Patterns of Birth Cohort-Specific Smoking Histories by Race and Ethnicity in the U.S.

Abstract

Introduction:

U.S. smoking prevalence varies greatly by race/ethnicity. However, little is known about how smoking initiation, cessation, and intensity varies by birth cohort and race/ethnicity.

Methods:

Adult smoking data were obtained from the 1978‒2018 National Health Interview Surveys. Age-period-cohort models with constrained natural splines were developed to estimate historical smoking patterns among non-Hispanic White (NHW), non-Hispanic Black (NHB), Hispanic, non-Hispanic Asian and Pacific Islander (API), and non-Hispanic American Indian and Alaskan Native (AIAN) individuals. Annual smoking prevalence and probabilities of smoking initiation, cessation, and intensity by age, year, gender, and race/ethnicity were estimated for the 1900 to 2000 birth cohorts. Analysis was conducted in 2020‒2021.

Results:

Smoking initiation probabilities were highest for the AIAN population, second highest among the NHW population, and lowest among API and Hispanic populations across birth cohorts. Historically, initiation probabilities among NHB populations were comparable to that of NHW populations, but have decreased since the 1970 birth cohort. Cessation probabilities were lowest among AIAN and NHB populations and highest among NHW and API populations across cohorts and ages. Initiation and cessation probabilities produce observed patterns of smoking where prevalence among AIAN populations is highest across all ages and cohorts. Across cohorts, smoking prevalence among NHB populations, particularly males, is lower than among NHW populations at younger ages, but higher at older ages.

Conclusions:

There are important and persistent racial/ethnic differences in smoking prevalence, initiation, cessation, and intensity across U.S. birth cohorts. Targeted interventions should address widening smoking disparities by race/ethnicity, particularly for AIAN and NHB populations.

INTRODUCTION

Smoking prevalence is decreasing across all sociodemographic groups in the U.S., but the prevalence and the magnitude of change varies greatly by race/ethnicity and gender.^1,2 American Indian and Alaskan Native populations have historically had the highest smoking prevalence among racial/ethnic groups in the U.S., with an estimated prevalence of 25.2% (95% CI=14.4%, 35.9%) in 2018.³ In contrast, Asian and Hispanic populations have the lowest levels of smoking in the U.S., with a reported prevalence in 2018 of 8.2% (95% CI=6.3%, 10.0%) and 12.3% (95% CI=10.8%, 13.8%), respectively. In between these extremes lie non-Hispanic Black and non-Hispanic White populations. Historically, these populations have had similar prevalence,¹ with some variation by gender.⁴ In 2018, smoking prevalence was 17.9% (95% CI=17.1%, 18.6%) in non-Hispanic White adults and 18.2% (95% CI=16.3%, 20.1%) in non-Hispanic Black adults.³

While trends in overall smoking prevalence by year are well characterized, less is known about how smoking prevalence or the rates of smoking initiation, cessation, and intensity (cigarettes smoked per day) vary by birth cohort across different racial and ethnic groups. Previous studies have shown that age-specific smoking patterns for some U.S. populations vary dramatically by birth cohort.^1,5‒12 In particular, Holford et al. estimated historical birth cohort patterns in smoking prevalence, initiation, cessation, and intensity for the U.S. population by gender. They found that smoking patterns vary largely by cohort with important differences by gender; males had consistent declines in smoking prevalence and initiation since the 1920 birth-cohort, whereas female smoking prevalence and initiation increased considerably through the 1940 birth-cohort.⁷ In a subsequent study of U.S. White versus Black populations, not excluding Hispanic populations, Holford et al.⁴ found that initiation probabilities exhibits higher and earlier age-peaks in White than Black populations, and that cessation rates are higher in the White than in the Black population across ages. These patterns translate into higher prevalence in the White population at younger ages, but higher prevalence in the Black population at older ages.⁴ These differences, together with the lower rates of smoking intensity among Black adults, translate into important differences in pack-years and duration, which in turn affect their rates of smoking-related diseases, such as lung cancer.^13‒17 Earlier studies explored birth cohort patterns in smoking prevalence and initiation among the Hispanic population^8,9,18 and among White, Black, and Asian/Pacific Islander populations.¹⁹ However, no studies have examined and simultaneously compared cohort and age patterns of smoking prevalence, initiation, cessation, and intensity for multiple racial and ethnic groups.

