Estimating Causal and Time-Varying Effects of Maternal Smoking on Youth Smoking

Abstract

Introduction:

Maternal smoking is a well-known risk factor for youth smoking, yet whether this relationship is causal remains unresolved. This study utilizes propensity score methods for causal inference to robustly account for shared risk factors between maternal and offspring smoking.

Methods:

An 8-year longitudinal cohort of 900 adolescents in the Chicago area were followed starting from approximately age 15.6. The effects of maternal lifetime smoking (MLS) and prenatal tobacco exposure (PTE) among participants reporting MLS on offspring’s past 30-day smoking, daily smoking status and smoking frequency were examined by logistic regression and Poisson regression after nearest-neighbor propensity matching. Age dependency of this relationship was then examined across the age range of 15–25 using time-varying effect modeling.

Results:

Propensity matching yielded 438 and 132 pairs for MLS and PTE study samples, respectively. MLS demonstrated significant associations with past 30-day smoking (RR 1.09; 95% CI 1.04–1.14), daily smoking (RR 1.08; 95% CI 1.05–1.12), and smoking frequency of offspring (RR 1.32; 95% CI 1.15–1.52), with stable effects across age. Among participants reporting MLS, having PTE showed significant additional effects on daily smoking (RR 1.09; 95% CI 1.02–1.17) and age-dependency that showed significance during young adulthood but not adolescence.

Conclusion:

The relationship between maternal and offspring smoking was not fully accounted for by shared risk factors, suggesting a causal mechanism with PTE having a delayed effect across age. Targeted prevention efforts should be made on maternal smoking-exposed adolescents to mitigate their risks of developing heavy smoking habits in adulthood.

INTRODUCTION

Smoking remains the leading cause of preventable morbidity and mortality in the United States.¹ While past 30-day established (≥100 cigarettes/lifetime) smoking prevalence has decreased significantly from 41.9% in 1965 to 14.1% in 2017 with extensive tobacco control efforts, the current smoking rate has plateaued around 15%² and is estimated to remain steady if current tobacco control policies remain unchanged.³ As a further reduction of smoking prevalence becomes more challenging, elucidating mechanisms by which youth initiate smoking and progress to chronic smoking becomes more important.

Parental smoking is one of the most well-documented risk factors for youth smoking initiation,⁴ progression to daily/heavy smoking,^5,6 and nicotine dependence.^5,6 Interestingly, maternal (vs paternal) smoking seems to be more strongly associated with offspring smoking.^6,7 Mothers who are smokers play a vital role in adolescents’ smoking behaviors through numerous environmental^8–11 and genetic mechanisms,^12–14 as well as their interplay.^13,15 However, stronger maternal influences may be explained, at least in part, by prenatal tobacco exposure (PTE) from maternal smoking during pregnancy.

PTE is a special case of maternal lifetime smoking (MLS) as tobacco toxicants and nicotine are directly transferred via the placenta to the fetus,¹⁶ compared to the indirect exposure from postnatal secondhand smoking. However, it remains unknown whether PTE poses an additional risk (on top of MLS’s effect) for offspring smoking. Some studies have found a significant, independent effect of PTE on offspring’s smoking frequency, quantity and nicotine dependence even after adjusting for maternal postnatal smoking.^5,7,17 Furthermore, studies examining PTE in relation to frequently co-occurring disruptive behavior outcomes show that combining within- family designs with high-quality exposure measurement yields a modest causal effect of PTE.¹⁸ However, some studies have failed to support an independent effect of PTE, concluding that mother’s prenatal smoking merely represents common environmental/genetic factors shared between mother and offspring.^19–22

Taken together, the existing literature remains ambiguous as to whether MLS and especially PTE introduce a causal effect on offspring smoking outcomes. This relationship may be attributable to characteristics of mothers who smoke, such as confounding by socioeconomic indicators.^23–25 Given this severe confounding, comparisons between smoking and non-smoking families are inappropriate due to large differences along with these covariates (‘imbalance’). In other words, traditional methods of risk adjustment (e.g. regression control) may be insufficient and leave residual bias in the overall findings.²⁶ Propensity score methods more effectively ‘balance’ covariates between treated (i.e., mothers with smoking history) and control (i.e., never-smoker mothers) units.²⁷ This process makes the comparison more appropriate (by mimicking random assignment to treatment and exposure groups) and allows the estimation of causal effects.

