Case-control study of cumulative cigarette tar exposure and lung and upper aerodigestive tract cancers

Abstract

The development of comprehensive measures for tobacco exposure is crucial to specify effects on disease and inform public health policy. In this population-based case-control study, we evaluated the associations between cumulative lifetime cigarette tar exposure and cancers of the lung and upper aerodigestive tract (UADT). The study included 611 incident cases of lung cancer; 601 cases of UADT cancers (oropharyngeal, laryngeal and esophageal cancers); and 1,040 cancer-free controls. We estimated lifetime exposure to cigarette tar based on tar concentrations abstracted from government cigarette records and self-reported smoking histories derived from a standardized questionnaire. We analyzed the associations for cumulative tar exposure with lung and UADT cancer, overall and according to histological subtype. Cumulative tar exposure was highly correlated with pack-years among ever smoking controls (Pearson coefficient = 0.90). The adjusted odds ratio (95% confidence limits) for the estimated effect of about 1 kg increase in tar exposure (approximately the interquartile range in all controls) was 1.61 (1.50, 1.73) for lung cancer and 1.21 (1.13, 1.29) for UADT cancers. In general, tar exposure was more highly associated with small, squamous and large cell lung cancer than adenocarcinoma. With additional adjustment for pack-years, positive associations between tar and lung cancer were evident, particularly for small cell and large cell subtypes. Therefore, incorporating the composition of tobacco carcinogens in lifetime smoking exposure may improve lung cancer risk estimation. This study does not support the claim of a null or inverse association between “low exposure” to tobacco smoke and risk of these cancer types.

Tobacco smoking has been identified as a causal factor for 15 organ sites, including the lung and upper aerodigestive tract (UADT).¹ In addition, smoking is associated with all major histological subtypes of lung cancer, although a higher association has been reported for small cell cancer and squamous cell carcinoma than for large cell lung cancer and adenocarcinoma.^2,3 With respect to UADT cancer, smoking is associated with squamous cell carcinoma of the head and neck, and with both squamous cell carcinoma and adenocarcinoma of the esophagus.¹ Tobacco smoke is a complex mixture of over 7,000 compounds, of which 81 are considered carcinogenic or potentially carcinogenic in humans.^4–8 “Tar” is a common term for the total particulate matter in tobacco smoke— excluding nicotine and water—that contains these putative carcinogens such as benzo[a]pyrene.⁴ While standard measures of tobacco exposure (e.g., pack-years) treat tobacco smoke as homogeneous, emissions have been shown to vary not only between filtered and unfiltered cigarettes but also between brands of each type.^4,9–11

Reviewing the extensive literature on the relationship between cigarette tar content and health, reports have concluded that “low-tar” cigarettes do not reduce risk for lung cancer and should not be recommended as healthy alternatives.^11,12 Only seven studies have investigated the association between cancer and cumulative tar exposure, an index accounting for changing smoking behaviors over time as well as tar content for different brands.^13–19 Positive associations with cumulative tar exposure were reported for lung,^13,15,16,19 pancreatic¹⁴ and oral cancer,¹⁷ but not for bladder cancer.¹⁸ Limitations of these prior studies include very few years of measured tar content,^13–18 hospital-based control selection^13,15–17 and limited or no analysis by histological subtype for lung cancer.^13,15,16,19 In addition, only two studies have adjusted for other measures of tobacco exposure.^13,18 In this study, we modified and applied the cumulative tar index to evaluate associations with cancers of the lung and UADT in a population-based case-control study conducted in Los Angeles County. In addition, we compared the associations with these cancers between cumulative tar and pack-years, as well as between histological subtypes. Furthermore, we measured the associations for cumulative tar and cancer after adjusting for pack-years to evaluate this additional information on cigarette composition.

Material and Methods

Study design and population

Investigators conducted a population-based case-control study of lung and UADT cancers in Los Angeles County from 1999 to 2004. The Institutional Review Boards of the University of California, Los Angeles (UCLA) and the University of Southern California (USC) approved the study, and all participants provided their written informed consent. Further details of the original study design are available in earlier references.^20,21 In brief, newly diagnosed lung and UADT cancer patients were recruited from the USC Cancer Surveillance Program for Los Angeles County (USC CSP), a National Cancer Institute Surveillance, Epidemiology and End-Results (SEER) Program cancer registry, through a rapid ascertainment system. Participants met the following inclusion criteria: (1) Residence in Los Angeles County at the time of diagnosis; (2) diagnosis age of 18–65 during the study period; (3) either English or Spanish speaking or accompanied by a translator during the interview. Among eligible patients, the recruitment rates for cases were 39% (611 of 1,556) for lung and 46% (601 of 1,301) for UADT cancer cases. The USC CSP collects pathology reports (over 95% of patients) and other diagnostic methods including magnetic resonance imaging and computed tomography scan with cancer reporting. In addition, the USC CSP classifies cancer diagnoses according to the International Classification of Diseases for Oncology, Third Edition (ICD-O-3). Among the 601 recruited UADT cases, there were 497 (82.7%) patients with squamous cell carcinoma of the oropharynx, larynx and esophagus. In addition, 74 UADT patients were diagnosed with adenocarcinoma, all confined to the esophagus. The 611 lung cancer cases consisted of 508 (83.1%) patients with non-small cell lung cancer and 75 patients with small cell lung cancer. Non-small cell lung cancer includes adenocarcinoma (n = 290), squamous cell carcinoma (n = 95) and large cell carcinoma (n = 115). Neighborhood-ascertained controls were matched to cases on sex and age (within ten-years) and had a 79% recruitment rate among identified eligible matches.