Using data from the National Health Interview Survey, the age-period-cohort approach developed by Holford et al.⁷ was used to estimate smoking prevalence and annual probabilities of smoking initiation, cessation, and intensity by age, birth cohort, year, and gender for 5 U.S. racial/ethnic groups: non-Hispanic White (NHW), non-Hispanic Black (NHB), Hispanic, non-Hispanic Asian and Pacific Islander (API), and non-Hispanic American Indian and Alaskan Native (AIAN). These estimates could characterize differences in smoking outcomes for public health surveillance purposes, and also serve as input parameters for computational modeling analyses of smoking and the impact of prevention and control interventions for specific racial and ethnic groups.^6,20‒30

METHODS

Study Sample & Measures

Publicly available de-identified data were obtained from the 1978–2018 National Health Interview Survey (NHIS). The NHIS is an annual nationally-representative cross-sectional household survey of the non-institutionalized population of U.S. adults ages ≥18 years.³¹ An established model for smoking was used, where individuals who have never smoked may transition to current smoking, and then may quit to former smoking.^4,6,7,25,32 Each survey provided data on current, former, or never smoking status, which was used for cross-sectional estimates of ever smoking prevalence by age and survey year. A subset of surveys collected age at smoking initiation (1978–1980, 1987,1988, 1992, 1995, 1997–2018), age at cessation (1978–1980, 1983, 1985, 1990, 1992, 1994, 1995, 1997–2018), and cigarettes per day smoked as an indicator of smoking intensity (1983, 1985, 1988,1991–1995, 1997–2018), which were used to retrospectively reconstruct smoking histories until the date of survey. More details about tobacco use definitions and the underlying model are provided in the Appendix.

Prior to 1978, the NHIS coded respondent race based on the interviewer’s observation (White, Black, Other), while national origin was asked in a single question starting in 1976. Because of the inadequacy of these data, in 1978 the NHIS began asking respondents directly about race and Hispanic origin as 2 separate questions in annual surveys.³³ Therefore, 1978‒2018 NHIS surveys were used, since during this period data were collected consistently for the racial/ethnic groups included in the analysis. Five racial/ethnic groups were considered: NHW, NHB, Hispanic, API, and AIAN. Individuals reporting multiple races were not included in the analysis. For brevity, the descriptor “non-Hispanic” was omitted in language describing API and AIAN groups. Data from the 2019 and 2020 NHIS were excluded because of survey design changes that became effective in 2019.³⁴

Statistical Analysis

For each racial/ethnic group, the approach developed by Holford et al.⁷ was independently applied to estimate the following age-specific (single ages 0‒99 years) smoking outcomes by birth cohort (single birth-year 1900 to 2000 cohorts), calendar year and gender: smoking initiation and cessation probabilities; prevalence of ever, never, current and former smoking; and distribution of cigarettes per day (CPD) smoked for current smokers. The following categories of CPD smoked were considered (approximate mean CPD within a category): CPD≤5 (3); 5< CPD≤15 (10); 15<CPD≤25 (20); 25<CPD≤35 (30); 35<CPD≤45 (40); and 45<CPD (60). This framework has been previously applied to obtain estimates of smoking histories for the U.S. population by gender,^6,7 for U.S. Black and White populations,⁴ and for the Canadian population by gender.⁵

Briefly, age-period-cohort weighted-logistic regression models (PROC SURVEYLOGISTIC in SAS 9.4) were used to estimate each of the smoking parameters. Temporal effects (age, period and cohort) were modeled using constrained natural splines. To resolve the well-known identifiability problem of distinguishing age, period, and cohort, (1) the period slope was constrained to be zero for estimating initiation probabilities, (2) the cohort slope was constrained to be zero for cessation probabilities, and (3) an age-cohort model was used for estimating ever-smoker prevalence. To address recall bias in retrospective self-reported age at smoking initiation, and to enforce consistency of the estimates across smoking outcomes, the initiation probabilities by cohort were calibrated to produce predictions of ever smoking prevalence consistent with estimated ever smoker prevalence at age 30 years. Information about the specific models used for each sociodemographic/smoking parameter combination, including the number and location of spline knots and other details, are presented in the Appendix.

