Intersectionality and Smoking Cessation: Exploring Various Approaches for Understanding Health Inequities

Abstract

Objective

Many marginalized groups smoke at higher rates and have greater difficulty quitting than less marginalized groups. Most research on smoking cessation inequities has focused on a single sociodemographic attribute (eg, race or socioeconomic status), yet individuals possess multiple attributes that may increase risk. The current study used an intersectionality framework to examine how the interplay between multiple marginalized attributes may impact smoking cessation outcomes.

Methods

A diverse sample of 344 adults enrolled in a smoking cessation program and reported on sociodemographic attributes (eg, race/ethnicity, gender, income) and continuous smoking abstinence on their quit date and at 1, 2, and 4 weeks postquit date. A Cox proportional hazard regression model was used to estimate whether intersectional links among race/ethnicity, gender, and income were related to smoking cessation outcomes.

Results

Lower household income may be related to higher risk of smoking cessation failure. There were no significant interactions among race/ethnicity, gender, and income in predicting relapse. Pairwise intersectional group differences suggested some groups may be at higher risk of relapse. Number of marginalized sociodemographic attributes did not predict relapse.

Conclusions

Intersectionality may be a promising framework for addressing health inequities, and may help elucidate how to best design and target intervention efforts for individuals characterized by sociodemographic intersections that concur particularly high risk for poor tobacco cessation outcomes.

Implications

Despite an overall decline in smoking rates, socioeconomic inequities in smoking prevalence and cancer mortality are widening. Efforts targeting tobacco cessation should incorporate new theory to capture the complex set of factors that may account for tobacco cessation inequities (eg, multiple aspects of identity that may influence access to tobacco cessation treatment and exposure to certain stressors that impede cessation efforts). Intersectionality may be a promising framework for addressing health inequities in tobacco use and cessation and may help elucidate how to best design and target intervention efforts for individuals that concur particularly high risk for poor tobacco cessation outcomes.

Introduction

Nearly 40 million adults in the US smoke cigarettes and smoking remains the leading cause of preventable death and disease. Smoking is linked to approximately 20 different types of cancer (eg, liver, esophageal, stomach, lung, kidney, bladder, etc.) and is the cause of almost a third of cancer deaths in the United States. In addition, smoking is associated with other chronic illnesses like respiratory and cardiovascular disease, diabetes, and blindness—touching nearly every organ of the body.¹

Smoking prevalence has declined dramatically over the last 50 years, and over half of all current smokers attempt to quit each year.² However, many historically marginalized populations continue to smoke at higher rates and/or have greater difficulty quitting than less marginalized groups.^3,4 For example, although the smoking prevalence rate among African Americans (16.5%) is similar to that of whites (16.6%),⁵ African Americans display greater difficulty quitting.⁶ Similarly, a recent review suggests that women may have a more difficult time achieving smoking cessation success than do men.⁷ In addition, a well-established body of literature compellingly demonstrates that individuals with lower socioeconomic status (SES) are less successful at quitting. This effect has been shown across multiple indicators of SES such as income, employment status, education level, financial strain, and subjective social status.^8–11

Importantly, most research on smoking cessation inequities has focused on a single marginalized sociodemographic attribute, such as SES or gender. Although attention to broad group differences is vital for understanding population trends (ie, cessation differences based on the level of SES), focus on a single attribute does not allow for examination of heterogeneity within marginalized groups or how multiple aspects of a person’s identity may confer disadvantage. In other words, individuals possess multiple sociodemographic attributes that may place them at both increased and decreased risk for health inequities, such as their race and ethnicity, gender, SES, religion, mental health status, sexual orientation, and more.