The intergenerational transmission of smoking is a function of both genetics and environmental effects. Studies have shown that the contribution of each domain changes significantly over time (i.e., the proportion of genetic influence increases while that of environmental effects decreases in older age);^28,29 therefore, the effect of maternal smoking may also demonstrate a significant variation across the age of offspring (i.e. significant in some age groups but undetectable in others), potentially explaining these divergent results in the literature. That is, previous analyses that used age as a covariate or a categorical moderator may have failed to detect the age-dependent effect of maternal smoking as a result of aggregation.

This study utilizes advanced statistical methods to account for above-mentioned limitations of the previous research. Specifically, propensity score methods can reduce bias that originates from the differences in baseline characteristics and provide a promising chance to isolate causal mechanisms in observational studies²⁶ as has been shown in developmental studies of PTE.^30,31 Time-varying effect modeling also can demonstrate how an association varies over a continuous moderator, allowing a more flexible examination of age-dependent changes. Driven by these innovative methodologies, this study aims to investigate the following research questions: 1) do maternal smoking measures (MLS and PTE) have causal effects on offspring’s smoking? 2) If so, does the strength of the relationship change across offspring’s age?

MATERIAL AND METHODS

Sample

The data were drawn from the Social and Emotional Contexts of Adolescent Smoking Patterns (SECASP) Study,³² a multi-aim project designed to capture early stages of smoking etiology (see Supplementary Material for details). Briefly, 9^th- and 10^th-grade students were recruited from 16 high schools in the greater Chicago metropolitan area in 2006. The SECASP Study oversampled current experimenters and current regular light smokers, and included random samples of never smokers and former experimenters at baseline.

The SECASP cohort was prospectively followed for 8 years (through 2014) with 10 surveys into young adulthood. A total of 1,263 adolescents completed the baseline questionnaire, and 1,022 adolescents had one parent also complete the parental questionnaire. For analyses of MLS, 900 adolescents of these 1,022 with parental data also had complete data on MLS and all covariates (described below). For analyses of PTE, which examine an additional exposure on top of having a biological mother who is a lifetime smoker, N=438 respondents reported positive MLS and had valid data on PTE. Figure 1 summarizes the process of sample selection throughout the analytic stages.

Figure 1.

Flowchart of the sample by analytic stages

As a secondary analysis of previously collected data, this study was approved by the Institutional Review Board (IRB) of the University of Illinois-Chicago and was exempt from IRB approval from Sanford Research.

Measures

Predictor: Maternal smoking measures

Maternal smoking status included two smoking behaviors: maternal lifetime smoking (MLS) and prenatal tobacco exposure (PTE).

MLS was proxy-reported by adolescents at baseline. MLS was collected in three levels: current, former, or never smoker, which was dichotomized into 1) lifetime smoker (current or former) and 2) never smoker. Previous studies using this sample justify this dichotomization.⁶

PTE was also assessed at baseline by the parent questionnaire reporting the frequency and quantity of smoking per trimester. PTE was dichotomized into having smoked for two or three trimesters vs. having smoked for zero or one trimester based on previous studies on this cohort which show an insignificant effect of transient tobacco exposure.⁷ Nearly all (26/30) mothers who smoked during only one trimester did so during the first trimester (vs. second or third), consistent with previous reports that the majority of pregnant smokers either quit in the first trimester or smoke throughout the pregnancy.³³

Outcome: Offspring smoking behaviors

Offspring smoking behaviors were defined independently for each 10 assessments based on the number of days smoked in the past 30 days (smoking frequency), which best indicates problematic smoking behavior among adolescents, especially experimental and light smokers^.34 The following outcome variables were included: 1) smoking frequency (number of days smoked in the past 30 days), 2) current smoking (1+vs. 0 days smoked in the past 30 days), 3) daily smoking (all 30 days vs. <30 smoked in the past 30 days).