Research staff interviewed each participant in-person using a standardized questionnaire. Questionnaire items included demographic characteristics; lifetime history of exposure to tobacco, alcohol, marijuana and other recreational drugs; medical and occupational histories; and family history of cancer. Cigarette smoking information was collected on a yearly basis, including age at starting and quitting, brand and sub-brand details, number and frequency (i.e., cigarettes smoked per day/week/month/year), usual length of unsmoked cigarette (explained below) and smoke inhalation depth (deep, moderate, shallow, did not inhale). Participants also reported details for the lifetime use of cigars, pipes, chewing tobacco and snuff.

Exposure estimation

We defined “ever-smokers” as participants who smoked at least 100 cigarettes in their lifetime. To estimate their cumulative tar exposure, we first created a historical database of machine-measure tar yields from 39 reports of the Federal Trade Commission (FTC) between 1967 and 2000. We ascertained these reports from the University of California, San Francisco online archive, the Truth Tobacco Industry Documents, formerly known as the Legacy Tobacco Documents Library.²² The FTC started collecting these ratings in 1967 according to the standardized machine smoking protocol of the Cambridge filter method.²³ Next, we used this longitudinal database to estimate cumulative tar exposure for all smoking participants. For each reported sub-brand in the questionnaire, we identified the closest match from the FTC report with respect to calendar year, size (Regular, King, 100 mm, 120 mm), design (Filter/Non-Filter), additive (Menthol/Non-menthol) and flavor (Full flavor, Light, Ultra-light). We calculated the average values for the ratings of multiple matches in the FTC report and for reports covering the same testing period. Missing tar ratings for years of reported exposure were imputed with the most recent rating in the database. For example, for pre-1967 smoking histories, we imputed tar ratings with values from 1967, when reporting began. Then, we modified Zang and Wynder’s cumulative exposure index for tar by accounting for cigarette portion size and tar ratings by calendar year.¹³ We have reproduced the original index below:

$$T={\displaystyle \sum _{i=1}^B(t_i\ast D_i\ast C_i)}\ast {10}^{-6}$$ (1)

where T is cumulative tar exposure (in kg), t is tar level per cigarette sub-brand (mg), D is days of smoking, C is cigarettes smoked per day and B is all of the cigarette sub-brands smoked during the participant’s lifetime. We summed tar exposure per year across all years of smoking to estimate cumulative tar. Study participants reported the portion of the unsmoked cigarette including the butt as “less than one-quarter,” “about one-quarter,” “about one-third” and “about one-half or more,” which we specified as consumed portions of 7/8, 3/4, 2/3 and 1/4, respectively. While marijuana smoke also contains tar with many of the same components that are found in tobacco tar, including procarcinogenic polycyclic aromatic hydrocarbons,²⁴ we have not observed clear associations between marijuana smoking and cancer risk in this study population²¹ and did not estimate tar exposure from this source.

Statistical analysis

First, we calculated pack-years of cigarette smoking, cumulative tar exposure and drink-years of alcohol consumption, lagged one year before the diagnosis year or reference year for controls. Then, we estimated the associations of cumulative tar exposure on risk for cancers of the lung and UADT in continuous and categorical analyses. The continuous measure was one interquartile range (IQR) increase in cumulative tar exposure; the categorical analysis used never smokers as the reference group and tertiles of exposure in ever smokers. Both the IQR and tertiles were based on tar distribution in the controls. We also analyzed associations with pack-years to compare cumulative tar exposure with the conventional measure of cumulative tobacco exposure. We used unconditional logistic regression to estimate odds ratios (OR’s) and 95% confidence limits (CL’s), adjusting for potential covariates: age and sex (the matching variables), race/ethnicity, education level and alcohol drink-years. In addition, we repeated the analyses for histological subtypes in both cancer groups. For lung cancer, we analyzed the common subtypes: squamous cell, small cell, large cell and adenocarcinoma. For UADT cancer, we separately analyzed squamous cell carcinoma and esophageal adenocarcinoma. The logistic regression equation takes the general form:

$$Y={\mathrm{\beta }}_0+{\mathrm{\beta }}_1{X}_1+\dots +{\mathrm{\beta }}_p{X}_P$$ (2)

where Y is the natural log odds of disease status (case vs. control), β₀ is the intercept when all predictors are zero and β_i is the regression coefficient for each predictor i multiplied by some value X of the predictor. Furthermore, β_i is the natural logarithm of the odds ratio for the association between disease status with a unit increase in the covariate. Each coefficient is conditional on other coefficients in the model. Models estimating the association for cumulative tar exposure and each cancer subtype included beta coefficients for cumulative tar exposure, age, sex, race/ethnicity, education level and alcohol drink-years. We also modeled the comprehensive smoking index (CSI), a function of smoking duration, intensity and time since cessation, including a half-life parameter (τ) of 10 years for the smoking effect.²⁵ We calculated the area under the curve (AUC) for the receiver operating characteristic (ROC) to compare models for different measures of smoking exposure, including the linear combination of pack-years and tar (the sum of each coefficient multiplied by the record value). Furthermore, we evaluated the modification of the association between cumulative tar exposure and cancer by race/ethnicity and smoking status (current/former). We tested for a residual association of cumulative tar exposure in models adjusted for key covariates as well as pack-years (an additional beta coefficient in the logistic regression model) and according to subtype. In order to correct for potential false-positive findings, we re-ran these adjusted models using semi-Bayes “shrinkage” estimation.^26–29 In this analysis, prior coefficients with null associations are updated with coefficients from observed data to shrink associations in the logistic regression model toward the null. We assigned independent normal priors for targeted coefficients of tar exposure and cancer risk, with mean zero and variance 0.5 (corresponding to OR = 1, 95% prior limits= 0.25, 4). Then, we combined the prior data with the observed data to calculate posterior estimates and 95% posterior limits. We performed our analyses using SAS version 9.4 (SAS Institute, Cary, NC).

Results

Distributions of socioemographic characteristics, cigarette smoking and alcohol consumption are presented in Table 1. Age and sex are matched overall but lung and UADT cancers have different distributions. Sixty-five percent of all study participants reported smoking at least 100 cigarettes in their lifetime. As expected, we observed positive associations between cigarette pack-years and risk of both lung and UADT cancers, stronger for lung than for UADT cancer. We observed positive associations between former versus never smoking and both cancer types. For current smoking, we observed a positive association for lung cancer and an inverse association for UADT cancer (OR = 0.59, 95% CL’s = 0.41, 0.85). The association for former smoking was higher than current smoking for both cancer types, possibly due to the induction time between smoking exposure and cancer onset. In addition, we observed a positive association between the highest level of drink-years (>40 drink-years vs. never drinkers) and UADT cancer (OR = 1.78, 95% CL’s = 1.26, 2.51) after adjusting for covariates including cigarette pack-years. The test for trend across categories of drink-years also suggested a positive association (p-trend = 0.0007), consistent with published reports.³⁰ We did not observe associations for consumption of cigars, pipes, snuff, marijuana or chewing tobacco or for passive smoking duration in either cancer type probably due to sparse exposure data (data not shown).

Table 1.

Distributions of sociodemographic and consumption characteristics among lung cancer cases (n = 611) and UADT cases (n = 601) compared to cancer-free controls (n = 1,040)