RESULTS

Figure 1 shows age-specific smoking initiation probabilities for each racial/ethnic group and gender combination for selected birth cohorts (1910, 1930, 1950, 1970, 1990). Figures showing patterns for additional cohorts by calendar year are available in the Appendix (Appendix Figure 1). For all racial/ethnic groups, smoking initiation probabilities increased with age during adolescence, peaked in the late teens and then declined (Figure 1). Among females, initiation probabilities increased from the 1910 to the 1950 birth cohort, except for the API population, and have since decreased for more recent birth cohorts. Among males, initiation probabilities increased from the 1910 to the 1930 birth cohort, except for the Hispanic population, but have likewise declined by cohort. In terms of race/ethnicity, age-specific smoking initiation probabilities were generally highest among the AIAN population, followed by the NHW population and lowest among Hispanic and API populations. Initiation probabilities were higher among NHB than Hispanic and API populations in earlier birth cohorts, but the differences were comparable across the 3 groups in recent birth cohorts. All groups had decreasing initiation by cohort, except for the female AIAN population, which show increased initiation probabilities in the 1990 birth cohort, the most recent cohort displayed.

Figure 1.

Age-specific smoking initiation probabilities for selected birth cohorts by race/ethnicity and gender (females – top panels, males – bottom panels).

Notes: Lines represent the initiation probabilities for Non-Hispanic White (NHW, red), Non-Hispanic Black (NHB, orange), Hispanic (sky blue), Asian and Pacific Islander (API, blue), and American Indian and Alaskan Native (AIAN, black) populations. An interactive version of this figure’s data can be found at: https://apps.cisnetsmokingparameters.org/race/.

Figure 2 shows age-specific smoking cessation probabilities for each racial/ethnic group and gender combination for selected birth cohorts (1910, 1930, 1950, 1970, 1990). Figures showing patterns for additional cohorts by calendar year are available in the Appendix (Appendix Figure 2). Smoking cessation probabilities generally increased with age and birth cohort (Figure 2). In terms of race/ethnicity, cessation probabilities were lowest among the NHB population, and highest among NHW, Hispanic and API populations. In general, while cessation rates increased by cohort across racial/ethnic groups, the differences in cessation probabilities between the groups with the highest and lowest rates widened in recent cohorts, particularly at younger ages. Of note, cessation probabilities among all other race/ethnic groups have increased relative to those among the NHB population.

Figure 2.

Age-specific smoking cessation probabilities for selected birth cohorts by race/ethnicity and gender (females – top panels, males – bottom panels).

Notes: Lines represent the cessation probabilities for Non-Hispanic White (NHW, red), Non-Hispanic Black (NHB, orange), Hispanic (sky blue), Asian and Pacific Islander (API, blue), and American Indian and Alaskan Native (AIAN, black) populations. An interactive version of this figure’s data can be found at: https://apps.cisnetsmokingparameters.org/race/.

Current smoking prevalence is derived from the aforementioned smoking initiation and cessation probabilities, yielding patterns of smoking prevalence by age, birth cohort, year, gender, and race/ethnicity. Figure 3 shows age-specific current smoking prevalence for each racial/ethnic group and gender combination for selected birth cohorts (1910, 1930, 1950, 1970, 1990). Figures showing patterns for additional cohorts by calendar year are shown in Appendix Figure 3. Among males, current smoking prevalence decreased by birth cohort since the birth cohort of 1930, but, among females, prevalence increased until the 1950 birth cohort and then decreased. In terms of race/ethnicity, current smoking prevalence was generally highest across ages and cohorts among the AIAN population and lowest among Hispanic and API populations. The relatively high smoking prevalence among the AIAN population was driven by higher initiation probabilities and relatively low cessation probabilities. Smoking prevalence among the NHB population, particularly in males, was lower at younger ages, but higher at older ages than that of the NHW population. These differences result from the lower initiation and cessation probabilities for the NHB population compared to the NHW population.

Figure 3.

Age-specific smoking prevalence (percentage) for selected birth cohorts by race/ethnicity and gender (females – top panels, males – bottom panels).

