Average and Heterogeneous Associations between Secondhand Smoking and Tooth Loss

Abstract

Secondhand smoke (SHS) is a pervasive health concern increasingly associated with tooth loss, yet studies assessing its population-level associations and potential heterogeneity are limited. This study aimed at estimating the average and subgroup-specific effects of SHS exposure on tooth loss among US adults using modern causal inference–based methods. This study analyzed cross-sectional data from 6,019 nonsmokers aged ≥20 y, representing 48.2 million individuals, in the National Health and Nutrition Examination Survey (2009 to 2018). SHS was defined by serum cotinine. This study quantified the association between SHS exposure and tooth loss using targeted minimum loss estimation, a doubly robust estimator that enhances confounding adjustment in observational data. Confidence intervals were obtained via bootstrap resampling. Heterogeneity of the association was evaluated by a data-adaptive causal forest method that identifies subgroup-specific variation. SHS exposure was associated with having 0.74 fewer teeth (average association, −0.739; 95% CI, −1.053 to −0.426) and with a 4.6-percentage higher probability of having <20 teeth (average association, 0.046; 95% CI, 0.027 to 0.065). Larger associations were observed among adults aged ≥45 y, those with lower socioeconomic position, individuals with higher sugar intake or medical comorbidities, and participants at the extremes of flossing frequency. This study found that SHS exposure was associated with greater tooth loss, with disproportionate burdens observed among adults facing socioeconomic disadvantages or chronic health conditions. These findings underscore the potential value of integrating oral health outcomes into tobacco control initiatives and guiding targeted prevention for groups at elevated risk.

Keywords

passive smoking health equity nutrition surveys potential outcomes socioeconomic factors tobacco control

Introduction

Smoking remains a major global health concern, with well-established detrimental effects on multiple organ systems (Dai et al 2022). Importantly, the harm of tobacco use extends beyond smokers themselves: secondhand smoke (SHS) exposure contributes substantially to the global disease burden (Flor et al 2024). While public health initiatives have made progress in reducing active smoking, SHS continues to pose risks for nonsmokers, particularly vulnerable populations (GBD 2017 Risk Factor Collaborators 2018). Oral health is increasingly recognized as an essential component of overall health and quality of life (Aida 2021), as tooth loss has been linked to malnutrition, weight loss, functional limitations, cardiovascular diseases, and higher mortality (Zelig et al 2022; Kusama et al 2023; Matsuyama et al 2023; Matsuyama et al 2024; Wang et al 2026). The burden of tooth loss also mirrors broader socioeconomic disparities (Mbulo et al 2016).

Accumulating evidence has linked SHS exposure to adverse oral health outcomes, including periodontal disease and dental caries (Sanders and Slade 2013; González-Valero et al 2018), raising concerns that SHS could influence tooth loss (Umemori et al 2020). Yet, despite well-established evidence supporting a causal link between active smoking and tooth loss (Baumeister et al 2025), analogous evidence for SHS remains limited. Most existing studies rely on self-reported measures of SHS and have not examined heterogeneity in the observed associations, thereby limiting their relevance for policy formulation and preventive interventions (Oliveira et al 2022; Luo et al 2025). These limitations are particularly concerning because reliance on self-reported exposure without objective biomarkers and the absence of heterogeneity assessment may lead to exposure misclassification and obscure population subgroups with stronger associations. The World Health Organization’s Oral Health Action Plan 2023–2030 calls for the integration of oral health within national tobacco control efforts, highlighting the importance of clarifying how tobacco exposure, including SHS, affects oral health. Because experimental studies on SHS exposure are ethically infeasible, observational data play a crucial role in advancing understanding of its impact. Recent developments in causal inference methodology provide opportunities to strengthen inference from such data by improving adjustment for confounding and enabling data-adaptive exploration of heterogeneity (Nugent et al 2024). These tools allow researchers to approximate aspects of experimental designs and to identify subgroups who may be disproportionately affected.

To bridge the gap, the present study aimed to evaluate the association between SHS exposure and tooth loss in a nationally representative sample of US adults, using modern causal inference approaches. To our knowledge, this is the first study that leveraged causal inference methods, including targeted minimum loss estimation and causal forest, within the Neyman-Rubin potential outcomes framework to estimate average and heterogeneous treatment effects of SHS on tooth loss. The findings may underscore the value of incorporating oral health into tobacco control efforts and inform public health policies and clinical supports that prioritize high-risk populations.

Materials and Methods

Study Population

The National Health and Nutrition Examination Survey (NHANES) is an ongoing cross-sectional survey that employs complex multistage sampling weights to reflect the national condition. NHANES 2009 to 2018 are the latest consecutive waves that incorporated dentition examination data, from which 8,063 nonsmokers who participated in the survey on household smoking were included. Participants without available data on serum cotinine level and complete dentition examination (n = 2,044) were excluded. Eventually, 6,019 nonsmokers were incorporated for the analytic procedure. The flowchart in Figure 1 illustrates the inclusion strategy.

Figure 1.

Workflow of the inclusion criteria. NHANES, National Health and Nutrition Examination Survey.

Measurements of Tooth Loss

A full-mouth dentition examination was performed by calibrated examiners. The third molar was excluded from the tooth count. A permanent tooth present (excluding a permanent root fragment) was considered a “tooth present.” Our primary outcome was total tooth count, which was analyzed as a continuous variable to maximize statistical power (Lopez-Ayala et al 2025). As the secondary analyses, the continuous tooth count was dichotomized into having <20 natural teeth and having at least 20 natural teeth.