Variable	Controls N (%)	Lung N (%)	Adjusted OR (95% CL’s)	UADT N (%)	Adjusted OR (95% CL’s)
Age1
17–34	51 (4.9)	4 (0.7)	–	32 (5.3)	–
35–44	170 (16.3)	57 (9.3)	–	77 (12.8)	–
45–54	500 (48.1)	301 (49.3)	–	267 (44.4)	–
>54	319 (30.7)	249 (40.8)	–	225 (37.4)	–
Sex1
Male	623 (59.9)	303 (49.6)	–	454 (75.5)	–
Female	417 (40.1)	308 (50.4)	–	147 (24.5)	–
Race/Ethnicity2
Caucasian	634 (61)	359 (58.9)	1 (Reference)	341 (56.9)	1 (Reference)
African–American	102 (9.8)	96 (15.7)	1.99 (1.38, 2.89)	69 (11.5)	1.05 (0.72, 1.53)
Hispanic	204 (19.6)	70 (11.5)	0.95 (0.62, 1.45)	109 (18.2)	0.70 (0.48, 1.00)
Asian/Pacific-Islander	62 (6.0)	70 (11.5)	4.70 (3.05, 7.22)	64 (10.7)	2.71 (1.80, 4.09)
Other	37 (3.6)	15 (2.5)	0.67 (0.32, 1.38)	16 (2.7)	0.63 (0.32, 1.23)
Education (years of schooling)2
<12	116 (11.2)	107 (17.5)	1 (Reference)	126 (21.0)	1 (Reference)
12	184 (17.7)	158 (25.9)	0.57 (0.36, 0.89)	147 (24.5)	0.60 (0.40, 0.90)
13–15	272 (26.2)	186 (30.4)	0.56 (0.36, 0.88)	156 (26.0)	0.49 (0.33, 0.73)
16	209 (20.1)	89 (14.6)	0.49 (0.30, 0.80)	103 (17.1)	0.43 (0.28, 0.67)
>16	258 (24.8)	71 (11.6)	0.37 (0.22, 0.61)	69 (11.5)	0.26 (0.17, 0.42)
			p-trend=0.0004		p-trend <0.0001
Cigarette smoking status3
Never	491 (47.2)	110 (18.0)	1 (Reference)	182 (30.3)	1 (Reference)
Former	371 (35.7)	390 (63.8)	4.34 (3.27, 5.75)	338 (56.2)	1.71 (1.32, 2.21)
Current	177 (17.1)	111 (18.2)	2.30 (1.60, 3.30)	81 (13.5)	0.59 (0.41, 0.85)
Cigarette pack-years3
Never Smokers	491 (47.2)	110 (18.0)	1 (Reference)	182 (30.3)	1 (Reference)
20	355 (34.1)	105 (17.2)	1.41 (1.02, 1.96)	150 (25.0)	0.96 (0.73, 1.27)
>20–40	137 (13.2)	213 (34.9)	8.36 (5.86, 11.92)	147 (24.5)	1.88 (1.34, 2.64)
>40	56 (5.4)	183 (30.0)	21.59 (13.85, 33.66)	122 (20.3)	3.37 (2.21, 5.14)
			p-trend < 0.0001		p-trend <0.0001
Alcohol drink-years4
Never Drinkers	264 (25.4)	170 (27.8)	1 (Reference)	117 (19.5)	1 (Reference)
40	586 (56.3)	260 (42.6)	0.67 (0.49, 0.90)	232 (38.6)	0.94 (0.70, 1.27)
>40	189 (18.2)	180 (29.5)	0.77 (0.52, 1.14)	250 (41.6)	1.78 (1.26, 2.51)
			p-trend = 0.14		p-trend=0.0007

¹Age and sex are matching variables and their odds ratios are not valid.

²Models for race/ethnicity and education adjusted for each other, plus age, sex, pack-years and drink-years.

³Models adjusted for age, race/ethnicity, sex, education and drink-years.

⁴Models adjusted for age, race/ethnicity, sex, education and smoking pack-years.

Distributions of cumulative tar exposure and pack-years for both cancer types, as well as controls, are shown in Table 2. Overall mean (standard deviation) tar exposure (in kg) was 0.86 (1.58) in controls, 2.90 (2.86) in lung cancer cases and 2.19 (2.78) in UADT cancer cases. Corresponding statistics for pack-years were 9.09 (15.39), 30.50 (24.33) and 21.88 (23.69). Point estimates for the mean of both measures were higher in men than women for all three groups. Cumulative tar exposure was highly correlated with pack-years, based on Pearson r = 0.90 in ever smoking controls. Nearly 25% of the 1,468 smokers reported a brand that was unknown or unlisted in the FTC Reports; 76 smokers had completely unknown/unlisted brand information; 5 smokers had missing information for portion size (length of unsmoked cigarette).

Table 2.

Distributions of cumulative tar and pack-years for cancer cases and cancer-free controls, stratified by sex

Variable	Controls					Lung cancer		UADT cancer
	N (%)	Mean (SD)	IQR	Smokers		N (%)	Mean (SD)	N (%)	Mean (SD)
				Tertile 1	Tertile 2
Cumulative tar (kg)
Overall	997 (95.9)	0.86 (1.58)	0.96	0.43	2.08	598 (97.9)	2.90 (2.86)	575 (95.7)	2.19 (2.78)
Men	589 (56.6)	1.06 (1.77)	1.55	0.55	2.48	297 (48.6)	3.73 (3.17)	435 (72.4)	2.50 (2.89)
Women	408 (39.2)	0.58 (1.21)	0.53	0.30	1.45	301 (49.3)	2.09 (2.25)	140 (23.3)	1.24 (2.17)
Missing	43 (4.1)					13 (2.1)		26 (4.3)
Cigarette pack-years
Overall	1039 (99.9)	9.09 (15.39)	12.81	5.25	21.00	611 (100)	30.50 (24.33)	601 (100)	21.88 (23.69)
Men	622 (59.8)	10.68 (16.46)	17.00	6.50	25.29	303 (49.6)	36.90 (25.27)	454 (75.5)	24.29 (23.66)
Women	417 (40.1)	6.73 (13.31)	6.47	3.75	17.10	308 (50.4)	24.20 (21.63)	147 (24.5)	14.43 (22.26)
Missing	1 (0.1)					0		0