An intersectionality framework is useful for understanding how the interplay between multiple marginalized sociodemographic attributes may shape health inequities. Intersectionality suggests it is insufficient to study inequities by a single attribute alone because the effect of one factor (eg, sexual orientation) on health is interconnected with the effect of other factors (eg, race/ethnicity and gender).¹² For example, an African American lesbian woman may experience racism, heterosexism, and sexism, and experiences based on each attribute may have unique independent effects on health.¹³ However, attributes are also interconnected, such that they may substantiate and reinforce one another to influence health. Seminal research examining gendered racism has demonstrated this point by showing that the interaction of oppression faced because of race and sex are related to psychological distress.¹⁴ Intersectionality also emphasizes that sociodemographic attributes interact with social, political, regulatory, and contextual factors to shape privilege, power, and the lived experiences among individuals at certain sociodemographic intersections that may increase the risk for health inequities.^15,16 In this way, the constraints placed upon an individual because of an attribute like race may prevent that individual from accessing privilege associated with another, such as higher SES.¹⁷ For example, highly educated African Americans fare no better on health outcomes such as depressive symptoms, infant mortality, and homicide compared to the least educated whites. This suggests that certain marginalized groups may be less able to access or benefit from privilege normally associated with high levels of education (eg, access to quality health care or safe neighborhoods).^18,19 This work highlights the importance of moving beyond prioritizing one category of social status as the basis for health inequities research (ie, examining differences by SES alone may not reveal inequities within higher SES subgroups).²⁰

An intersectionality framework may be useful for shifting the focus of smoking cessation inequities research from broad sectors of the population to understudied intersectional groups at disproportionate risk for poor outcomes. Several noteworthy studies have examined how other smoking-related outcomes differ by race/ethnicity and gender. For example, one study found main effects for gender on smoking expectancies (women scored higher on negative reinforcement and weight control smoking expectancies than did men) and for ethnicity (Hispanics scored higher on negative reinforcement smoking expectancies than did non-Hispanic African Americans and non-Hispanic whites). There was also a race by gender interaction on weight control smoking expectancies, such that non-Hispanic white women endorsed greater weight control smoking expectancies than did non-Hispanic white men (there were no gender differences among non-Hispanic African Americans nor Hispanics).²¹ Another study found that females were more likely to be light intermittent smokers regardless of race/ethnicity, but that white females (vs. black females and Latinas) experienced a greater increase in light intermittent smokers over time.²² Finally, differences by race and gender have been shown in abstinence-induced negative affect, where females show greater abstinence-induced negative affect than men among non-Hispanic white women but not African American women.²³ To the best of our knowledge, very few studies to-date have examined smoking cessation outcomes with an intersectionality lens. Nollen and colleagues examined factors that may contribute to racial differences in abstinence among black and white smokers. They found that black individuals were less likely to achieve abstinence and this was related to a subset of social determinants indicative of lower SES (eg, lack of home ownership, lower income, neighborhood problems) that were disproportionately represented in black individuals compared to white individuals.²⁴ Taken together, this small body of literature suggests that intersectionality may improve the understanding of the confluence of factors that influence inequities in smoking behavior change.²⁵ It is important to note that while we are using “smoking cessation” as our phenomenon for studying intersectionality, the models and approach apply to studying tobacco cessation, initiation, and maintenance, as well as behavior change more broadly.

Approaches to Studying Intersectionality

Rooted in feminist psychology and critical race theory, intersectionality research has largely relied on qualitative data.²⁶ As such, there is limited guidance as to what analyses may be best to examine intersectionality-related research questions. However, a growing body of research suggests there may be a variety of approaches (eg, additive and multiplicative) that could improve the understanding of intersectional influences on health.^27–30 Additive approaches may include testing the main effects of sociodemographic attributes on health, which would reveal the unique contribution of those attributes to health outcomes. Testing the effect of the number of marginalized statuses on health has been used to show the level of burden associated with possessing multiple marginalized statuses. Multiplicative approaches may include testing statistical interactions between multiple sociodemographic factors to elucidate whether the effect of one sociodemographic attribute (eg, SES) is exacerbated or mitigated by another. In addition, examining differences between individuals at particular intersectional locations within larger groups (eg, higher SES African American men vs. higher income Latino women) may reveal whether there are fundamental differences in health outcomes within subgroups of the population that may not typically be examined.^20,26–28

There is a burgeoning interest in the application of intersectionality in quantitative research seeking to better understand health inequities across biobehavioral, social, psychological, and population health sciences.²⁰ Although there have been advances in this area, there is still a considerable dearth of work exploring approaches to intersectionality-focused research on inequities in smoking cessation outcomes.