Identification of Exposure to SHS and Smoking Status

Participants who were currently smoking or had quit, with histories of smoking >100 cigarettes in their lives, were identified as either current or former smokers, while their counterparts were identified as nonsmokers. Serum cotinine is the primary metabolite of nicotine and is widely accepted as an optimal biomarker for tobacco smoke exposure due to its half-life of approximately 16 to 20 h, which may reflect recent exposure over 1 to 3 d (Benowitz 1996). Thus, the present study identified nonsmokers as being exposed to SHS if their serum cotinine levels were ≥0.05 ng/mL per a previous study (Tsai et al 2021). The alternative definition considered the presence of household smokers based on serum cotinine levels.

Ascertainment of the Confounders

According to existing evidence (Bashir et al 2026), we included the following as potential confounders: sex, age, race, poverty-to-income ratio, educational attainment, weekly dental floss by frequency, daily sugar consumption, body mass index, diabetes, hypertension, and hyperlipidemia. Detailed definitions can be found in Appendix Table 1.

Statistical Analysis

All statistical analyses were performed in R software (version 4.3.0). The modeling assumption was visualized by the directed acyclic graph in Appendix Figure 1. First, a descriptive analysis was performed on all participants. According to the potential outcome framework, the “What if” scenarios hypothesized in this study were as follows: “What if all nonsmokers were exposed to SHS?” and “What if all nonsmokers were not exposed to SHS?” The average treatment effect (ATE) was calculated by the difference in the mean outcome (ie, tooth count) values under the 2 scenarios (Hernan and Robins 2025).

Targeted minimum loss estimator (TMLE) is a doubly robust estimator that combines the outcome model of g-computation and the exposure model of inverse probability weighting and therefore may provide valid estimates where either model holds (Schuler and Rose 2017). We applied the Super Learner ensemble within the TMLE framework, stacking regularized linear regression, extreme gradient boosting (XGBoost), and neural networks. To estimate the nuisance functions, we employed a stratified 10 × 10 cross-validation (Brown et al 2025): for each outer 10-fold used to obtain cross-fitted Q estimates, the Super Learner was tuned on a prespecified hyperparameter grid (Appendix Table 3) in the remaining outer folds via an inner 10-fold cross-validation, in which candidate learners with suboptimal hyperparameters were effectively downweighted (Bodnar et al 2022; Polley et al 2024). This ensemble strategy enhances robustness and mitigates bias from misspecification by leveraging the complementary strengths of diverse algorithms (Moccia et al 2024). ATE was estimated by TMLE in the tmle3 R package (Gruber and Laan 2012; van der Laan 2023); the conditional ATE (CATE) was derived by a causal forest from a generalized random forest algorithm in the grf R package (Athey et al 2019); and hyperparameter was tuned prior to growing 2,000 trees. To avoid overfitting, the data were divided into 5 folds; for each fold, prediction was derived from trees trained by the other 4 folds. Model performance of the generalized random forest was assessed by the group ATE and the best linear predictor analysis (Athey et al 2017; Athey and Wager 2019).

To explore heterogeneity, we employed 2 strategies. First, participants were stratified into different groups based on the interquartile range (IQR) of CATEs. For the tooth count outcome, individuals of a lower percentile of the IQR were those experiencing greater tooth loss from SHS. In contrast, for the dentition status outcome (probability of having <20 teeth), individuals of a lower percentile of the IQR were those who retained more teeth, per the outcome coding (Matsuyama et al 2024). Second, participants were classified into different groups by employing causal distillation tree in the causalDT package, which distills complex heterogeneity from CATEs derived from a generalized random forest and summarizes the subgroups with various treatment effects into a tree structure, offering additional clinical interpretability (Rehill 2024; Huang et al 2025). The optimal maximum tree depth for the student model was based on a Jaccard similarity index. For ATE, the 95% confidence interval (95% CI) was calculated by a bootstrapping method with 1,000 iterations (Tibbe and Montoya 2022), and E value was applied to estimate the residual confounding (VanderWeele and Ding 2017). Specifically, E values were calculated by the formula for linear regression coefficients for the continuous tooth count outcome and the formula for risk differences for dentition status, implemented via a publicly available online calculator (https://www.evalue-calculator.com/evalue/; accessed December 25, 2025). A random forest algorithm was used to complete the missing data in the missForest R package (Tang and Ishwaran 2017). The missing rates and distribution of the missing data are illustrated in Appendix Table 2 and Appendix Figure 2. P < 0.05 was considered statistically significant. The NHANES weight was applied to all analyses per the NHANES guideline. R code used in the study can be found at https://github.com/Jackieyueqi/SHS_tooth. The results were reported following STROBE (Strengthening the Reporting of Observational Studies in Epidemiology).

Sensitivity Analysis

Sensitivity analyses were conducted to evaluate the robustness of the findings by 1) applying alternative serum cotinine thresholds to define SHS; 2) defining SHS by urinary 4-methylnitrosamino-1-3-pyridyl-1-butanol (NNAL), a biomarker of tobacco exposure with a half-life of 10 to 18 d, allowing it to reflect SHS exposure over 4 to 8 wk (Goniewicz et al 2009); 3) completing missing values in the exposure and outcome; and 4) reestimating models without survey weights. More details are provided in the Appendix 2.