Tables (3–5) display adjusted odds ratios (OR’s) and 95% confidence limits (CL’s) for associations of lung and UADT cancers, overall and by histological subtype, with cumulative tar exposure and pack-years. Models were adjusted for age (in fine categories), race/ethnicity, sex, years of education and drink-years. Positive associations were evident in both cancer types and corresponding subtypes, by trend tests (all p < 0.05) and by OR estimates for categorical and continuous analyses. OR’s for one-IQR increase of tar exposure (0.96 kg) and pack-years (12.81) did not vary much by subtype within each cancer group. For lung cancer and subtypes (Table 3: overall, small cell, squamous cell; Table 4: adenocarcinoma, large cell), the OR estimates for pack-years were greater than tar exposure by one-IQR increase (overall: pack-years- OR= 2.16, 95% CL’s= 1.96, 2.39; tar- OR= 1.61, 95% CL’s= 1.50, 1.73). However, the confidence intervals for corresponding tertiles of smoking exposures overlapped, suggesting no obvious difference. Compared to never smokers, the second and third tertiles of exposure were associated with overall lung cancer and with the major histological subtypes. For squamous cell lung cancer, the second tertile excluded the null for tar (OR= 5.09, 95% CL’s= 2.08, 12.41) but not for pack-years (OR= 2.49, 95% CL’s= 0.93, 6.65). Furthermore, associations in the higher tertiles for tar and pack-years were generally higher for small cell, squamous and large cell carcinoma of the lung than for lung adenocarcinoma.

Table 3.

Cumulative tar and pack-years and risk of overall, small cell and squamous cell lung cancer

Variable	Controls N (%)	Lung N (%)	Adjusted OR (95% CL’s)1	Small cell N (%)	Adjusted OR (95% CL’s)1	Squamous N (%)	Adjusted OR (95% CL’s)1
Cumulative tar
Per IQR Increase			1.61 (1.50, 1.73)		1.63 (1.44, 1.84)		1.71 (1.51, 1.92)
Never Smokers	491 (47.2)	110 (18)	1 (Reference)	4 (5.3)	1 (Reference)	8 (8.4)	1 (Reference)
Tertile 1	169 (16.3)	34 (5.6)	0.95 (0.60, 1.49)	2 (2.7)	1.70 (0.29, 9.80)	3 (3.2)	0.93 (0.24, 3.65)
Tertile 2	169 (16.3)	130 (21.3)	3.48 (2.47, 4.90)	18 (24)	17.63 (5.22, 59.61)	18 (18.9)	5.09 (2.08, 12.41)
Tertile 3	168 (16.2)	324 (53)	10.43 (7.34, 14.81)	49 (65.3)	45.97 (13.68, 154.48)	63 (66.3)	19.00 (8.02, 45.03)
			p-trend<0.0001		p-trend<0.0001		p-trend<0.0001
Cigarette pack-years
Per IQR Increase			2.16 (1.96, 2.39)		2.26 (1.89, 2.70)		2.41 (2.03, 2.88)
Never Smokers	491 (47.2)	110 (18)	1 (Reference)	4 (5.3)	1 (Reference)	8 (8.4)	1 (Reference)
Tertile 1	183 (17.6)	27 (4.4)	0.75 (0.47, 1.21)	1 (1.3)	0.86 (0.09, 8.06)	3 (3.2)	1.06 (0.27, 4.15)
Tertile 2	183 (17.6)	86 (14.1)	2.14 (1.49, 3.07)	12 (16)	10.58 (3.14, 35.58)	10 (10.5)	2.49 (0.93, 6.65)
Tertile 3	182 (17.5)	388 (63.5)	11.82 (8.42, 16.61)	58 (77.3)	54.30 (16.76, 175.96)	74 (77.9)	22.68 (9.79, 52.57)
			p-trend<0.0001		p-trend<0.0001		p-trend<0.0001

¹Models adjusted for age, race/ethnicity, sex, years of education and drink-years.

Table 5.