Notes: Lines represent the smoking prevalence for Non-Hispanic White (NHW, red), Non-Hispanic Black (NHB, orange), Hispanic (sky blue), Asian and Pacific Islander (API, blue), and American Indian and Alaskan Native (AIAN, black) populations. An interactive version of this figure’s data can be found at: https://apps.cisnetsmokingparameters.org/race/.

Figure 4 shows age-specific mean cigarettes per day consumed by smokers for each racial/ethnic group and gender combination for selected birth cohorts (1910, 1930, 1950, 1970, 1990). Figures showing patterns for additional cohorts by calendar year are shown in Appendix Figure 4. Age-specific mean CPD among current smokers decreased since the 1930 birth cohort. In terms of age, for older birth cohorts, the mean CPD increased until about age 50, and then decreased. However, for recent birth cohorts, the mean CPD has been relatively constant or decreasing with age with no clear peak. In terms of race/ethnicity, the NHW population had the highest mean CPD for both males and females followed by the AIAN and the NHB populations. The Hispanic and API populations had the lowest mean CPD across birth-cohorts.

Figure 4.

Age-specific mean cigarettes per day among current smokers for selected birth cohorts by race/ethnicity and gender (females – top panels, males – bottom panels).

Notes: Lines represent the mean cigarettes per day for Non-Hispanic White (NHW, red), Non-Hispanic Black (NHB, orange), Hispanic (sky blue), Asians and Pacific Islander (API, blue), and American Indian and Alaskan Native (AIAN, black) populations. An interactive version of this figure’s data can be found at: https://apps.cisnetsmokingparameters.org/race/.

Appendix Figure 5 shows mean cumulative years of smoking (duration) and Appendix Figure 6 shows mean cumulative pack years for selected birth cohorts by race/ethnicity and gender. Across all cohorts, the AIAN female population had the longest smoking duration, followed by the NHW and NHB populations. Among males, smoking duration was similar in the earlier cohorts across groups, except for API which had lower smoking duration, however, in later cohorts, AIAN had the longest smoking durations. The NHW and AIAN populations both had the highest levels of pack years, consistent with their higher mean CPD and duration.

DISCUSSION

Age- and cohort-specific smoking prevalence and annual probabilities of smoking initiation, cessation, and intensity by gender were estimated for 5 U.S. racial/ethnic groups using historical cross-sectional, nationally representative data. This is the first study to estimate cohort-specific smoking patterns for racial/ethnic subgroups, including the API and AIAN populations. The age-period-cohort modeling approach by Holford et al. was applied,⁷ which was previously used for analyses of the U.S. population,^6,7 the U.S. White and Black populations,⁴ and the Canadian population.⁵ The results show important variations in smoking initiation and cessation probabilities by race/ethnicity that shape differences in age-specific smoking prevalence by calendar year and birth cohort.

This study results indicate large reductions in initiation and increases in cessation among recent cohorts for all racial/ethnic groups. However, disparities in all smoking indicators across racial/ethnic groups remain, and in some cases are even widening by birth cohort. Consequently, racial/ethnic disparities in smoking will likely persist as recent cohorts age, unless efforts are made to further reduce smoking in high prevalence groups. Across cohorts, the AIAN population consistently had the highest smoking prevalence and initiation probabilities, duration of smoking and pack years, and the lowest cessation probabilities. Relatively low cessation probabilities among the NHB population have led to higher prevalence at older ages than the NHW population, especially for males. NHW and AIAN individuals who currently smoke had higher smoking intensities (cigarettes per day), which persisted across birth cohorts. These findings highlight the need for targeted tobacco control interventions for AIAN and NHB populations, such as developing culturally appropriate cessation programs for AIAN populations³⁵ or banning of menthol cigarettes and flavored cigars, which have been used to target the NHB population by the tobacco industry.^1,36,37