Results

Characteristics of Participants

In total, 6,019 nonsmokers, representing 48.2 million US adults, were included. Among them, 36% were identified as being exposed to SHS based on their serum cotinine levels, corresponding to 16.0 million adults. When compared with unexposed participants, those exposed to SHS more frequently reported living with household smokers and had a higher prevalence of having <20 natural teeth. SHS-exposed individuals also tended to be younger, female, of lower socioeconomic position, and non-Hispanic Black; they had higher sugar intake and more comorbid conditions; and they flossed less often (Table 1). Of the included NHANES cycles, only 2009 to 2014 contained laboratory data of urinary NNAL, resulting in a subsample of 4,512 participants for this sensitivity analysis.

Table 1.

Baseline Characteristics of Participants Stratified by Exposure to Secondhand Smoking.

	Group, n (%) or Mean ± SD
Characteristic	Overall (N = 48,230,321)^a	Unexposed (n = 32,264,986)^a	Exposed (n = 15,965,336)^a
Frequency (unweighted) ^b	6,019	3,838	2,181
Sex: female	3,558 (57)	2,418 (61)	1,140 (49)
Age, y
20 to 30	1,056 (18)	463 (13)	593 (29)
30 to 44	1,793 (31)	1,161 (31)	632 (32)
45 to 64	2,033 (36)	1,384 (39)	649 (30)
≥65	1,137 (15)	830 (17)	307 (9.8)
Race
Non-Hispanic White	2,172 (63)	1,457 (66)	715 (57)
Non-Hispanic Black	1,406 (12)	626 (7.9)	780 (22)
Mexican American	887 (9.0)	645 (9.6)	242 (7.6)
Other Hispanic	675 (6.7)	520 (7.4)	155 (5.1)
Other	879 (9.0)	590 (9.1)	289 (8.8)
PIR
≤1	1,240 (14)	614 (10)	626 (21)
1 to 3	2,516 (35)	1,541 (32)	975 (41)
≥3	2,263 (51)	1,683 (58)	580 (37)
Educational attainment
College graduate or above	3,463 (66)	2,369 (72)	1,094 (54)
High school graduate or equivalent	1,238 (20)	672 (16)	566 (28)
Less than high school	1,318 (14)	797 (12)	521 (17)
BMI, kg/m²
Normal weight	1,677 (28)	1,133 (30)	544 (24)
Overweight	1,940 (33)	1,280 (34)	660 (31)
Obesity	2,308 (37)	1,375 (34)	933 (42)
Underweight	94 (1.7)	50 (1.5)	44 (2.0)
Floss, times/wk
Never	1,864 (26)	1,058 (23)	806 (32)
1 to 2	1,281 (25)	811 (25)	470 (26)
3 to 4	783 (14)	503 (14)	280 (14)
5 to 7	2,091 (35)	1,466 (38)	625 (28)
Diabetes: yes	891 (10)	573 (10)	318 (11)
Hypertension: yes	2,303 (34)	1,508 (34)	795 (33)
Hyperlipidemia: yes	3,451 (57)	2,338 (60)	1,113 (50)
Household smokers (person)
0	5,336 (90)	3,735 (98)	1,601 (75)
1	493 (6.6)	69 (1.3)	424 (17)
2	149 (2.5)	29 (0.9)	120 (5.7)
≥3	41 (0.9)	5 (0.2)	36 (2.3)
Sugar consumption, g/d	111 ± 62	108 ± 56	118 ± 71
Natural tooth count	25 ± 6	25 ± 6	24 ± 6
Dentition status: natural teeth
≥20	4,982 (89)	3,231 (91)	1,751 (87)
<20	1,037 (11)	607 (9.5)	430 (13)

BMI, body mass index; PIR, poverty-to-income ratio; SHS, secondhand smoking.

Weighted US frequency.

Data in the table represent the unweighted frequency (weighted column-wise proportion).

Average Associations between SHS Exposure and Tooth Loss

When serum cotinine was the primary SHS definition, exposure was associated with 0.74 fewer teeth (95% CI, −1.053 to −0.426) and a 4.6-percentage higher prevalence of having <20 teeth (95% CI, 0.027 to 0.065). Estimates obtained per the alternative definition requiring elevated serum cotinine and a household smoker were similar in magnitude and direction (Table 2).

Table 2.

Average Associations Between Secondhand Smoking and Tooth Loss.

Outcome: Exposure	ATE^a	95% CI	E Value
Tooth count: continuous
SHS: serum cotinine levels	–0.739	–1.053, −0.426	1.48
SHS: serum cotinine levels and presence of household smokers	–0.753	–1.063, −0.444	1.49
Dentition status: binary
SHS: serum cotinine levels
≥20 natural teeth^b	0.046	0.027, 0.065	1.80
SHS: serum cotinine levels and presence of household smokers
≥20 natural teeth^b	0.050	0.031, 0.069	1.82

ATE, average treatment effect; SHS, secondhand smoking.

ATE denotes the estimand of the doubly robust estimator and is interpreted as an adjusted association in this cross-sectional study.

Reference: <20 natural teeth.

Performance of Causal Forest Model

Figure 2 depicts the distribution of CATEs. CATEs for the primary SHS definition centered on a median reduction of 0.653 teeth (IQR, −1.036 to −0.335) and a 3.5–percentage point increase in the probability of having <20 teeth (IQR, 0.018 to 0.058).

Figure 2.

Distribution of the conditional average treatment effects of secondhand smoking on tooth loss. (A) Distribution of conditional average treatment effects of secondhand smoking on tooth count: the left side of the x-axis indicates greater tooth loss (higher effects), while the right side indicates less tooth loss (lower effects). (B) Distribution of conditional average treatment effects of secondhand smoking on the dentition status (≥20 vs <20 natural teeth): the left side of the x-axis indicates a lower possibility of having <20 natural teeth. CATE, conditional average treatment effects; Q25, 25% quartile of the conditional average treatment effects; Q75, 75% quartile of the conditional average treatment effects; SHS, secondhand smoking.