Current Study

The purpose of the current study is to contribute to the dialog on how intersectionality may be used in quantitative population health sciences research. We demonstrate this by focusing on how multiple marginalized statuses and their intersections may impact long term smoking cessation outcomes. Given that race/ethnicity, SES, and gender have been established as robust predictors of inequities in smoking-related outcomes, we chose to use these factors as a case example in analyses because their combined effects may reveal differences in smoking cessation outcomes that may not be evident when examining each factor on its own.^{27–29,31,32} It is important to note that the approach used in this study could be applied to other factors useful for studying tobacco cessation inequities within an intersectionality framework (eg, sexual orientation, mental health status, social support, proximity to tobacco outlets, and neighborhood disadvantage).

Methods

Participants and Procedures

The current study used data from a longitudinal cohort study designed to examine racial/ethnic differences in smoking cessation. Participants were recruited from the general population in the Houston, TX area through the use of print media (eg, newspapers), radio and TV advertisements, and community outreach (eg, distributing flyers to primary care practices, local medical societies, health fairs, etc.). Inclusion criteria included age 21 or above, current smoker with a history of at least five cigarettes per day for the past year, motivated to quit within the next 30 days, home address and functioning home telephone number, and could speak, read, and write in English at a sixth grade literacy level. Exclusion criteria included contraindication for nicotine patch use, use of tobacco products other than cigarettes, an active substance use disorder, use of nicotine replacement products other than the patch, participation in a cessation program in the past 90 days, or a household member enrolled currently in the study.

Participants were 424 male and female adult smokers who were followed from 1 week prior to quit date (ie, baseline) through 26 weeks postquit date, with in-person visits occurring at baseline (week −1), on quit date (week 0), and weeks 1, 2, 4, and 26 postquit. At baseline, participants answered standard questionnaires on demographic information, tobacco dependence, and smoking behavior. At each in-person visit, participants answered questions related to smoking behavior, as well as other psychosocial and contextual constructs. The smoking-related questions were used to assess continuous smoking abstinence across the duration of the study. Participants received one $20 gift card at each in-person visit through week 2, and one $40 gift card at week 4 and week 26. The study was approved by the appropriate Institutional Review Board.

Measures and Analyses

Sociodemographic Statuses

Sociodemographic attributes included gender, race/ethnicity (white, African American, or Latino), and annual household income on an 11-point scale from less than $10 000 per year to $100 000 or more per year.

Smoking Behavior and Dependence

Participants were asked, “How many cigarettes per day do you smoke on average?,” and “How soon after you wake up do you smoke your first cigarette?” with response options, 1 = more than 60 minutes; 2 = 31 to 60 minutes; 3 = 6–30 minutes; 4 = 5 minutes or less.

Continuous Abstinence

At each postquit in-person visit, participants were asked one item to measure self-reported continuous abstinence, “Since your quit date, have you smoked even a puff?.” Biochemical verification of abstinence was defined as self-reported complete abstinence from smoking since their quit date and an expired carbon monoxide less than or equal to 6 ppm. Our primary outcome was coded as abstinence = 0 and relapse = 1. We also ran the analyses using an even lower cutoff of 3 ppm, as suggested by some researchers.^33,34 The results remained unchanged from those at the 6 ppm cut off. Results for the 3 ppm cut off are available upon request.

Analytic Plan

All analyses were conducted with SAS Version 9.4 using a Cox proportional hazard regression model to estimate the hazard rate of smoking relapse across the study period. In each step described below, the sociodemographic attributes of race/ethnicity, gender, and income were used as independent variables to estimate whether these were significantly related to relapse. All models were also tested with age, time to first cigarette, and number of cigarettes per day included as covariates, as these constructs may differ by racial/ethnic groups and may influence smoking outcomes.

