63 481 (25.96)	21.53	3.03 (2.87 to 3.21)c	5.76	3.43 (3.05 to 3.86)c
Perceived frequency of peer substance use
No	144 029 (58.90)	2.69	1 (Referent)	0.45	1 (Referent)
Yes	100 490 (41.10)	14.57	2.21 (2.09 to 2.33)c	3.76	2.57 (2.29 to 2.89)c
Past-year conduct problems
No	237 212 (97.01)	6.64	1 (Referent)	1.48	1 (Referent)
Yes	7307 (2.99)	37.16	1.91 (1.73 to 2.11)c	12.69	1.80 (1.59 to 2.05)c
Poor academic performance
No	181 387 (74.18)	5.12	1 (Referent)	1.02	1 (Referent)
Yes	63 132 (25.82)	14.68	1.83 (1.74 to 1.93)c	4.12	1.90 (1.65 to 1.97)c
Inconsistent seat belt use
No	193 156 (78.99)	5.94	1 (Referent)	1.27	1 (Referent)
Yes	51 363 (21.01)	13.88	1.58 (1.50 to 1.66)c	3.92	1.56 (1.43 to 1.71)c
No participation in prosocial activities
No	227 840 (93.18)	6.74	1 (Referent)	1.41	1 (Referent)
Yes	16 679 (6.82)	17.59	1.46 (1.35 to 1.57)c	6.96	2.23 (2.00 to 2.47)c
Poor overall health
No	187 079 (76.51)	6.13	1 (Referent)	1.23	1 (Referent)
Yes	57 440 (23.49)	11.96	1.46 (1.39 to 1.54)c	3.62	1.82 (1.67 to 1.99)c
Minimal parental involvement
No	210 169 (85.95)	6.08	1 (Referent)	1.33	1 (Referent)
Yes	34 350 (14.05)	16.13	1.22 (1.15 to 1.29)c	4.61	1.22 (1.11 to 1.34)c
No religious beliefs
No	169 414 (69.28)	5.30	1 (Referent)	1.07	1 (Referent)
Yes	75 105 (30.72)	12.58	1.22 (1.16 to 1.28)c	3.46	1.32 (1.21 to 1.45)c

a

Weighted to the US population. AOR = adjusted odds ratio; CI = confidence interval.

b

All predictors were entered into the model simultaneously, adjusting for covariates.

c

Statistically significant at a P value less than .001.

Past-month alcohol binge drinking. Responses to 1 question on past-month alcohol binge drinking, defined as 5 or more drinks (2002-2014) and 5 or 4 or more drinks on the same occasion for males and females (2015-2019), respectively, were scored as yes-no.

Past-month marijuana use. Responses to 1 question on past-month marijuana use were scored as yes-no.

Perceivedindifference toward peer substance use. Responses to 4 questions where participants indicated how they would feel about someone their age (1) smoking a pack of cigarettes daily, (2) trying marijuana, (3) using marijuana monthly, and (4) drinking at least 2 alcoholic beverages daily on a 3-point scale (1 = neither approve nor disapprove; 2 = somewhat disapprove; 3 = strongly disapprove). Averaged responses of at least 2.01 were categorized as “no” (disapproving) and responses of no more than 2 as “yes” (some indifference).

Perceived frequency of peer substance use. Responses to 4 questions where participants estimated how many grade-level peers used (1) cigarettes, (2) marijuana, (3) alcohol, and (4) “get drunk at least weekly” on a 4-point scale (1 = none; 2 = a few; 3 = most; 4 = all). Averaged responses of no more than 2 were categorized as “no”(not many) and responses at least 2.01 as “yes” (common).

Past-year conduct problems. Responses to 7 questions where participants estimated over the past year how many times they had (1) fought with a parent, (2) serious fight at school or work, (3) fought against another group, (4) carried a handgun, (5) sold illegal drugs, (6) stolen or tried to steal anything of at least $50, and (7) attacked someone with intent to cause serious harm on a 5-point scale (1 = 0 times; 2 = 1-2 times; 3 = 3-5 times; 4 = 6-9 times; 5 = 10 or more times). Averaged responses of no more than 2 were categorized as “no” (none or few) and responses of at least 2.01 as “yes” (conduct problems).