Higher- and lower-effect subgroups stratified by CATEs showed clear separation for the continuous tooth count outcome (difference, ~0.7 teeth), with more limited separation for the binary dentition outcome under the primary SHS definition (difference, ~0.005). Best linear predictor analyses indicated that the causal forests for tooth count were well calibrated and captured treatment effect heterogeneity, whereas forests for dentition status, although well calibrated, did not successfully capture the heterogeneity (Appendix Tables 7 and 8).

Appendix Figure 3 depicts the models’ stability. The student decision tree distilled from the causal forest achieved optimal performance at a depth of 2 and was constructed only for models in which the best linear predictor analysis indicated detectable treatment effect heterogeneity, providing a stable and interpretable summary of the heterogeneity structure.

Heterogeneous Association with Tooth Count

Descriptive analyses based on the CATEs indicated that adults aged ≥45 y, those with higher sugar intake, and those with metabolic conditions exhibited larger associations between SHS exposure and tooth loss. A U-shaped pattern was observed for flossing frequency, with greater vulnerability among participants who never flossed and those who flossed nearly daily (Table 3, Appendix Tables 4–6).

Table 3.

Characteristics across Subgroups Stratified by Conditional Average Treatment Effects.

	Effect,^a n (%) or Mean ± SD
Characteristic	Overall (N = 48,230,321)^b	Higher (n = 10,428,647)^b	Higher-Medium (n = 11,813,775)^b	Medium-Lower (n = 12,497,850)^b	Lower (n = 13,490,049)^b	P Value^c
Frequency (unweighted)	6,019	1,505	1,505	1,504	1,505	—
Sex: female	3,558 (57)	720 (45)	879 (56)	958 (61)	1,001 (64)	<0.001
Age, y						<0.001
20 to 30	1,056 (18)	127 (9.8)	154 (10)	256 (16)	519 (35)
30 to 44	1,793 (31)	308 (23)	443 (29)	604 (41)	438 (31)
45 to 65	2,033 (36)	654 (44)	615 (45)	472 (34)	292 (22)
≥65	1,137 (15)	416 (23)	293 (16)	172 (8.7)	256 (12)
Race						<0.001
Non-Hispanic White	2,172 (63)	545 (61)	494 (61)	533 (63)	600 (66)
Non-Hispanic Black	1,406 (12)	347 (13)	353 (13)	382 (14)	324 (11)
Mexican American	887 (9.0)	250 (9.9)	267 (10)	228 (9.7)	142 (6.3)
Other	879 (9.0)	163 (7.5)	202 (8.6)	225 (8.9)	289 (11)
Other Hispanic	675 (6.7)	200 (8.2)	189 (7.6)	136 (5.1)	150 (6.1)
PIR						0.052
≤1	1,240 (14)	465 (23)	334 (15)	262 (12)	179 (7.6)
1 to 3	2,516 (35)	632 (36)	645 (37)	629 (35)	610 (33)
≥3	2,263 (51)	408 (41)	526 (48)	613 (53)	716 (60)
Educational attainment						<0.001
College graduate or above	3,463 (66)	619 (53)	798 (61)	964 (71)	1,082 (76)
High school graduate or equivalent	1,238 (20)	311 (21)	329 (23)	310 (19)	288 (19)
Less than high school	1,318 (14)	575 (27)	378 (17)	230 (9.3)	135 (4.7)
BMI, kg/m²						0.3
Normal weight	1,677 (28)	409 (26)	383 (27)	420 (30)	465 (29)
Overweight	1,940 (33)	525 (36)	515 (35)	455 (32)	445 (32)
Obesity	2,308 (37)	558 (37)	591 (37)	608 (37)	551 (36)
Underweight	94 (1.7)	13 (1.3)	16 (0.9)	21 (1.5)	44 (2.9)
Floss, times/wk						<0.001
Never	1,864 (26)	661 (40)	472 (29)	356 (20)	375 (20)
1 to 2	1,281 (25)	157 (12)	201 (17)	350 (25)	573 (43)
3 to 4	783 (14)	108 (6.4)	199 (13)	248 (18)	228 (17)
5 to 7	2,091 (35)	579 (41)	633 (42)	550 (37)	329 (21)
Diabetes: yes	891 (10)	301 (14)	233 (12)	197 (9.3)	160 (6.7)	<0.001
Hypertension: yes	2,303 (34)	829 (53)	661 (42)	446 (26)	367 (19)	<0.001
Hyperlipidemia: yes	3,451 (57)	983 (67)	964 (66)	853 (59)	651 (40)	<0.001
Sugar consumption, g/d	111 ± 62	141 ± 94	110 ± 52	105 ± 42	95 ± 41	<0.001

The definition of secondhand smoking considered serum cotinine levels only. The outcome was tooth count. Data in the table represent the unweighted frequency (weighted column-wise proportion).

BMI, body mass index; PIR, poverty-to-income ratio.

Participants with conditional average treatment effect values: <25% quartile, higher effect (ie, the left side of the x-axis indicated greater tooth loss); >75% quartile, lower effect (ie, the right side of the x-axis indicated fewer tooth loss); 25% to 50% quartile, higher-medium effect; 50% to 75%, medium-lower effect.

Weighted US frequency.

Differences in binary and nominal variables were tested by a weighted chi-square test, while those in ordinal and continuous variables were tested by a weighted Kruskal-Wallis rank sum test.