Cumulative tar and pack-years and risk of overall UADT cancer, UADT squamous cell carcinoma and esophageal adenocarcinoma

Variable	Controls N (%)	UADT N (%)	Adjusted OR (95% CL’s)1	Squamous N (%)	Adjusted OR (95% CL’s)1	Esophageal adenocarcinoma N (%)	Adjusted OR (95% CL’s)1
Cumulative tar
Per IQR Increase			1.21 (1.13, 1.29)		1.18 (1.10, 1.27)		1.27 (1.13, 1.43)
Never smokers	491 (47.2)	182 (30.3)	1 (Reference)	149 (30)	1 (Reference)	18 (24.3)	1 (Reference)
Tertile 1	169 (16.3)	50 (8.3)	0.70 (0.48, 1.03)	41 (8.2)	0.69 (0.46, 1.05)	6 (8.1)	0.82 (0.31, 2.19)
Tertile 2	169 (16.3)	122 (20.3)	1.53 (1.11, 2.10)	98 (19.7)	1.43 (1.02, 2.01)	17 (23.0)	2.52 (1.21, 5.25)
Tertile 3	168 (16.2)	221 (36.8)	1.97 (1.42, 2.74)	185 (37.2)	1.79 (1.26, 2.54)	31 (41.9)	3.54 (1.69, 7.43)
			p-trend<0.0001		p-trend=0.0005		p-trend=0.0003
Cigarette pack-years
Per IQR Increase			1.36 (1.25, 1.49)		1.33 (1.21, 1.46)		1.46 (1.24, 1.73)
Never smokers	491 (47.2)	182 (30.3)	1 (Reference)	149 (30)	1 (Reference)	18 (24.3)	1 (Reference)
Tertile 1	183 (17.6)	50 (8.3)	0.69 (0.48, 1.01)	41 (8.2)	0.69 (0.46, 1.04)	7 (9.5)	0.86 (0.34, 2.19)
Tertile 2	183 (17.6)	106 (17.6)	1.24 (0.90, 1.70)	85 (17.1)	1.18 (0.84, 1.67)	14 (18.9)	1.83 (0.85, 3.91)
Tertile 3	182 (17.5)	263 (43.8)	2.39 (1.75, 3.26)	222 (44.7)	2.22 (1.59, 3.09)	35 (47.3)	3.69 (1.82, 7.52)
			p-trend<0.0001		p-trend<0.0001		p-trend=0.0002

¹Models adjusted for age, race/ethnicity, sex, years of education and drink-years.

Table 4.

Cumulative tar and pack-years and risk of lung adenocarcinoma and large cell lung cancer

Variable	Controls N (%)	Adenocarcinoma N (%)	Adjusted OR (95% CL’s)1	Large cell N (%)	Adjusted OR (95% CL’s)1
Cumulative tar
Per IQR Increase			1.42 (1.30, 1.54)		1.62 (1.45, 1.81)
Never smokers	491 (47.2)	77 (26.6)	1 (Reference)	14 (12.2)	1 (Reference)
Tertile 1	169 (16.3)	17 (5.9)	0.69 (0.38, 1.23)	5 (1.7)	1.12 (0.38, 3.27)
Tertile 2	169 (16.3)	70 (24.1)	2.64 (1.76, 3.96)	19 (6.6)	4.02 (1.88, 8.60)
Tertile 3	168 (16.2)	121 (41.7)	5.71 (3.75, 8.70)	75 (25.9)	17.25 (8.48, 35.10)
			p-trend<0.0001		p-trend<0.0001
Cigarette pack-years
Per IQR Increase			1.77 (1.58, 1.98)		2.18 (1.86, 2.55)
Never smokers	491 (47.2)	77 (26.6)	1 (Reference)	14 (12.2)	1 (Reference)
Tertile 1	183 (17.6)	15 (5.2)	0.59 (0.32, 1.08)	3 (2.6)	0.68 (0.19, 2.43)
Tertile 2	183 (17.6)	46 (15.9)	1.66 (1.07, 2.58)	13 (11.3)	2.47 (1.09, 5.56)
Tertile 3	182 (17.5)	152 (52.4)	6.56 (4.39, 9.81)	85 (73.9)	18.16 (9.11, 36.22)
			p-trend<0.0001		p-trend<0.0001

¹Models adjusted for age, race/ethnicity, sex, years of education and drink-years.

For overall UADT cancer (Table 5), estimated associations for tar and pack-years were generally less than for lung cancer (one IQR increased tar exposure- OR= 1.21, 95% CL’s= 1.13, 1.29). With respect to overall disease and subtypes, the third tertiles for tar and pack-years were associated with increased cancer risk. The second tertile for tar exposure, not pack-years, was associated with overall disease (OR= 1.53, 95% CL’s= 1.11, 2.10), UADT squamous cell carcinoma (OR= 1.43, 95% CL’s= 1.02, 2.01) and esophageal adenocarcinoma (OR = 2.52, 95% CL’s= 1.21, 5.25). The subtypes did not appear to differ by tertiles of tar or pack-years. For the ROC analysis, the area under the curve (AUC) was lower for cumulative tar compared to pack-years, except for esophageal adenocarcinoma (p > 0.05; Supporting Information Table 1). The AUC’s for the CSI were higher than pack-years for overall lung (77.3% vs. 76.7%) and UADT cancers (66.6% vs. 66.1%). However, the AUC’s for combined cumulative tar and pack-years were no different than pack-years alone.