Poor academic performance. Responses to 1 question on the past semester’s grades were scored as “no” (good grades) (≥B) or “yes” (poor grades) (≤C).

Inconsistent seat belt use. Responses to 2 questions where participants reported their seat belt–wearing frequency when (1) riding in the front passenger seat and (2) driving using a 4-point scale (1 = never; 2 = seldom; 3 = sometimes; 4 = always). Averaged responses of at least 3.01 were categorized as “no” and responses of no more than 3 as “yes.”

No prosocial activities participation. Responses to 4 questions where participants estimated past year participation in (1) school-based activities (eg, sports, clubs), (2) community-based activities, (3) church and/or faith-based activities, and (4) other activities (eg, dance lessons) on a 4-point scale (0 = none; 1 = 1; 2 = 2; 3 = 3 or more). Averaged responses of at least .01 were categorized as “no” and responses equal to 0 as “yes.”

Poor overall health. Participants responded to 1 overall health question on a 5-point scale (1 = excellent; 2 = very good; 3 = good; 4 = fair; 5 = poor). Responses of no more than 3 were categorized as “no” (good health) and responses of at least 4 as “yes” (poor health).

Minimal parental involvement. Responses to 7 questions where participants reported over the past year if parents (1) checked homework completion, (2) helped with homework, (3) made them do work or chores, (4) limited TV time, (5) limited time with friends, (6) provided encouragement, and (7) expressed pride in an accomplishment on a 4-point scale (1 = always; 2 = sometimes; 3 = seldom; 4 = never). Averaged responses of no more than 2 were categorized as “no” and responses of at least 2.01 as “yes.”

No religious beliefs. Responses to 3 questions asking about religious beliefs and their (1) importance, (2) influence in decision making, and (3) importance others share in them on a 4-point scale (1 = strongly disagree; 2 = disagree; 3 = agree; 4 = strongly agree). Averaged responses of at least 2.01 were categorized as “no” and responses of no more than 2 as “yes.”

Risk index and cumulative risk categories

For each participant, each risk factor was added to calculate an index composite measure of cumulative risk. Participants were categorized into 3 levels of risk based on the distribution of cumulative risk index scores. Those with 0, 1, or 2 risk factors were categorized as low risk, 3 or 4 as moderate risk, and 5 or more as high risk.

Covariates

To address confounding of other demographics on relationships between cumulative risk and cigarette smoking, gender, race and ethnicity, family income, and residence (rural or urban) were included as covariates.⁷ Residence was coded using rural and urban continuum codes from the 2000 Census and 2013 Office of Management and Budget classifications.

Statistical analysis

Descriptive statistics for outcomes were generated by sociodemographic characteristics and by cumulative risk. All models were adjusted for all covariates and applied sampling weights based on sociodemographics of the US Census between 2002 and 2019. Logistic regression analyses tested whether past-month and daily smoking prevalence varied by sociodemographic characteristics and cumulative risk category across time. Linear regression analyses tested whether the proportional change in past-month and daily smoking prevalence compared with baseline (ie, 2002-2003) varied by sociodemographics and cumulative risk category across time. A sensitivity analysis was conducted to compare the 3-group categorization with a 5-group categorization for past-month and daily smoking prevalence.

Univariable analyses were conducted to test for statistically significant associations between each risk factor and the outcome variables. Each predictor with a statistically significant association (P < .05) was included as a predictor in the multivariable models. Multivariable logistic regression analyses were conducted to predict past-month and daily cigarette smoking status. Multivariable linear regression analyses were conducted to predict proportional change in past-month and daily smoking prevalence. All 11 risk variables were simultaneously included in all models.

Three cumulative risk models were tested for each outcome. In model 1, cumulative risk category was the only predictor. In model 2, time was entered as a predictor in addition to cumulative risk. Model 3 assessed a 2-way interaction of cumulative risk by time. Based on the interaction results, pairwise contrasts compared the low-risk category (referent) to the moderate-risk and high-risk categories. All statistical analyses were conducted using SAS version 9.4 statistical analysis software (SAS Institute, Inc, Cary, NC, USA) using the Proc SurveyMeans, Proc SurveyLogistic, and ProcSurveyReg procedures for descriptive statistics, logistic regression analyses, and linear regression analyses, respectively. Statistical significance level alpha was set a priori to 0.05 (2-tailed).