Covariates

Propensities for 1) MLS and 2) PTE were assessed with 9 covariates at baseline based on the previous literature on risk factors of adolescent smoking.^35–38

Socio-demographics included sex, race, parental marital status, parental educational attainment, and familial income (see Tables 1 and 2 for categories used). Parental marital status was dichotomized into single-parent (divorced/separated, widowed, or never-married) or dual-parent status (married or living with a romantic partner). Proximity to other smokers was defined as having at least one closest friend who has ever tried a cigarette or cohabitating family members (i.e. father, step-parents, grandparents, siblings, etc) who currently smoke. For substance use, lifetime use of alcohol, marijuana, and other illicit drugs (cocaine, amphetamine, hallucinogen, inhalant, injection drug, steroid, or any prescription medication misuse) at baseline were dichotomized into ever use vs. never use.

Table 1.

Descriptive statistics of sample, stratified by maternal lifetime smoking status

		Mother with lifetime smoking history (N=438)	Mother with no smoking history (N=462)	Total (N=900)	P-value
Race/ethnicity	White	297 (67.81% )	264 (57.14% )	561 (62.33% )	<.01
	Black	59 (13.47% )	74 (16.02% )	133 (14.78% )
	Hispanic	50 (11.42% )	75 (16.23% )	125 (13.89% )
	Others	32 (7.31% )	49 (10.61% )	81 (9.00% )
Sex	Male	252 (57.53%)	248 (53.68%)	500 (55.56%)	>.05
	Female	186 (42.47%)	214 (46.32%)	400 (44.44%)
Parental marital status	Single-parent	136 (31.05% )	90 (19.48% )	226 (25.11% )	<.01
	Dual-parent	302 (68.95% )	372 (80.52% )	674 (74.89% )
Parental educational attainment	HS/GED or less	144 (32.88% )	100 (21.65% )	244 (27.11% )	<.01
	Some college or more	294 (67.12% )	362 (78.35% )	656 (72.89% )
Familial income (in thousand dollars)	<$12k	23 (5.25% )	17 (3.68% )	40 (4.44% )	<.01
	$12k-$25k	49 (11.19% )	43 (9.31% )	92 (10.22% )
	$25k-$50k	100 (22.83% )	98 (21.21% )	198 (22.00% )
	$50k-$75k	96 (21.92% )	82 (17.75% )	178 (19.78% )
	$75k-$100k	68 (15.53% )	75 (16.23% )	143 (15.89% )
	$100k-$150k	59 (13.47% )	84 (18.18% )	143 (15.89% )
	>$150k	43 (9.82% )	63 (13.64% )	106 (11.78% )
Have at least one closest friend or household member who smokes		421 (96.12% )	413 (89.39% )	834 (92.67% )	<01
Lifetime substance use	Alcohol	404 (92.24% )	384 (83.12% )	788 (87.56% )	<.01
	Marijuana	224 (51.14%)	225 (48.70%)	449 (49.89%)	>.05
	Other substances	155 (35.39% )	105 (22.73% )	260 (28.89% )	<.01

Note: statistics represent frequency (percentage) for categorical. P-values are computed from chi-squared tests. Bold: p<.05.

Table 2.

Descriptive statistics of the subpopulation with positive maternal lifetime smoking history, stratified by prenatal tobacco exposure status

		Exposure to tobacco in utero (N=132)	No exposure to tobacco in utero (N=289)	Subtotal with positive maternal lifetime smoking (N=421)	P-Value
Race/ethnicity	White	98 (74.24%)	194 (67.13%)	292 (69.36%)	>.05
	Black	14 (10.61%)	39 (13.49%)	53 (12.59%)
	Hispanic	11 (8.33%)	35 (12.11%)	46 (10.93%)
	Others	9 (6.82%)	2 (7.27%)	30 (7.13%)
Sex	Male	74 (56.06%)	169 (58.48%)	243 (57.72%)	>.05
	Female	58 (43.94%)	120 (41.52%)	178 (42.28%)
Parental marital status	Single-parent	45 (34.09%)	85 (29.41%)	130 (30.88%)	>.05
	Dual-parent	87 (65.91%)	204 (70.59%)	291 (69.12%)
Parental educational attainment	HS/GED or less	50 (37.88%)	86 (29.76%)	136 (32.30%)	>.05
	Some college or more	82 (62.12%)	203 (70.24%)	285 (67.70%)
Familial income (in thousand dollars)	<$12k	10 (7.58%)	11 (3.81%)	21 (4.99%)	>.05
	$12k-$25k	15 (11.36%)	32 (11.07%)	47 (11.16%)
	$25k-$50k	30 (22.73%)	68 (23.53%)	98 (23.28%)
	$50k-$75k	36 (27.27%)	55 (19.03%)	91 (21.62%)
	$75k-$100k	16 (12.12%)	49 (16.96%)	65 (15.44%)
	$100k-$150k	15 (11.36%)	43 (14.88%)	58 (13.78%)
	>$150k	10 (7.58%)	31 (10.73%)	41 (9.74%)
Have at least one closest friend or household member who smokes		129 (97.73%)	276 (95.50%)	405 (96.20%)	>.05
Lifetime substance use	Alcohol	125 (94.70% )	263 (91.00% )	388 (92.16% )	>.05
	Marijuana	67 (50.76%)	149 (51.56%)	216 (51.31%)	>.05
	Other substances	41 (31.06%)	109 (37.72%)	150 (35.63%)	>.05