CDT-derived decision trees consistently identified lower educational attainment and higher sugar consumption as key factors delineating higher-risk subgroups (Appendix Fig. 3).

Results of Sensitivity Analyses

When serum cotinine thresholds varied from 0.05 to 3 ng/mL, the associations with tooth count grew increasingly negative at higher cutoffs—for example, with ATEs of −0.836 at >1.5 ng/mL and −0.884 at >2 ng/mL, reaching −1.164 at >3.0 ng/mL. However, the associations with dentition status remained statistically significant across most cutoffs but did not show a consistent monotonic trend. The associations with tooth count were consistently significant across thresholds when SHS was defined by NNAL, with ATEs ranging from −0.677 to −0.692. For dentition status, the estimate was 0.054 (95% CI, 0.025 to 0.083) and statistically significant at the 3-pg/mL threshold but attenuated and became nonsignificant at higher cutoffs (0.033 at 6 pg/mL, 0.021 at 9 pg/mL; Appendix 2: Tables 1–3).

Sensitivity analyses based on complete case data for serum cotinine and tooth count and models fitted without survey weights yielded smaller but directionally consistent estimates: ATEs on tooth count ranged from −0.449 to −0.653 and ATEs on dentition status from 0.032 to 0.043, supporting the robustness of the primary findings. For the heterogeneous associations, both sensitivity analyses identified broadly similar profiles of vulnerable subgroups. However, in the unweighted analysis, the causal forests were well calibrated but did not capture meaningful heterogeneity, whereas in the complete case analysis, the forests were well calibrated and detected statistically significant heterogeneity, albeit with evidence of underdispersion in the predicted effects (Appendix 2: Tables 4–16).

Overall, SHS exposure was consistently associated with modest reductions in tooth count and higher prevalence of having <20 teeth, with larger associations observed in several socioeconomically and medically vulnerable subgroups.

Discussion

To our knowledge, this is the first study to apply advanced causal inference methods to evaluate the association between SHS exposure and tooth loss. SHS exposure, as assessed by a biomarker, was associated with fewer natural teeth and a higher probability of having <20 teeth. Analyses of heterogeneous associations indicated that several demographic, socioeconomic, behavioral, and metabolic characteristics were linked to greater susceptibility, and these patterns were broadly supported across sensitivity analyses.

Average Associations

Our results align with previous observational studies that linked SHS to oral health outcomes. Previous work has shown elevated risks of edentulism and periodontal disease among nonsmokers exposed to environmental tobacco smoke, although most studies relied on self-reported exposure and conventional regression approaches (Umemori et al 2020). By incorporating an objective biomarker of exposure and a doubly robust estimator, the present study extends this evidence base and provides more robust population-level estimates. Our results also agree with studies documenting the benefits of tobacco control measures for oral health (Baumeister et al 2025), underscoring the importance of minimizing involuntary exposure among nonsmokers. Additionally, evaluations of smoke-free legislation have shown that public smoking bans do not compromise economic growth and may confer broad population benefits (Fu et al 2024). Importantly, there remains a lack of studies that directly quantify the impact of smoking control policies on dental outcomes via causal inference methods, highlighting an important gap for future research.

Heterogeneous Associations

Heterogeneity in the associations was observed. The causal distillation tree and descriptive analysis suggested that individuals with lower educational levels and higher sugar consumption appeared to be more susceptible to the adverse association between SHS exposure and tooth loss. A possible explanation is that those with higher education tend to possess greater health literacy, preventive behaviors, and access to professional dental care (Ghanbari-Jahromi et al 2023), which may attenuate the SHS-related harm and tooth loss and thereby support tooth retention (Baumeister et al 2025). These findings emphasize that public interventions for oral health should be tailored to socioeconomic backgrounds to reduce inequalities in oral health (Thomson et al 2018). This notion was supported by a study revealing that public intervention such as water fluoridation may yield larger benefits on caries reduction in people from lower educational backgrounds (Matsuyama et al 2025). Additionally, SHS has been linked to shifts toward cariogenic oral microbiota and to impaired oral immune defenses, creating conditions conducive to dental caries (Golpasand Hagh et al 2013; Al-Marzooq et al 2022). In the presence of high sugar consumption, caries risk and progression may increase. Our finding provides additional rationale for taxing sugar-sweetened products, with potential revenue allocated to oral health promotion in vulnerable subgroups (Andreyeva et al 2022).

Individuals who never used dental floss and those who reported flossing >5 d per week appeared to be more susceptible to SHS-related tooth loss. This seemingly paradoxical pattern may reflect, at least in part, that frequent flossing is more commonly reported among those who are already engaged with dental care due to existing oral health problems, rather than indicating a protective behavior per se (Chapple et al 2015). While this paradox was found in another study in a similar population (Cepeda et al 2017), it should be interpreted prudently without robust validation. Similarly, common comorbidities, including diabetes, hypertension, and hyperlipidemia, were identified as features of the vulnerable group, suggesting that SHS may exacerbate the oral health burden in persons already at elevated systemic risk (Kocher et al 2018; Bitencourt et al 2023). This observation is biologically plausible, given that these conditions are characterized by chronic inflammation and metabolic dysregulation, which are implicated in periodontitis (Hajishengallis and Chavakis 2021). On the grounds that these comorbidities are particularly prevalent among elderly adults (Cao et al 2023) and that our results indicate that the elderly are more susceptible to SHS, clinicians should systematically assess not only personal but also household smoking exposure when managing patients with these conditions. At a broader level, these findings reinforce the need for public health strategies that integrate smoking control with chronic disease management, thereby improving oral health and overall resource allocation (Neophytou et al 2024).