Results

Descriptive characteristics

This cross-sectional sample pooled across years included 244 519 adolescents ages 12-17 years (mean [SD] age = 14.6 [1.7]; 49.4% female) at survey completion from 2002 to 2019 (Table 1). Most respondents were non-Hispanic White (60.5%) and had an annual family income less than $75 000 (65.1%).

Table 2 summarizes the prevalence of the psychosocial- and health-related risk and smoking status aggregated across survey years, first by cumulative risk category then by each type of risk. All 11 risk factors were statistically significant predictors of past-month and daily smoking. The top 3 risk factors were past-month alcohol binge drinking, past-month marijuana use, and perceived indifference toward peer substance use. The weighted proportion of respondents (mean) in the low-, moderate-, and high-risk categories was 67.8% (95% confidence interval [CI] = 67.6% to 68.1%), 21.9% (95% CI = 21.6% to 22.1%), and 10.3% (95% CI = 10.2% to 10.5%), respectively.

Across risk categories, past-month smoking prevalence decreased in a linear trend from 12.4% (95% CI = 12.0% to 12.9%) in 2002-2003 to 2.5% (95% CI = 2.2% to 2.8%) in 2018-2019. Daily smoking decreased linearly from 3.7% (95% CI = 3.4% to 3.9%) in 2002-2003 to 0.3% (95% CI = 0.2% to 0.4%) in 2018-2019. Overall proportional (ie, relative) change in past-month smoking decreased from -6.1% (95% CI = -13.0% to 0.8%) in 2004-2005 to -82.2% (95% CI = -86.0% to -78.4%) in 2018-2019; daily smoking decreased from 2002-2003 to 2018-2019 by -94.3% (95% CI = -98.4% to -90.3%).

Smoking prevalence by cumulative risk

Prevalence decreased statistically significantly between 2002-2003 and 2018-2019 in each risk category for past-month (Figure 1, A) and daily smoking (Figure 1, B). Sensitivity analyses indicated similar results for the 3-group and 5-group categorizations (Figure S1). Table 3 summarizes the prevalence models for past-month and daily smoking (top half). Cumulative risk and time were statistically significant predictors of cigarette smoking in each model, suggesting greater risk was associated with higher smoking prevalence and decreasing prevalence from 2002-2003 to 2018-2019. The interaction of cumulative risk and time was statistically significant for past-month smoking indicating differences in prevalence change over time between cumulative-risk categories, but was not statistically significant for daily smoking.

Figure 1.

Past-month smoking and daily smoking percent prevalence and proportional change in prevalence over time by cumulative risk group.

Table 3.

Regression model outcomes for past-month and daily smoking prevalence organized by absolute prevalence and relative rates of change over time

Modela
Prevalence	β	OR (95% CI)
Past-month smoking
Cumulative risk	1.86b	6.40 (6.22 to 6.58)
Time	−.18b	.84 (.83 to .84)
Cumulative risk x time	.01c	1.01 (1.00 to 1.02)
Daily smoking
Cumulative risk	2.09b	8.10 (7.65 to 8.59)
Time	−.20b	.82 (.80 to .83)
Cumulative risk x time	.02	1.02 (.99 to 1.04)
Proportional change in prevalence	t	β (95% CI)
Past-month smoking
Cumulative risk	6.83b	.07 (.05 to .09)
Time	−25.63b	−.10 (−.11 to −.09)
Cumulative risk x time	4.34b	.02 (.01 to .02)
Daily smoking
Cumulative risk	−1.93c	−.06 (−.13 to .00)
Time	−10.18b	−.12 (−.14 to −.09)
Cumulative risk x time	2.44c	.03 (.01 to .05)

a

Weighted to the US population. All predictors were entered into the model simultaneously, adjusting for covariates. CI = confidence interval; OR = odds ratio.

b

Statistically significant at a P value of less than .001.

c

Statistically significant at a P value of less than .05.