Note: “No exposure to tobacco in utero” group indicates those with positive maternal lifetime smoking history but no prenatal tobacco exposure. Statistics represent frequency (percentage). P-values are computed from chi-squared tests. Bold: p<.05.

Time-varying variable: Age

Age was calculated from the self-reported birthday and the day of the survey at each follow-up. The median ages of participants were 15.6 (interquartile range (IQR): 15.1–16.1) at baseline and 24.4 (IQR: 23.8–24.9) at the 8th-year follow up.

Analysis

Aim 1: causal relationship

To examine the causal effects of MLS and PTE on offspring’s current smoking, daily smoking and smoking frequency across time, propensity matching was conducted to account for systemic differences and achieve approximate similarity between “treatment” (i.e. positive MLS or PTE status) and “control” (i.e. negative MLS or PTE status) groups. Nearest-neighbor matching was done using “MatchIt” package³⁹ in R software.⁴⁰. Details and balance diagnostics of the propensity matching are described in the Supplementary Material. Logistic regression analyses were conducted in the matched sample for the effect of MLS and PTE on offspring’s current and daily smoking statuses (4 separate analyses). Each regression analysis utilized all-available time points, and the within-subjects longitudinal correlation was accounted for by including subject-level random intercepts using the “nlme” package.⁴¹ For smoking frequency, generalized linear mixed Poisson regression analyses were performed using “glmmTMB” package.⁴² Smoking frequency was analyzed among past 30-day smokers only to account for excessive zero-day frequencies that originate from never or non-current smokers in the sample.

Aim 2: age-dependency

Since preliminary analyses on interaction effects with age revealed different effects of MLS and PTE at different ages, time-varying effect models (TVEMs) were utilized to examine how these associations vary across participant age.⁴³ Age was used as the “time” variable, i.e. as a continuous moderator across the continuous range of ages 13.9 to 27.5 in regression models described above. Models were also run using a publicly available TVEM macro for SAS, version 3.1.1.⁴⁴ It has the advantage of estimating the strength and direction of the association across the continuous time-varying variable, without making parametric assumptions about the trend as traditional models do (e.g. that changes over time are linear). Adolescents’ smoking frequency shows bimodal distribution (peaks at 0–5 days and 25–30 days), which TVEM Macro was not able to accommodate. Therefore, only current smoking and daily smoking were examined with the macro. Graphs were created with “ggplot2”⁴⁵ package from the output data of the macro.

RESULTS

Among 900 eligible participants, 48.7% (N=438) had a mother with a lifetime smoking history. PTE is a special subgroup of MLS, as those with positive PTE logically have positive MLS. A quality check assessed whether any adolescents were in the logically impossible category of being MLS- but PTE+ (perhaps due to measurement error from adolescents reporting on MLS and parents reporting on PTE); <5 adolescents met this criteron and were excluded from all analyses. Therefore, PTE was examined among the subset of MLS-positive adolescents; 31.4% (N=132) were PTE-positive (See Figure 1 for details).

Descriptive statistics of participants based on maternal lifetime smoking status are shown in Table 1. Those with MLS history were more likely to be non-Hispanic white, in a single-parent household with lower familial income, to have a parent with a lower level of educational attainment, have reported proximity to other smokers, and have tried alcohol and other illicit substances, compared to offspring of mothers without such smoking history (all p<.01). The baseline characteristics of those with positive MLS are further stratified based on their PTE history and presented in Table 2. Environmental factors of those with positive MLS history were highly comparable, and did not show significant difference regardless of PTE status.