Main Effects

First, three separate Cox proportional hazard regression models were estimated to test the main effect of race/ethnicity, gender, and income on smoking relapse. In each model, the sociodemographic attribute was entered as the independent variable and smoking relapse was entered as the dependent variable. Next, the main effects of race/ethnicity, gender, and income were entered simultaneously in a single Cox proportional hazard regression model to determine whether any attributes were reliably associated with relapse above and beyond the presence of the others.

Two- and Three-Way Interactions

Two-way interactions between race/ethnicity, gender, and income, and their association with smoking relapse were estimated by entering all possible two-way combinations and their main effects as the independent variables in a single Cox proportional hazard regression model. For the three-way interaction, a model was estimated that contained the three-way, all possible two ways, and main effects as the independent variables and smoking relapse as the dependent variable.

Intersectional Group Differences

To estimate intersectional group differences in smoking relapse, we first constructed a “group” variable by using every possible combination of categorical sociodemographic attributes (eg, white/African American/Latino; female/male; lower income/higher income) to assign individuals to a unique intersectional group. This resulted in 12 unique intersectional groups. For this step, income was dichotomized into higher and lower income groups. Although the entire sample is lower income, indicators of SES such as income have been noted as robust predictors of smoking cessation inequities. Therefore, it was important to determine whether there were intersectional differences based on income. To construct the group variable, income was dichotomized at a value closer to the median to create a higher income group (greater than or equal to $25 000 per year) and lower income groups (less than $25 000 per year). To estimate the effect of intersectional group, the “group” variable was entered as the independent variable in the Cox proportional hazard regression model and smoking relapse as the dependent variable. The model estimates hazard rate comparisons relative to the reference group (0, white males with higher income). However, in order to mitigate concern in intersectionality research to only compare less privileged intersectional groups to a group with a higher social status, we also examined all possible pairwise comparisons to determine whether any combination of intersectional groups were significantly different from one another.³¹

Count of Attributes

To estimate the effect of the number of marginalized attributes with relapse, we constructed a “count” variable by adding up the number of potentially marginalized attributes an individual may possess. For race/ethnicity, African American and Latino were treated as potentially marginalized attributes, as were female gender and lower income. Thus, the count could range from zero to three (race/ethnicity + gender + income). Individuals who did not indicate having any of the marginalized attributes (eg, white males with higher income) were given a value of zero, individuals with one of the attributes (eg, white male with lower income) was given a value of one, and so on. We then estimated a Cox proportional hazard regression model with “count” entered as the independent variable and smoking relapse as the dependent variable. The model estimates the hazard rate comparison of each count group relative to the reference group (0 marginalized attributes). We then examined all pairwise comparisons to determine whether there were any significant differences in relapse among those with different numbers of marginalized statuses.

In order to conduct analyses across each set of constructed intersectional groups as described above, participants with missing data on any of the main sociodemographic attributes were not included in the final sample (N = 49). Additionally, participants who did not attend any of the in-person visits were excluded (N = 31). The final sample included in analyses was N = 344.

Results

Participant Characteristics

Table 1 presents sample characteristics. The final sample (N = 344) consisted of 182 (52.91%) females and 162 (47.09%) males; 115 (33.43%) were African American, 106 (30.81%) were Latino, and 115 (33.43%) were white. The average age was 42.14 years (SD = 10.81) and the average annual income was $34 172 (SD = 26 291); 146 (42.44%) participants had an annual income below $25,000 per year. The sample smoked an average of 21 cigarettes per day and nearly half the sample (47.67%) smoked their first cigarette within 5 minutes of waking. There were 37 participants with zero marginalized attributes (10.76%), 110 with 1 attribute (31.98%), 144 with two attributes (41.86%), and 53 with three attributes (15.41%). Cell sizes in the intersectionality grouping variable ranged from 13 (3.78%) to 41 (11.92%) participants each. Follow-up rates were such that 96.22% (N = 331) of the sample attended the quit day visit; 89.24% (N = 307) attended the week 1 visit; 84.56% (N = 291) attended the week 2 visit; and 87.79% (N = 302) attended the week 4 visit.