Table 6 displays estimates for associations between cumulative tar exposure and cancer, with additional adjustment for pack-years in maximum-likelihood and semi-Bayes corrected models. Associations with lung cancer were evident in the second exposure tertiles even after semi-Bayes adjustment: overall lung cancer (semi-Bayes odds ratio-SBOR = 1.55, 95% posterior limits-PL’s = 1.07, 2.24); small cell (SBOR = 2.73, 95% PL’s = 1.33, 5.61); adenocarcinoma (SBOR = 1.56, 95% PL’s =1.02, 2.37); and in the third tertile of large cell lung cancer (SBOR= 2.51, 95% PL’s = 1.20, 5.25). We also observed positive trends for cumulative tar in small cell and large cell lung cancer in these models (p-trend < 0.05). However, we did not observe associations between tar and cancer by per-IQR increase in exposure. Moreover, we did not observe positive associations for cumulative tar exposure in UADT cancer or subtypes after adjusting for pack-years in maximum likelihood or semi-Bayes models.

Table 6.

Cumulative tar exposure and risk of lung and UADT cancer and subtypes, adjusted for pack-years in maximum-likelihood and semi-Bayes models

Cancer type	Maximum-likelihood adjusted OR (95% CL’s)1	Semi-Bayes adjusted OR (95% posterior limits)1
Overall lung cancer
Per IQR Increase	0.99 (0.86, 1.13)	0.96 (0.84, 1.10)
Never smokers	1 (Reference)	1 (Reference)
Tertile 1	0.73 (0.46, 1.17)	0.73 (0.47, 1.12)
Tertile 2	1.47 (0.98, 2.21)	1.55 (1.07, 2.24)
Tertile 3	1.56 (0.88, 2.76)	1.47 (0.88, 2.45)
	p-trend = 0.09	p-trend=0.05
Small cell lung cancer
Per IQR Increase	0.93 (0.73, 1.18)	0.93 (0.74, 1.18)
Never Smokers	1 (Reference)	1 (Reference)
Tertile 1	1.48 (0.25, 8.61)	0.82 (0.30, 2.22)
Tertile 2	8.79 (2.47, 31.37)	2.73 (1.33, 5.61)
Tertile 3	7.61 (1.81, 32.08)	2.06 (0.88, 4.82)
	p-trend= 0.002	p-trend=0.004
Squamous cell lung cancer
Per IQR Increase	1.12 (0.92, 1.36)	1.07 (0.89, 1.29)
Never Smokers	1 (Reference)	1 (Reference)
Tertile 1	0.52 (0.11, 2.44)	0.67 (0.26, 1.73)
Tertile 2	1.66 (0.62, 4.44)	1.59 (0.78, 3.21)
Tertile 3	1.53 (0.47, 5.03)	1.30 (0.57, 2.97)
	p-trend= 0.35	p-trend=0.26
Lung adenocarcinoma
Per IQR Increase	0.93 (0.78, 1.10)	0.91 (0.77, 1.07)
Never Smokers	1 (Reference)	1 (Reference)
Tertile 1	0.59 (0.33, 1.07)	0.62 (0.37, 1.05)
Tertile 2	1.50 (0.94, 2.42)	1.56 (1.02, 2.37)
Tertile 3	1.56 (0.79, 3.08)	1.41 (0.78, 2.54)
	p-trend=0.15	p-trend=0.10
Large cell lung cancer
Per IQR Increase	1.07 (0.88, 1.29)	1.06 (0.88, 1.27)
Never Smokers	1 (Reference)	1 (Reference)
Tertile 1	0.92 (0.31, 2.74)	0.80 (0.36, 1.79)
Tertile 2	2.07 (0.91, 4.74)	1.57 (0.83, 2.98)
Tertile 3	3.77 (1.43, 9.91)	2.51 (1.20, 5.25)
	p-trend=0.007	p-trend=0.003
Overall UADT cancer
Per IQR Increase	0.96 (0.84, 1.10)	0.98 (0.86, 1.12)
Never Smokers	1 (Reference)	1 (Reference)
Tertile 1	0.61 (0.41, 0.90)	0.64 (0.44, 0.93)
Tertile 2	0.91 (0.63, 1.34)	0.95 (0.67, 1.35)
Tertile 3	0.59 (0.33, 1.05)	0.66 (0.40, 1.10)
	p-trend= 0.13	p-trend=0.14
UADT squamous cell carcinoma
Per IQR Increase	0.94 (0.82, 1.09)	0.96 (0.83, 1.10)
Never Smokers	1 (Reference)	1 (Reference)
Tertile 1	0.60 (0.39, 0.92)	0.64 (0.43, 0.94)
Tertile 2	0.86 (0.57, 1.30)	0.92 (0.64, 1.33)
Tertile 3	0.56 (0.30, 1.03)	0.64 (0.37, 1.10)
	p-trend=0.09	p-trend=0.10
Esophageal adenocarcinoma
Per IQR Increase	1.02 (0.80, 1.30)	1.07 (0.85, 1.34)
Never Smokers	1 (Reference)	1 (Reference)
Tertile 1	0.75 (0.28, 2.01)	0.83 (0.38, 1.78)
Tertile 2	1.69 (0.76, 3.78)	1.44 (0.76, 2.72)
Tertile 3	1.22 (0.39, 3.82)	1.12 (0.49, 2.56)
	p-trend=0.45	p-trend=0.43