The matched sample consisted of 438 treatment-control pairs for MLS (total N=876) and 132 pairs for PTE (total N=264). Matching by MLS reduced, but did not eliminate, imbalance in adolescents’ race/ethnicity, alcohol and other substance use, and parental education status, and marital status. Matching by PTE was successful, with adequate balance achieved on all covariates (see Supplement for details).

Results on Aim 1 (causal effects of maternal smoking) are summarized in Table 3, and indicate that maternal smoking statuses had significant and consistent main effects on smoking behaviors of offspring. Specifically, participants with positive MLS had a 1.09-fold significantly higher risk of currently smoking and a 1.08-fold significantly greater risk of daily smoking relative to MLS negative participants. Similarly, having a mother with a lifetime smoking history was associated with a significant 32% increase in the number of days smoked in the past 30 days. Among MLS-positive participants, PTE imposed an additional 1.09-fold higher risk for daily smoking but not for current smoking or smoking frequency (both p>.05). Maternal smoking also showed significant interaction effects with age, specifically MLS with daily smoking and smoking frequency, and PTE with all offspring smoking behaviors.

Table 3.

Effects of maternal smoking statuses on offspring’s smoking behavior over time

		Maternal lifetime smokinga		Prenatal tobacco exposure
		Risk ratio	95% CI	Coefficient	SE/95% CI
Current smoking	Maternal smoking	1.09	1.04 – 1.14***	1.04	0.96 – 1.13
	Age	1.01	1.00 – 1.01***	1.00	0.99 – 1.01
	Interaction	1.00	0.99 – 1.00	1.01	1.00 – 1.02*
Daily smoking	Maternal smoking	1.08	1.05 – 1.12***	1.09	1.02 – 1.17*
	Age	1.01	1.01 – 1.02***	1.01	1.01 – 1.02***
	Interaction	1.01	1.00 – 1.01***	1.02	1.01 – 1.03***
Smoking frequency	Maternal smoking	1.32	1.15 – 1.52***	1.21	0.96 – 1.53
	Age	1.07	1.07 – 1.08***	1.04	1.03 – 1.05***
	Interaction	0.99	0.98 – 0.99***	1.02	1.01 – 1.03***

^a: Regression models were adjusted for parental educational attainment, marital status, adolescents’ race, alcohol use, and other substance use due to covariate imbalance.

Note: The effect of prenatal tobacco exposure was examined in a subgroup of participants who reported having a biological mother with lifetime smoking history (N=421). Smoking frequency was examined among past 30-day smokers only. Statistically significant results are in bold (*: 0.01<p<0.05, **: 0.001<p<0.1, ***: p<0.001). “Interaction” indicates the interaction effect between the independent variable (maternal lifetime smoking or prenatal tobacco exposure) and age.

Aim 2 (age-dependency) was further investigated with TVEM to demonstrate the changes in the strength of relationships across the age range (Figures 2 and 3). In Figure 2, MLS has a significant, positive effect on both smoking behaviors among offspring, as indicated by the entire 95% confidence interval staying above 1 throughout most of the age range. However, the effect of MLS on two smoking behaviors remained similar over time without visible age-dependency. Figure 3 shows a notable age-varying association between PTE and daily smoking, consistent with the strong interaction effect present in the regression model. Specifically, PTE was not associated with daily smoking during adolescence, but became a significant risk factor in young adulthood, as evidenced by the lower band of 95% confidence interval excluding 1 around the age of 18.2 for daily smoking, and progressively higher risk above that age. PTE did not impose additional risk for current smoking, however.

Figure 2.

Age-varying effects of maternal lifetime smoking on offspring’s smoking status

Note: The y-axis represents the strength (risk ratio) of the association between maternal lifetime smoking and current (Figure 1a) or daily (Figure 1b) smoking status of offspring. The central solid line indicates the point estimate of the risk ratio, and the gray ribbon with dashed borders shows the 95% confidence interval. The horizontal line shows the risk ratio of 1 (no predictive validity).

Figure 3.