Strengths and Limitations

This study has several strengths. We used a nationally representative dataset, applied biomarker-based exposure assessment, and employed advanced causal inference methods under the potential outcomes framework. The incorporation of a generalized random forest and causal distillation tree allowed us to estimate heterogeneous treatment effects and offer clinically interpretable insights into population subgroups most affected (Matsuyama et al 2024; Rehill 2024; Huang et al 2025). With application of NHANES weights, our estimates are generalizable to the US adult population, providing policy-relevant information (Burgos-Ochoa and Clouth 2025).

Several limitations warrant consideration. First, tooth loss is cumulative and biologically plausible as an outcome influenced by SHS; however, the cross-sectional design precludes establishing temporality, and reverse causation cannot be excluded. Second, despite the use of E values to evaluate robustness, residual confounding from unmeasured factors may persist. Third, dichotomizing serum cotinine may not capture cumulative exposure patterns relevant to tooth loss. Future studies employing panel data or longitudinal modified treatment policy strategies offer a more appropriate framework to characterize dose-response relationships and account for within-individual variation in exposure over time. Beyond the methodological limitations, the results of the present study were derived from the US population, which limits the generalizability to populations with distinctive smoking behaviors or from different oral health systems. Future studies are warranted to examine the external validity in other populations or combine results from various datasets with meta-analytic approaches.

Conclusion

This study provides evidence from causal inference methods that SHS exposure was associated with tooth loss among nonsmokers, and it highlights that individuals with disadvantaged socioeconomic status, higher sugar consumption, and comorbidities are particularly vulnerable. These findings reinforce the need for integrating dental outcomes into tobacco control policy making and prioritizing the susceptible subpopulations in public health interventions. Clinicians may consider assessing SHS exposure, particularly for vulnerable subgroups, in dental settings and mitigating such exposure as a supporting measure within oral health management.

Author Contributions

Y.Q. Chen, contributed to conception and design, data acquisition, analysis, and interpretation, drafted and critically revised manuscript; J.M. Zheng, contributed to design, data analysis and interpretation, critically revised the manuscript; Y.Z. Chen, H.Y. Zhang, contributed to design, data acquisition and interpretation, drafted the manuscript; J.Y. Li, H. Deng, contributed to data interpretation, drafted the manuscript; Y.Y. Kong, contributed to conception and design, data interpretation, critically revised the manuscript. All authors gave their final approval and agree to be accountable for all aspects of the work.

Supplemental Material

sj-docx-1-jdr-10.1177_00220345261419027 – Supplemental material for Average and Heterogeneous Associations between Secondhand Smoking and Tooth Loss

Supplemental material, sj-docx-1-jdr-10.1177_00220345261419027 for Average and Heterogeneous Associations between Secondhand Smoking and Tooth Loss by Y.Q. Chen, J.M. Zheng, Y.Z. Chen, H.Y. Zhang, J.Y. Li, H. Deng and Y.Y. Kong in Journal of Dental Research

Supplemental Material

sj-docx-2-jdr-10.1177_00220345261419027 – Supplemental material for Average and Heterogeneous Associations between Secondhand Smoking and Tooth Loss

Supplemental material, sj-docx-2-jdr-10.1177_00220345261419027 for Average and Heterogeneous Associations between Secondhand Smoking and Tooth Loss by Y.Q. Chen, J.M. Zheng, Y.Z. Chen, H.Y. Zhang, J.Y. Li, H. Deng and Y.Y. Kong in Journal of Dental Research

Footnotes

Acknowledgements

The authors acknowledge the endeavor of all participants in NHANES project. The authors thank Professor Aida Jun at the Institute of Science Tokyo for providing suggestions on the results interpretation. The authors thank Professor Jiang Hongbo at Guangdong Pharmaceutical University for his assistance in the implementation of statistical approaches.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported by the Guangdong Basic and Applied Basic Research Foundation Project (2022A1515110601).

ORCID iDs

Y.Q. Chen

Y.Y. Kong

Ethical Considerations

The NHANES project was approved by the Ethics Review Committee of the National Center for Health Statistics of the United States. All participants provided their informed consent.

Data Availability Statement

NHANES data are publicly available at .

A supplemental appendix to this article is available online.

References

Aida

. 2021. Tooth loss. In: Peres

Antunes

JLF

Watt

, editors. Oral epidemiology: a textbook on oral health conditions, research topics and methods. Springer International Publishing. p 223–233. https://doi.org/10.1007/978-3-030-50123-5_13

Al-Marzooq

, et al. 2022. Supragingival microbiome alternations as a consequence of smoking different tobacco types and its relation to dental caries. Sci Rep. 12(1):2861. https://doi.org/10.1038/s41598-022-06907-z

Andreyeva

Marple

Marinello

Moore

Powell

. 2022. Outcomes following taxation of sugar-sweetened beverages: a systematic review and meta-analysis. JAMA Netw Open. 5(6):e2215276. https://doi.org/10.1001/jamanetworkopen.2022.15276

Athey

Imbens

Pham

Wager

2017. Estimating average treatment effects: supplementary analyses and remaining challenges [preprint]. arXiv. https://doi.org/10.48550/arXiv.1702.01250

Athey

Tibshirani

Wager

2019. Generalized random forests. Ann Statist. 47(2):1148–1178. https://doi.org/10.1214/18-AOS1709

Athey

Wager

2019. Estimating treatment effects with causal forests: an application [preprint]. arXiv. https://doi.org/10.48550/arXiv.1902.07409

Bashir

Woolf

BAR

Burgess

Bernabé

2026. Periodontitis and systemic disease: the impact of covariate selection. J Dent Res. 105(1):59–66. https://doi.org/10.1177/00220345251356469

Baumeister

Listl

Holtfreter

Nascimento

Leite

FRM

. 2025. Causal effect of smoking and cessation on tooth loss. J Clin Periodontol. 52(4):539–546. https://doi.org/10.1111/jcpe.14106

Baumeister

S-E

Listl

Nascimento

Nolde

2025. Education, tooth loss, and dental visits: evidence from schooling reforms. J Dent Res. 104(5):489–494. https://doi.org/10.1177/00220345251316815

10.