For past-month smoking (Figure 1, A), prevalence (%) was greatest in the high-risk category from 2002-2003 through 2018-2019 (49.6% to 17.0%), followed by the moderate-risk (17.7% to 3.7%) and low-risk (2.7% to 0.4%) categories. Compared with the low-risk category, the adjusted odds ratio of past-month smoking was 9.14 (95% CI = 8.58 to 9.72) and 46.15 (95% CI = 43.38 to 49.10) in the moderate- and high-risk categories, respectively. A similar pattern was observed for daily smoking (Figure 1, B), with the greatest prevalence from 2002-2003 to 2018-2019 in the high-risk category (20.1% to 2.7%) followed by the moderate-risk (3.5% to 0.3%) and low-risk (0.3% to 0.0%) categories. Compared with the low-risk category, the adjusted odds of daily smoking were 14.11 (95% CI = 11.92 to 16.70) and 97.32 (95% CI = 83.06 to 114.03) times greater for the moderate- and high-risk categories, respectively.

Proportional change in smoking prevalence by cumulative risk

Proportional change from 2002-2003 prevalence in past-month (Figure 1, C) and daily smoking (Figure 1, D) varied by cumulative-risk category and time. The bottom half of Table 3 summarizes the models assessing proportional change for past-month and daily smoking. Cumulative risk and time were statistically significant predictors in each model, indicating greater cumulative risk was associated with less proportional change in smoking and time with decreasing smoking prevalence. The interaction of cumulative risk and time was also statistically significant for each outcome, indicating proportional changes in smoking prevalence over time varied between cumulative-risk categories.

Regarding past-month smoking (Figure 1, C; Table S1), the decline differed across cumulative-risk categories. The low-risk category exhibited the steepest proportional decline between 2002-2003 and 2018-2019 (−85.1%; β = −.11 95% CI = −.12 to −.10; t = −18.93; P < .001) followed by the moderate- (−79.2%; β = −.09 95% CI = −.10 to −.08; t = −25.43; P < .001) and high-risk (−65.7%; β = −.08, 95% CI = −.08 to −.07; t = −26.41; P < .001) categories. These differences in rate of decline statistically significantly differed between all pairs: low-risk vs moderate-risk (β = .02, 95% CI = .00 to .03; t = 2.47; P = .014), low-risk vs high-risk (β = .02, 95% CI = .01 to .02; t = 5.46; P < .001), and moderate-risk vs high-risk (β = .02, 95% CI = .01 to .03; t = 3.88; P < .001). For daily smoking (Figure 1, D; Table S2), the greatest proportional decline over time occurred in the low-risk (−96.5%; β = −.13, 95% CI = −.16 to −.10; t = −7.66; P < .001) and moderate-risk (−90.5%; β = −.09, 95% CI = −.11 to −.08; t = −12.39; P < .001) categories, with the least proportional change in the high-risk category (−86.4%; β = −.09, 95% CI = −.10 to −.08; t = −20.78; P < .001). The difference in rate of decline was statistically significant for low-risk vs moderate-risk (β = .04, 95% CI = .00 to .08; t = 2.15; P = .032) and low-risk vs high-risk (β = .02, 95% CI = .01 to .04; t = 2.54; P = .011), but not moderate-risk vs high-risk (β = .01, 95% CI = −.01 to .02; t = 0.57; P = .570).

Discussion

This study provides robust, novel evidence from a US nationally representative sample that adolescent smoking risk cumulatively increases corresponding to the number of psychosocial and health risk experienced. The study extends to adolescents a cumulative risk model of smoking risk previously documented in adults.^5-7 Moreover, this study documents that this cumulative risk model predicts risk for daily smoking, a pattern indicative of a transition to chronic smoking, dependence, and greater longer-term risk for smoking-related chronic disease and premature death.^12,13 Disparities in smoking risk associated with psychosocial- and health-related characteristics are well established in the tobacco control and regulatory literatures, including, for example, poverty; racial, ethnic, sexual, and minority status; and psychiatric illness. However, this cumulative risk model underscores that these risks often co-occur, which is novel and important. Thus, when assessing smoking risk, the number of smoking-related risks experienced is important to consider in addition to any specific risk.