Table 1.

Baseline Descriptive Statistics

	Mean (SD)/Frequency (%)
Age	42.1 y (10.81)
Gender (%female)	182 (52.91%)
Race/ethnicity
Non-Hispanic black	115 (33.43%)
Hispanic	106 930.81%)
Non-Hispanic white	115 (33.43%)
Annual income	$34 1715 (26 291)
Less than $25 000 per year	146 (42.44%)
Years of education	12.94 years (1.97)
Time to first cigarette
>60 min	39 (11.34%)
31–60 min	40 (11.63%)
6–30 min	101 929.36%)
5 min or less	164 (47.67%)
Number of cigarettes per day	21.22 (10.33)
Average CO in the sample	24.94 (12.77)

CO = expired-air carbon monoxide.

Cox Proportional Hazards Regression Effects

Main Effects

Separate main effects models indicated there was no significant effect of income, race/ethnicity, nor gender on hazard rate of smoking relapse. When all predictors were entered into the same model as covariates, income was a marginal predictor, such that for each unit increase in income, the hazard rate (ie, risk of failure) decreased by 1.00 % (b = −0.004, SE = 0.002, p = .06, 95% CI [0.991–1.00]; Table 2).

Table 2.

Survival Analysis of Smoking Relapse

	b	SE	p
Individual main effects
Race
African American	−0.15	0.14	.27
Latino	−0.15	0.14	.31
Gender
Female	0.01	0.11	.93
Income	−0.004	0.002	.10
Multiple main effects
Race
African American	−0.20	0.14	.15
Latino	−0.15	0.14	.28
Gender
Female	−0.03	0.12	.80
Income	−0.004	0.002	.06
Two-way interactions
Race
African American	−0.15	0.29	.60
Latino	−0.24	0.30	.42
Gender
Female	−0.11	0.27	.68
Income	−0.01	0.004	.23
Race × Gender
African American × Female	0.10	0.29	.73
Latino × Female	−0.05	0.29	.86
Race × Income
African American × Income	−0.004	0.01	.49
Latino × Income	0.003	0.01	.60
Gender × Income
Female × Income	0.002	0.005	.70
Three-way interaction
Race
African American	−0.09	0.35	.79
Latino	−0.22	0.37	.54
Gender
Female	−0.07	0.33	.84
Income	−0.005	0.005	.36
Race × Gender
African American × Female	−0.02	0.45	.97
Latino × Female	−0.08	0.50	.88
Race × Income
African American × Income	−0.01	0.01	.47
Latino × Income	0.002	0.01	.74
Gender × Income
Female × Income	0.001	0.01	.92
Race × Income × Gender
African American × Income × Female	0.004	0.01	.74
Latino × Income × Female	0.001	0.01	.95
Intersectional group differences (omnibus test)	b	HR	95% CI
White male, higher income (reference)	-	-	-
White male, lower income (1)	0.38	1.47	0.79–2.71
White female, higher income (2)	0.02	1.02	0.62–1.67
White female, lower income (3)	0.23	1.26	0.77–2.08
Black male, higher income (4)	−0.46	0.63	0.34–1.17
Black male, lower income (5)	0.17	1.19	0.72–1.95
Black female, higher income (6)	−0.11	0.90	0.54–1.49
Black female, lower income (7)	0.07	1.07	0.87–1.70
Latino male, higher income (8)	−0.01	1.00	0.62–1.60
Latino male, lower income (9)	0.01	1.01	0.56–1.84
Latino female, higher income (10)	−0.04	0.96	0.59–1.58
Latino female, lower income (11)	−0.19	0.82	0.42–1.63
Post hoc all group comparisons
White male, lower income vs. black male, higher income		2.31	1.12–4.81
White female, lower income vs. black male, higher income		3.73	1.08–3.83
Count of attributes (omnibus test)	b	HR	95% CI
0 marginalized attributes (reference)	-	-	-
1 marginalized attribute	−0.03	0.97	0.65–1.44
2 marginalized attributes	0.05	1.05	0.72–1.54
3 marginalized attributes	0.003	1.00	0.65–1.56