Benowitz

. 1996. Cotinine as a biomarker of environmental tobacco smoke exposure. Epidemiol Rev. 18(2):188–204. https://doi.org/10.1093/oxfordjournals.epirev.a017925

11.

Bitencourt

, et al. 2023. Co-occurrence of periodontitis and diabetes-related complications. J Dent Res. 102(10):1088–1097. https://doi.org/10.1177/00220345231179897

12.

Bodnar

, et al. 2022. Use of a doubly robust machine-learning-based approach to evaluate body mass index as a modifier of the association between fruit and vegetable intake and preeclampsia. Am J Epidemiol. 191(8):1396–1406. https://doi.org/10.1093/aje/kwac062

13.

Brown

, et al. 2025. Development and validation of super learner models to predict small and large for gestational age in the second generation. Sci Rep. 15(1):33212. https://doi.org/10.1038/s41598-025-18466-0

14.

Burgos-Ochoa

Clouth

. 2025 Jul 14. Causal inference and survey data in paediatric epidemiology: generalising treatment effects from observational data. Paediatr Perinat Epidemiol. [published online ahead of print]. https://doi.org/10.1111/ppe.70042

15.

Cao

, et al. 2023. Serum uric acid to creatinine ratio and metabolic syndrome in middle-aged and elderly population: based on the 2015 CHARLS. Nutr Metab Cardiovasc Dis. 33(7):1339–1348. https://doi.org/10.1016/j.numecd.2023.05.004

16.

Cepeda

Weinstein

Blacketer

Lynch

. 2017. Association of flossing/inter-dental cleaning and periodontitis in adults. J Clin Periodontol. 44(9):866–871. https://doi.org/10.1111/jcpe.12765

17.

Chapple

ILC

, et al. 2015. Primary prevention of periodontitis: managing gingivitis. J Clin Periodontol. 42 Suppl 16:S71–S76. https://doi.org/10.1111/jcpe.12366

18.

Dai

, et al. 2022. Health effects associated with smoking: a burden of proof study. Nat Med. 28(10):2045–2055. https://doi.org/10.1038/s41591-022-01978-x

19.

Flor

, et al. 2024. Health effects associated with exposure to secondhand smoke: a burden of proof study. Nat Med. 30(1):149–167. https://doi.org/10.1038/s41591-023-02743-4

20.

, et al. 2024. Effects of comprehensive smoke-free legislation on smoking behaviours and macroeconomic outcomes in Shanghai, China: a difference-in-differences analysis and modelling study. Lancet Public Health. 9(12):e1037–e1046. https://doi.org/10.1016/s2468-2667(24)00262-7

21.

GBD 2017 Risk Factor Collaborators. 2018. Global, regional, and national comparative risk assessment of 84 behavioural, environmental and occupational, and metabolic risks or clusters of risks for 195 countries and territories, 1990–2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet. 392(10159):1923–1994. https://doi.org/10.1016/s0140-6736(18)32225-6

22.

Ghanbari-Jahromi

Bastani

Jalali

Delavari

2023. Factors affecting oral and dental services’ utilization among elderly: a scoping review. BMC Oral Health. 23(1):597. https://doi.org/10.1186/s12903-023-03285-4

23.

Golpasand Hagh

Zakavi

Ansarifar

Ghasemzadeh

Solgi

. 2013. Association of dental caries and salivary sIgA with tobacco smoking. Aust Dent J. 58(2):219–223. https://doi.org/10.1111/adj.12059

24.

Goniewicz

, et al. 2009. Elimination kinetics of the tobacco-specific biomarker and lung carcinogen 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanol. Cancer Epidemiol Biomarkers Prev. 18(12):3421–3425. https://doi.org/10.1158/1055-9965.epi-09-0874

25.

González-Valero

, et al. 2018. Association between passive tobacco exposure and caries in children and adolescents: a systematic review and meta-analysis. PLoS One. 13(8):e0202497. https://doi.org/10.1371/journal.pone.0202497

26.

Gruber

Laan

MJVD

. 2012. tmle: an R package for targeted maximum likelihood estimation. J Stat Soft. 51(13):1–35. https://doi.org/10.18637/jss.v051.i13

27.

Hajishengallis

Chavakis

2021. Local and systemic mechanisms linking periodontal disease and inflammatory comorbidities. Nat Rev Immunol. 21(7):426–440. https://doi.org/10.1038/s41577-020-00488-6

28.

Hernan

Robins

. 2025. Causal inference: what if. CRC Press.

29.

Huang

Tang

Kenney

. 2025. Distilling heterogeneous treatment effects: stable subgroup estimation in causal inference [preprint]. arXiv. https://doi.org/10.48550/arXiv.2502.07275

30.