Age-varying effects of prenatal tobacco exposure on offspring’s smoking status, in participants with positive maternal lifetime smoking history

Note: The y-axis represents the strength (risk ratio) of the association between prenatal tobacco exposure and current (Figure 2a) or daily (Figure 2b) smoking status of offspring. The central solid line indicates the point estimate of the risk ratio, and the gray ribbon with dashed borders shows the 95% confidence interval. The horizontal line shows the risk ratio of 1 (no predictive validity).

DISCUSSION

This present study examines the potentially causal and age-dependent effects of maternal smoking on the smoking behaviors among offspring. Using propensity score methods that robustly account for shared confounding between maternal and youth smoking, MLS may have causal effects on offspring’s current and more frequent smoking. Furthermore, being exposed to tobacco in utero (PTE-positive) puts adolescents at additional risk for daily smoking. The effect of PTE shows marked age-dependency, becoming significant in young adulthood but not during adolescence. Taken together, the findings of this study suggest a potential causal role of maternal smoking (both MLS and PTE) on smoking behaviors among offspring.

This study confirms previous research suggesting a significant role of maternal smoking on smoking behaviors of offspring even after adjusting for common risk factors.^5,6,46–48 In particular, studies have shown offspring smoking is more strongly associated with biological parents’ smoking status than adoptive parents’, indicating that there are additional effects of parental smoking over and above shared environmental risk factors.^46,47 This study also echoes the previous findings of age-dependency in which the effect of maternal smoking increases with the age of offspring,^{6,7,12,15,48,49} although our findings only find evidence of this with PTE, not with MLS. However, other research has questioned the effect of intrauterine exposure on offspring’s smoking initiation,^19–21 progression to heavy smoking,^19,20 and nicotine dependence.²² These disagreements largely originate from a number of genetic, environmental commonalities shared between the mother and youth and limitations of conventional statistical methods that cannot fully account for this unique relationship.

The mechanism underlying our current findings that MLS and PTE have potentially causal effects on offspring smoking remains unclear. In fact, each of these associations may encompass several specific environmental and/or genetic mechanisms. Environmentally, maternal smoking may act on youth through behavioral modeling,⁸ relaxed parental control and permissiveness,^8–10 perceived availability,¹¹ and diminished expectations on negative outcomes of smoking.¹⁰ Individually unique environmental risks that are not shared within the family (i.e., perceived cigarette availability or safety)^10,11 may also explain the observed relationship between maternal and offspring smoking. While still in its infancy, genetic research has suggested candidate genes including intergenerational transmission of CHRNA5-A3-B4 cluster that predispose carriers for nicotine dependence,^14,50 and gene variations in dopamine receptors that augment nicotine-induced reward.^17,50 The interaction between the gene and environment also represents a possible mechanism in which the effect of inherited vulnerability depends on environmental exposures. For example, externalizing⁵¹/internalizing⁵² comorbidities or heritable personality traits (i.e., risk-seeking personalities)⁵³ that are heavily influenced by both genetic and environmental factors may mediate the observed, possibly causal relationship between maternal smoking and offspring’s smoking.^54,55

PTE is distinguished from general maternal postnatal smoking as the developing fetus is directly exposed to tobacco toxicants through the umbilical cord, indicating a possibility of direct biological mechanisms. In this study, the effect of PTE was exclusively examined in participants who reported having a mother with a lifetime smoking history. These MLS-positive respondents, regardless of their PTE statuses, share similar environmental vulnerabilities that arise from having a formerly or currently smoking mother, which accounts for the environmental influences in the analyses. This control of environmental risk factors suggests that the effect of PTE, if any, will arise from additional mechanisms that are distinct from environmental influences, such as behavioral teratologic mechanisms. The current study has demonstrated two notable findings on PTE: that 1) its effect has age-dependency such that it is dormant during adolescence but starts to take effect in young adulthood and 2) it has significant effects on daily smoking of offspring but not on current (past 30-day) smoking.

Previous studies have shown similar age-dependent effects of PTE,^{6,7,12,15,48,49} which are often explained with genetically-related mechanisms, such as upregulated nicotine receptors⁵⁶ or dopaminergic system¹⁷ and alterations in the DNA methylation patterns that may lead to epigenetic changes.⁵⁷ Although they were not directly tested in this study, these genetic mechanisms have the potential to explain the main findings of this study as well. First, the proportion of heritability in youth smoking is known to increase through adulthood and the contribution of family-shared environmental risks tends to decrease after adolescence.^15,58 Furthermore, genetically imposed risks are much more significant for heavy, problematic smoking than for smoking initiation,^58,59 which is consistent with the results of this study on daily smoking as well.