Individual main effects – Predictors were tested in separate models; Multiple main effects – All predictors were tested in one model; Two-way interaction – Each set of two-way interactions was entered in a single model; Three-way interaction – the three-way interaction was tested in a single model; Intersectional group differences – This table presents the omnibus Cox proportional regression results and significant post hoc group contrasts. Count of attributes – Similarly to the intersectional group differences, this table presents the omnibus Cox proportional regression results for count of attributes. There were no significant post hoc group contrasts. HR = hazard ratio, or the risk of relapse in one group compared to another.

Two- and Three-Way Interactions

In testing the two- and three-way interactions between sociodemographic attributes and their association with hazard rate of smoking relapse, no significant interactions were detected (Table 2).

Intersectional Group Differences

The omnibus test for intersectional group differences indicated no overall effect of sociodemographic group on hazard rate of smoking relapse (p = .70). However, individual group contrasts indicated that the hazard rate among white males with lower income is significantly higher than among black males with higher income (HR = 2.31, 95% CI [1.12–4.81]. This suggests the risk of relapse is 2.31 times higher in white males with lower income than black males with higher income. Similarly, the hazard rate among white females with lower income is significantly higher than that of black males with higher income (HR = 3.73, 95% CI [1.08, 3.83]. This suggests the risk of relapse was 3.73 times higher in white females with lower income than black males with higher income. There were no other significant group contrasts. Table 2.

Count of Attributes

Finally, we tested the association of count of marginalized attributes with hazard rate of smoking relapse and found no significant effect (Table 2).

All models above were also tested with age, time to first cigarette, and number of cigarettes per day included in the model as covariates. Results were unchanged, and therefore, not reported here but are available upon request.

Discussion

The current study sought to demonstrate how an intersectionality framework may be useful for investigating the influence of multiple marginalized statuses on risk of smoking relapse during a quit attempt. In addition, we sought to broaden the dialog and empirical evidence of potential intersectional effects in smoking cessation and health behavior change literature more broadly. We showed that lower household income was trending towards higher risk of smoking cessation failure. There were no interactions among marginalized characteristics in predicting relapse. Moreover, there was no indication that a simple count of marginalized statuses predicted higher risk of relapse. Although pairwise comparisons of intersectional group differences suggested that some groups might be at higher risk of relapse, these findings should be considered exploratory as the overall effect of intersectional group was not significant, cell sizes for each of these groups were relatively small, and power for these comparisons was low. In sum, although intersectionality can provide a more nuanced understanding of how certain aspects of identity may exacerbate or mitigate poor health outcomes, our findings did not yield clear, meaningful indications that multiple sociodemographic attributes, nor specific combinations of these increased risk for smoking cessation failure in our sample.