Kocher

König

Borgnakke

Pink

Meisel

2018. Periodontal complications of hyperglycemia/diabetes mellitus: epidemiologic complexity and clinical challenge. Periodontol 2000. 78(1):59–97. https://doi.org/10.1111/prd.12235

31.

Kusama

, et al. 2023. Weight loss mediated the relationship between tooth loss and mortality risk. J Dent Res. 102(1):45–52. https://doi.org/10.1177/00220345221120642

32.

Lopez-Ayala

Riley

Collins

Zimmermann

2025. Dealing with continuous variables and modelling non-linear associations in healthcare data: practical guide. BMJ. 390:e082440. https://doi.org/10.1136/bmj-2024-082440

33.

Luo

Sun

Chan

SSC

Chu

CH.

2025. Secondary smoking and early childhood caries: a systematic review and meta-analysis. Int Dent J. 75(1):23–31. https://doi.org/10.1016/j.identj.2024.08.013

34.

Matsuyama

Aida

Kondo

Shiba

2024. Heterogeneous association of tooth loss with functional limitations. J Dent Res. 103(4):369–377. https://doi.org/10.1177/00220345241226957

35.

Matsuyama

, et al. 2025. Water fluoridation as a population strategy for reducing oral health inequalities: high-dimensional effect heterogeneity analysis using machine learning. Int J Epidemiol. 54(4):dyaf080. https://doi.org/10.1093/ije/dyaf080

36.

Matsuyama

Jürges

Listl

2023. Causal effect of tooth loss on cardiovascular diseases. J Dent Res. 102(1):37–44. https://doi.org/10.1177/00220345221120164

37.

Mbulo

, et al. 2016. Secondhand smoke exposure at home among one billion children in 21 countries: findings from the Global Adult Tobacco Survey (GATS). Tob Control. 25(e2):e95–e100. https://doi.org/10.1136/tobaccocontrol-2015-052693

38.

Moccia

, et al. 2024. Machine learning in causal inference for epidemiology. Eur J Epidemiol. 39(10):1097–1108. https://doi.org/10.1007/s10654-024-01173-x

39.

Neophytou

, et al. 2024. Bridging differences in cohort analyses of the relationship between secondhand smoke exposure during pregnancy and birth weight: the transportability framework in the ECHO program. Environ Health Perspect. 132(5):57007. https://doi.org/10.1289/ehp13961

40.

Nugent

Marquez

Charlebois

Abbott

Balzer

. 2024. Blurring cluster randomized trials and observational studies: two-stage TMLE for subsampling, missingness, and few independent units. Biostatistics. 25(3):599–616. https://doi.org/10.1093/biostatistics/kxad015

41.

Oliveira

MDM

Ardenghi

Zanatta

. 2022. Is secondhand smoke exposure associated with poor periodontal status in children and adolescents? A systematic review and meta-analysis. Eur Arch Paediatr Dent. 23(4):513–525. https://doi.org/10.1007/s40368-022-00709-7

42.

Polley

LeDell

Kennedy

Lendle

van der Laan

2024. SuperLearner: Super Learner Prediction. R package version 2.0-40. [accessed 2025 Sept 7]. https://cran.r-project.org/web/packages/SuperLearner/index.html

43.

Rehill

. 2024. Distilling interpretable causal trees from causal forests [preprint]. arXiv. http://arxiv.org/abs/2408.01023

44.

Sanders

Slade

2013. State cigarette excise tax, secondhand smoke exposure, and periodontitis in US nonsmokers. Am J Public Health. 103(4):740–746. https://doi.org/10.2105/ajph.2011.300579

45.

Schuler

Rose

2017. Targeted maximum likelihood estimation for causal inference in observational studies. Am J Epidemiol. 185(1):65–73. https://doi.org/10.1093/aje/kww165

46.

Tang

Ishwaran

2017. Random forest missing data algorithms. Stat Anal Data Min. 10(6):363–377. https://doi.org/10.1002/sam.11348

47.

Thomson

, et al. 2018. The effects of public health policies on health inequalities in high-income countries: an umbrella review. BMC Public Health. 18(1):869. https://doi.org/10.1186/s12889-018-5677-1

48.

Tibbe

Montoya

. 2022. Correcting the bias correction for the bootstrap confidence interval in mediation analysis. Front Psychol. 13:810258. https://doi.org/10.3389/fpsyg.2022.810258

49.

Tsai

, et al. 2021. Trends in secondhand smoke exposure, 2011–2018: impact and implications of expanding serum cotinine range. Am J Prev Med. 61(3):e109–e117. https://doi.org/10.1016/j.amepre.2021.04.004

50.

Umemori

, et al. 2020. Does second-hand smoke associate with tooth loss among older Japanese? JAGES cross-sectional study. Int Dent J. 70(5):388–395. https://doi.org/10.1111/idj.12577

51.

van der Laan

, et al. 2023. Welcome. Targeted learning in R. tlverse; [accessed 2025 Nov 29]. https://tlverse.org/tlverse-handbook/

52.

VanderWeele

Ding

2017. Sensitivity analysis in observational research: introducing the E-value. Ann Intern Med. 167(4):268–274. https://doi.org/10.7326/m16-2607

53.

Wang

Matsuyama

Kiuchi

Kusama

Aida

2026. Oral frailty as a predictor for weight loss in older Japanese: a cohort study. J Dent Res. 105(1):87–94. https://doi.org/10.1177/00220345251356437

54.

Zelig

, et al. 2022. Tooth loss and nutritional status in older adults: a systematic review and meta-analysis. JDR Clin Trans Res. 7(1):4–15. https://doi.org/10.1177/2380084420981016