This study is novel in its use of advanced statistical methodologies to better account for shared risks between maternal and offspring smoking behaviors. The significant residual associations even after the propensity matching suggest potential causality of maternal smoking on offspring smoking. Examining above-mentioned environmental/genetic mechanisms that may explain this causal relationship, e.g. through mediation analysis, is an important avenue for future research.

These findings, however, should be interpreted within the context of several limitations. First, propensity score methods make a crucial assumption that there are no unmeasured confounders.²⁶ This assumption is frequently violated, with this study being no exception. Thus, the remaining association between maternal and offspring smoking may be further reduced in size or even rendered insignificant if relevant confounders are added to the analyses. The exact mechanisms of these potentially causal relationships remain unclear and warrant further research. Additionally, the small sample size may have limited the ability to detect the differences in baseline characteristics between exposed and non-exposed groups, the age-varying effects of MLS, or the effect of PTE on smoking frequency. The main effects reported in this study are relatively small in size. This may be attributable to more robust accounting of shared environmental effects. The study is also based on self- or proxy-reports, which are susceptible to under-reporting and may have affected the results in unknown directions. Variables used in propensity matching were collected at baseline, some of which may have changed over the study period (i.e., MLS, offspring’s substance use). However, effects of environmental factors accumulate throughout the adolescent’s life, and significant changes in shared environmental factors between the mother and offspring are rarely expected in late adolescence. Lastly, the SECASP study was launched in 2006 and followed then-high school students in the greater Chicago area, and may not be representative of other geographical areas or to current adolescents, although the effects of maternal behaviors on adolescents are expected to persist across generations.

The current study has a number of strengths that add to the literature as well. First, this study utilizes multiple innovative methodologies that include propensity score and time-varying effect modeling with mixed effects. Specifically, it is one of the first studies that attempted to use statistical methods to reduce the effect of environmental risk factors in the association of maternal and offspring smoking. This application shows promising potential to overcome the constraints of family-based research designs by reproducing consistent results with a general community sample. Furthermore, the SECASP study remains one of the most comprehensive long-term longitudinal studies on early stages of smoking etiology, providing unique opportunities to study the development of addiction and progression to heavy smokers. This study furthers research on the well-known association between maternal and offspring smoking for their possible causal relationship and suggests age-dependency as a possible mechanism for the controversial findings of the previous literature. Lastly, the results of this study raise concerns for possible nicotine-specific effects of prenatal use of diverse, non-cigarette tobacco products such as e-cigarettes, hookah, and cigar, considering the growing popularity of these products among pregnant women.⁶⁰ These results inform families regarding adolescents’ risks for current, heavy smoking and provide a foundation for optimal public health interventions. This study supports a targeted approach and timely education effort for tobacco-exposed adolescents before they progress to chronic smoking.

CONCLUSION

Cigarette smoking represents one of the biggest public health challenges in the United States. Even after accounting for shared risk factors using propensity score methods, this study demonstrates persistent and significant effects of maternal smoking on offfspring’s smoking behaviors and illustrates the age-dependent effect of prenatal smoking that becomes evident in offspring’s young adulthood. These findings indirectly, but consistently, may suggest genetic or heritable contributions to the etiology of smoking progression and addiction. Given that most chronic smokers initiate smoking and develop an addiction during adolescence, a history of exposure to maternal smoking can assist in the identification of adolescent smokers who are at a greater risk of developing persistent, heavy smoking habits in adulthood. This study calls for future research to explore specific genetic, environmental mechanisms or gene-environment interactions through which maternal smoking may cause offspring to smoke and provides promising evidence for more targeted anti-tobacco interventions.

Highlights.

• We use propensity score methods to estimate causal effects of maternal smoking

• This includes lifetime smoking (MLS) and prenatal tobacco exposure (PTE)

• Both PTE and MLS have effects on youth smoking after propensity adjustment

• MLE’s effect on offspring smoking is stable across adolescence and young adulthood

• PTE’s effect is null in adolescence but becomes significant in young adulthood