Much of the improvement in observed smoking indicators across cohorts reflect the impact of tobacco control policy progress since the 1960s.^1,27 However the health gains have not been uniformly distributed across sociodemographic subgroups. In particular, the persistently higher smoking initiation, prevalence and intensity and lower smoking cessation in the AIAN population can be explained by a variety of factors, including targeting by the tobacco industry, lack of strong tobacco control policies and low tobacco taxes in tribal lands, the use of commercial tobacco for traditional practices, and the relatively disadvantaged state of the AIAN population with lower socioeconomic status and higher rates of substance use, which correlate with higher smoking prevalence, compared with other racial/ethnic groups.^38‒41 These factors result in a high burden of tobacco-related morbidity and mortality.^1,40,42,43 The study findings reinforce calls for targeted strategies to reduce smoking in the AIAN population. It highlights the need for continuous funding and support for tribally-driven efforts to reduce commercial tobacco use through culturally appropriate cessation programs, smoke-free policies, higher tobacco prices in tribal lands, and educational media campaigns for AIAN youth.^{35,38,41,44‒47} These efforts should distinguish commercial tobacco from traditional tobacco which has cultural and spiritual importance to AIAN communities.⁴⁸

Many studies have examined patterns and trends in smoking prevalence, initiation and cessation by race/ethnicity,^{1,2,39,49‒51} but few have focused on variations by cohort in age-specific smoking behaviors. In a previous study, smoking patterns were compared between Black and White cohorts in the U.S. using NHIS data from 1965 to 2012.⁴ This study updates these estimates by including data through 2018 and disaggregating the Hispanic population as a separate group given their known differences in smoking patterns with NHW and NHB populations.^1,2

This study also updates smoking cohort patterns for the Hispanic population. Escobedo et al. previously analyzed birth cohort patterns in smoking prevalence among Mexican‒Americans, Cuban‒Americans and Puerto Rican‒Americans using data from the 1982‒1983 Hispanic Health and Nutrition Examination Survey (HHANES).⁸ They found that smoking prevalence decreased by birth cohort in males for all Hispanic groups, but that prevalence increased among successive cohorts of Cuban‒American and Puerto‒Rican females. This study updated these analyses and found that smoking prevalence is now also decreasing by cohort among Hispanic females. However, specific Hispanic populations were not analyzed independently. In a subsequent study, Escobedo et al. evaluated trends in smoking prevalence at ages 20‒24 years as a proxy for smoking initiation.⁹ They found differences in initiation rates between Hispanic subgroups, with some having higher rates than non-Hispanic White individuals in earlier cohorts. However, the rates of initiation generally either declined or leveled off for more recent Hispanic cohorts relative to the White population. This updated analysis found that initiation and prevalence remain lower for the Hispanic than the NHW population in recent cohorts.

A strength of this study is the use of historical NHIS data, a nationally representative survey of the U.S. civilian, non-institutionalized population providing data by race and ethnicity consistently since 1978. The timespan and resolution of the NHIS enabled us to produce detailed age, gender, and cohort-specific smoking patterns for different racial/ethnic groups, including API and AIAN populations that are often missing or aggregated in other analyses due to small sample sizes. Another strength is the use of a methodology that incorporates cohort as well as age and period trends in smoking initiation and cessation probabilities. These probabilities are estimated jointly with ever, never, current and former smoking prevalence, thereby ensuring that the resulting estimates are consistent across smoking outcomes. This critical feature guarantees that the estimates can be readily used to parameterize simulation models of smoking trends and generate smoking patterns representative of the studied populations.

The results highlight the importance of cohort variations when comparing trends in smoking behaviors between different sociodemographic groups. Cohort-based analyses are needed to understand the consequences of the varying tobacco market, tobacco control policy, cessation treatment environments and other social and environmental factors experienced by different generations. Ignoring cohort patterns could result in misinterpretation of trends and in erroneous projections of future smoking. However, analyses like this study require a large amount of data from multiple nationally representative surveys covering many years. The long-term surveillance of tobacco use behaviors across different sociodemographic groups requires continued support for multiple national health surveys, improving and maintaining survey response rates, and limiting changes in survey design and structure to ensure consistent tobacco product measures over time. While this study did not examine social determinants of health, or other structural factors that could influence smoking in the groups analyzed, its findings and parameter estimates can inform future work investigating the causal effects of these contextual factors.