Main effect models showed no association of race/ethnicity or gender with relapse, which is counter to some prior work suggesting lower probabilities of achieving smoking abstinence during a quit attempt among African Americans,⁶ Latinos,³⁵ and women.⁷ Income also had no association with relapse, but the direction of effects suggests that risk of relapse may be higher among those at lower levels of income. This is consistent with a large body of research showing that individuals with lower SES (ie, lower education, income, occupational and insurance status) are less likely to successfully quit smoking compared to higher SES individuals,³ reflecting that SES may be a more robust determinant of cessation than are race/ethnicity or gender.³⁶ Research on mechanisms linking SES to smoking cessation has shown that neighborhood disadvantage (ie, vandalism, litter, traffic, vigilance, social cohesion, and trust), social support, negative affect, stress, and agency mediate the relationship between SES and smoking cessation.⁸ Similarly, a recent study found that low SES smokers encounter more smoking conducive environments that are associated with smoking cessation failure, such as being around other smokers, in contexts where smoking is allowed, or where there is easier cigarette availability.³⁷ Studying these mechanisms may clarify why low SES groups experience more smoking cessation failure than higher SES groups, even when provided with identical treatment (as they were in the current study), and may inform policies and interventions that target the complex interdependence of factors that may impede health behavior change in this group. This may include, but is not limited to, institutional or structural circumstances (eg, smoke-free policies that de-normalize smoking behavior but stigmatize smokers), negative environmental/contextual factors (eg, poverty, violence), decreased availability of intrapersonal and social resources (eg, agency, social support), and stressors (eg, racism, sexism) that create situations where certain populations are more vulnerable to relapse.^8,37,38

Another possible explanation for the null effects of race/ethnicity, gender, as well as the two- and three-way interactions, is that self-reported demographic characteristics are likely to be broad proxies of complex underlying processes that play a more important role in health inequities than the “labels” per se. For example, categories of race/ethnicity are socially constructed labels, but in reality may encompass geographic origins, language, values, cultural norms and traditions, acculturation, ethnic identity, etc.^39,40 Acculturation has been noted as an important predictor of health behaviors including smoking, and a growing body of research suggests that certain facets of acculturation (eg, years living in United States, preferred language) predict greater smoking cessation success.^41,42 Similar hypotheses have been addressed in research on the role of sex and gender in tobacco exposure. In particular, vulnerability to tobacco exposure among women may be better understood by considering the gendered nature of psychological, social, and economic factors. For example, gendered roles and responsibilities may limit power to manage smoke exposure in the home or cause low-income women to spend more time in the home, which may contribute to lower rates of quitting success among low-income females.^43,44 In this way, our broad sociodemographic measures may not have been able to discern important intersectional, between-person differences in the underlying processes associated with sociodemographic attributes.

In addition to examining main effects and interactions, we also classified individuals into intersectional groups based on their race/ethnicity, gender, and income status and examined differences in relapse. This reflects the perspective that individuals are situated at distinct sociodemographic crossroads that may influence their health behaviors and experiences. These analyses allowed for comparisons across groups that may normally be understudied in research examining group differences based on a single sociodemographic factor (eg, income). In other words, each group is treated as unique and relationships between multiple factors are not assumed to be linear as they are in two- and three-way interactions.^27,28 Given the small cell sizes of intersectional groups and the lack of an overall significant “intersectional group” effect, our results should be interpreted with caution. However, our exploratory results suggest that there could be utility in this approach for clarifying cessation inequities. For example, Figure 1 shows intersectional groups with statistically significant differences in hazard rate of relapse. Groups that are significantly different from one another are notated with the same symbol in both the figure and legend. Although we recognize the limitations of making multiple group comparisons given our small sample size, an important facet of intersectionality is the exploration of within-group diversity (eg, within higher income group, looking at nuanced differences by race/ethnicity and gender) and movement away from using a majority group as a “normative control.” ²⁸ As such, we feel these comparisons may be an important first step with respect to studying heterogeneity in health behaviors and disease risk within marginalized groups.

Figure 1.

Abstinence rate by gender × race × income intersectionalities.

Finally, we compared hazard rates of relapse between individuals with zero, one, two, or three marginalized attributes. This approach mirrors work suggesting that possessing a greater number of marginalized attributes may confer greater burden on health,¹⁸ as well as a recent study showing that individuals who endorsed mistreatment due to a greater number of marginalized attributes were more likely to experience mental health morbidity and post-traumatic stress disorder symptoms than those who attributed mistreatment to a lesser number of attributes.³² There were no statistically significant differences in relapse based on number of marginalized statuses. We recognize the limitations of this method, such that counting statuses may be overly simplistic in that it treats all statuses the same (ie, all “twos” are created equal no matter which attributes it was comprised of). However, given that prior work using rudimentary measures of count of marginalized statuses and attributions does suggest greater health risk,^18,32 future research should explore this technique as a way to elucidate level of burden associated with possessing multiple marginalized statuses.⁴⁵