This study provides evidence that although the record-setting reductions in cigarette smoking among US adolescents are far-reaching across US regions, ethnicity and race, and socioeconomic strata, they nevertheless continue to include disparities. We observed statistically significant decreases in past-month and daily cigarette smoking prevalence from 2002-2003 to 2018-2019, with reductions evident across adolescents in all cumulative-risk categories. Indeed, these results align with overall decreases in adolescent risk behaviors in general^22-24 and with changing norms and adolescent approval of cigarette smoking in particular.^25,26 However, these findings also demonstrate that past-month and daily cigarette smoking prevalence and proportional decline varied in an orderly manner by cumulative risk group, with high odds ratios indicating inequitable decreases in smoking. Specifically, our results suggest a widening effect of disparities on past-month and daily cigarette smoking. From 2002-2003 to 2018-2019, smoking remained most prevalent among adolescents with greater cumulative risk, and the least proportional decline occurred among those with the greatest cumulative risk. Notably, the robust differences in past-month and daily cigarette smoking in this cumulative-risk model were observed when controlling for sociodemographics and family income.

By examining prevalence change and proportional change,²⁷ we believe these findings contribute to better understanding the temporal changes in adolescent smoking with implications highlighting the need for more targeted prevention and treatment interventions. As adolescent cigarette smoking continues to move toward the end goal of the tobacco endgame,^8,28,29 for efficiency, allocating future efforts to prioritize populations with known greater disparities is important. Even with the temporal decreases among those in the high-risk category, prevalence in 2018-2019 for past-month (17.0%) and daily (2.7%) smoking approximates starting rates in 2002-2003 among those in the moderate-risk category (past-month, 17.7%; daily, 3.5%) and is well above starting rates in the low-risk category (past-month, 2.7%; daily, 0.3%).

This study has several limitations. First, findings from this cross-sectional observational study cannot support causal inferences. Second, risk factors were selected from variables collected in NSDUH and are not necessarily representative of all potential risk factors. However, the overarching aim of the cumulative-risk model is to quantify the role of cumulative disparities as opposed to identifying which vulnerability is most predictive. Future studies could examine how cumulative-risk group categorization, including analysis in a summative manner,^5-7 affects smoking and other outcomes. Smoking status was determined exclusively by participant self-report and may underestimate smoking prevalence. Responses to questions about past-year behavior may be biased because of time. Finally, this study focused on cigarette smoking; future research should investigate associations with other tobacco and nicotine products and other classes of misused substances.

These results document persistent disparities in adolescent cigarette smoking prevalence among US adolescents with greater psychosocial and health risks, with these disparities widening over time as overall smoking prevalence has decreased. Findings highlight the potential utility of a cumulative-risk model as a robust predictor to better understand smoking disparities in adolescents and can inform prevention and treatment efforts to bring adolescent smoking closer to the tobacco endgame.

Acknowledgement

The funder did not play a role in the design of the study; the collection, analysis, and interpretation of the data; the writing of the manuscript; and the decision to submit the manuscript for publication.

Author contributions

Tyler G. Erath, PhD (Conceptualization; Formal analysis; Investigation; Methodology; Writing—original draft; Writing—review & editing), Fang Fang Chen, PhD (Data curation; Formal analysis; Methodology; Writing—original draft; Writing—review & editing), Michael DeSarno, MS (Methodology; Writing—original draft; Writing—review & editing), Derek Devine, MS (Methodology; Writing—original draft; Writing—review & editing), Adam M. Leventhal, PhD (Writing—original draft; Writing—review & editing), Warren K. Bickel, PhD (Writing—original draft; Writing—review & editing), and Stephen T. Higgins, PhD (Conceptualization; Funding acquisition; Investigation; Methodology; Writing—original draft; Writing—review & editing).

Supplementary material

Supplementary material is available at JNCI: Journal of the National Cancer Institute online.

Funding

This work was supported by the Tobacco Centers of Regulatory Science (TCORS) award U54DA036114 from the National Institute on Drug Abuse and Food and Drug Administration; National Institute of General Medical Sciences Center of Biomedical Research Excellence Award P30GM149331; and National Institute on Drug Abuse Institutional Training Award T32DA007242. The funder did not play a role in the design of the study; the collection, analysis, and interpretation of the data; the writing of the manuscript; and the decision to submit the manuscript for publication.

Conflicts of interest

The authors have no conflicts to disclose.

Data availability

Information regarding the National Survey on Drug Use and Health data is available at https://www.samhsa.gov/data/data-we-collect/nsduh-national-survey-drug-use-and-health.