Simulation models of tobacco use have become important tools to assess the impact of past and potential tobacco control policies and regulations.^{6,20,22‒25,27‒30,32,65‒67} However, the failure of most available models to consider race/ethnicity is a major gap, since policies have differential effects by race/ethnicity which may result in widening health disparities.^68‒70 A major challenge to the development of simulation models by sociodemographic group is the lack of data to inform these models, particularly group-specific estimates of smoking initiation and cessation, smoking intensity, absolute and relative risks of smoking-related morbidity and mortality, and policy effect sizes. This study estimates aim to address some of these gaps, thereby facilitating the development of simulation models by racial/ethnic status. Preliminary versions of the estimated smoking initiation and cessation probabilities have already been incorporated in a model of smoking and health outcomes from menthol cigarettes among Black populations in the U.S.²⁸

The observed differences in smoking intensity and duration by race/ethnicity have implications for the burden of tobacco-related diseases across groups; the risk of lung cancer, other tobacco-related cancers and chronic obstructive pulmonary disease (COPD) are known to increase with higher cumulative exposures.^1,14,71‒73 The relatively high smoking duration among NHW, NHB and AIAN populations, and high pack-years among NHW and AIAN populations across cohorts at least partially explain their correspondingly high rates of smoking-related health conditions. Differences in smoking prevalence and intensity by sociodemographic groups also have implications for lung cancer screening eligibility.^13,74,75 The U.S. Preventive Services Task Force currently recommends screening adults aged 50‒80 years with at least 20 pack years of exposure and no more than 15-years since quit.⁷⁴ Major differences in smoking intensity and duration by race/ethnicity suggest that current racial/ethnic disparities in eligibility for screening^13,74,75 will continue as newer cohorts reach the eligibility age.

Limitations

Limitations of the study include the use of self-reported data, which is subject to underreporting, particularly in more recent years when smoking became less socially accepted. Similarly, NHIS response rates have decreased in recent years,^52,53 which could result in higher non-response bias among recent cohorts. The analysis also relies on self-reported retrospective data on the age of smoking initiation and cessation, which are subject to recall bias. However, this data is essential for analyzing earlier birth cohorts and is critical to understand long-term use trends. The calibration process aimed to address this limitation (Appendix).

This study did not consider further variations within racial/ethnic subgroups. There are important differences in smoking patterns within Hispanic populations. In particular, Cuban-Americans and Puerto Ricans have higher smoking prevalence compared with Mexican Americans.^1,2,8,54 There is considerable heterogeneity in Asian American populations as well. For example, Filipino Americans have considerably higher smoking prevalence than Asian Indians and Chinese Americans.⁵⁵ Factors such as country of origin, level of acculturation, duration of U.S. residence and other socioeconomic factors, as well as temporal changes in the composition of the population due to in- and out- migration, translate into important variations across race/ethnicity groups.^2,56,57 There are also important differences in rates of commercial tobacco use and in ceremonial traditional tobacco practices across different AIAN subpopulations.⁵⁸ Future study is warranted to disaggregate Asian, Hispanic, and AIAN groups. Individuals from other racial groups or who reported multiple races were not included in the analysis, however these groups also constitute a small fraction of the whole sample. While small sample sizes create challenges, leveraging data from multiple nationally representative surveys may enable analyses for smaller sub-populations.⁵⁹

Last, while comprehensive analysis of cohort patterns of cigarette smoking by race and ethnicity was conducted, these analyses did not consider other relevant nicotine delivery products, such as the use of e-cigarettes, cigars and smokeless tobacco, which also vary by race and ethnicity and may affect cigarette smoking trends due to substitution effects or dual use.^1,2,60‒62 Also, menthol and non-menthol cigarette smoking were not distinguished in the analysis. Menthol smoking is known to be an important determinant of smoking initiation and cessation, with high rates of use among NHB populations.^36,63,64

CONCLUSIONS

This study of smoking patterns by race/ethnicity in the U.S. identifies important differences not only in prevalence, but also in cohort- and age-specific initiation and cessation probabilities, and smoking intensity. These differences merit strong consideration in developing comprehensive tobacco control strategies, especially those that aim to address tobacco-related health disparities. Among the groups analyzed, the AIAN population remains the group with the highest smoking prevalence and initiation and the lowest cessation probabilities, and should therefore be prioritized for targeted policy intervention. Similarly, the NHB population’s lower cessation rates translate into longer durations and higher prevalence at older ages, emphasizing the need for policies that promote their smoking cessation.