Limitations and Future Directions

Traditional intersectionality research has largely utilized qualitative data, which is vital for providing a comprehensive understanding of how sociodemographic factors interact with social processes to shape experiences and health outcomes. As such, there may not be one best quantitative approach to capturing the rich subjective nature of human experience.²⁸ However, given the growing interest in using intersectionality in health research, the current study demonstrates that there are various methods that could improve the understanding of how multiple factors relate to health outcomes. The current study represents an important step in broadening the dialogue about how an intersectionality framework can help tackle health inequities in tobacco cessation.

The current study utilized data collected across four in-person laboratory visits. Although this allowed us to examine cumulative hazard rate of smoking relapse, we were unable to make conclusions about how interpersonal stressors in daily life (eg, data on discrimination were not collected at each visit) may influence smoking relapse. More intensive longitudinal data collection methods (eg, ecologically momentary assessment [EMA]) could better provide information on the dynamic nature of momentary or daily experiences related to intersectional identities and their relation to smoking behavior. EMA data would also improve the statistical power necessary for detecting important variation due to between-person processes (eg, determining which intersectional group may be more at risk), as well as variation due to within-person processes (eg, determining when a particular intersectional group may be more at risk due to dynamic contextual factors). Future studies of intersectional processes in daily life would better capture the within-person, dynamic nature of how and when the saliency of various facets of identity may change depending on time and context (eg, depending on whether interpersonal experiences of discrimination occurred, or how one’s gender identity may change depending on the characteristics of social settings).¹⁶

We tested a number of models in the current study, which increases the risk of Type I error. Given there is very limited empirical work using an intersectional framework to understand how multiple sociodemographic attributes may relate to smoking behavior, we did not use any correction techniques. However, given the exploratory nature of using intersectionality to better identify at-risk groups, we believe this approach unwarranted. Future work should also incorporate measures to assess constructs that may contribute to inequities that may not be captured by broad labels of race/ethnicity, gender, or SES (eg, cultural norms, sexual orientation, ethnic identity, and indicators of SES like tangible assets or prestige).^39,46–49

Conclusion

Although there are well-established inequities in tobacco cessation outcomes based on broad sociodemographic attributes like race/ethnicity, gender, and income, there is still a significant gap in our understanding of how to reduce inequities. In fact, despite an overall decline in smoking rates, socioeconomic inequities in smoking prevalence and cancer mortality are widening, especially in preventable cancers like lung and colorectal cancer.^47,50 This suggests efforts targeting tobacco cessation should incorporate new theory and approaches to capture the complex set of factors that may account for tobacco cessation inequities (eg, multiple aspects of identity, their interaction with each other, and their interaction with social and contextual factors that influence access to tobacco cessation treatment and exposure to stressors that impede health behavior change).²⁵ In conclusion, intersectionality may be a promising framework for addressing health inequities in tobacco use and cessation and may elucidate how to better design and target interventions for individuals characterized by sociodemographic intersections that increase the risk for poor tobacco cessation outcomes.

Supplementary Material

A Contributorship Form detailing each author’s specific involvement with this content, as well as any supplementary data, are available online at https://academic.oup.com/ntr.

Funding

This work was supported by the National Cancer Institute of the National Institutes of Health (grant numbers R01DA014818, P30CA042014); and National Center for Advancing Translational Science (grant number 5TL1TR002540 to LNP). Additional support was provided by the Huntsman Cancer Foundation. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health or the Huntsman Cancer Foundation.

Declaration of Interests

We have no conflicts of interest to disclose. The authors have full control of all primary data and agree to allow the journal to review the data if requested. This manuscript has not been published and is not under consideration for publication elsewhere, nor have these data been previously reported.