¹Models adjusted for age, race/ethnicity, sex, years of education, drink-years and pack-years.

We observed a higher association between tar and overall lung cancer for Whites than non-Whites in the highest tertile of exposure compared to never smokers (Supporting Information Table 2: OR = 16.65, 95% CL’s= 10.22, 27.13 versus OR = 5.89, 95% CL’s = 3.43, 10.10; p for multiplicative interaction <0.05). The association did not differ for overall UADT cancer (Supporting Information Table 2: p for multiplicative interaction > 0.05). After we adjusted for pack-years in semi-Bayes corrected models, the positive trend between cumulative tar and lung cancer was apparent in Whites (overall, small cell, adenocarcinoma and large cell disease, p-trend <0.05) but not non-Whites. We did not observe modification of the association between cumulative tar and either cancer type by smoking status (current vs. former smokers), comparing the highest tertile of exposure to the first tertile (Supporting Information Table 3: p for multiplicative interaction > 0.05).

Discussion

Our study of 611 lung cancer patients, 601 UADT cancer patients and 1,040 controls found that cumulative tar exposure is highly correlated with pack-years and is positively associated with lung and UADT cancers. An increase of about 1 kg lifetime cumulative tar exposure was associated with approximately a 61% increased risk of lung cancer and about a 21% increased risk of UADT cancer. This concurs with prior evidence that tobacco smoking is a stronger risk factor for lung cancer than UADT cancer.³¹ To estimate cumulative tar exposure, we modified Zang and Wynder’s cumulative lifetime tar index by incorporating historical tar values and cigarette portion size. We believe that the modification is closer to the true tobacco exposure. The major advantage of using this index compared to pack-years is that cumulative tar accounts for the attributable risk of particulate carcinogens and could potentially sort out the remaining risk by other carcinogens not directly associated with tar exposure, such as those in gas-phase. Our reported positive association between cumulative tar and lung cancer risk is consistent with other reports.^13,15,16,19 While most studies of cumulative tar and cancer risk have relied on one or two years of reported cigarette tar ratings,^13–18 one study of lung cancer in Tasmania estimated cumulative tar based on 17 reports of machine-measured tar yields published between 1961and 1996.¹⁹ Our study collected yields from 39 reports covering 25 testing years between 1967 and 2000. Furthermore, this appears to be the first association reported for overall UADT cancer.

We detected associations between cumulative cigarette tar exposure and lung cancer subtypes after adjusting for pack-years, the standard measure of cumulative cigarette smoking. Zang and Wynder¹³ previously noted a positive trend for cumulative tar and lung cancer even after restricting to higher levels of pack-years. However, the residual association we observed could be a result of the strong correlation between cumulative tar and pack-years (r = 0.90). We applied semi-Bayes shrinkage estimation to reduce the potential for false positive findings from variance inflation or multiple comparisons. Cumulative tar neither improved risk models for any case group compared to pack-years in the ROC analysis, nor was cumulative tar associated with UADT cancer after we adjusted for pack-years.

Furthermore, we detected positive associations between cumulative tar exposure and major subtypes of lung cancer, with higher estimates for small cell, squamous cell and large cell cancer than for adenocarcinoma. This concurs with previous reports^13,15,16 for a higher association for cumulative tar in “Kreyberg type I” cancers (squamous, epidermoid, oat, small and large cell) compared to “Kreyberg type II” (adenocarcinoma). This also concurs with two published meta-analyses of cigarette smoking and lung cancer, which reported greater estimates for small cell and squamous cell cancer than for adenocarcinoma and large cell carcinoma.^2,3 However, we had limited sample size to detect small differences between lung and UADT cancer subtypes.

In the United States, although the adult smoking prevalence decreased by 60% between 1965 and 2014, the risk of smoking-related lung cancer and mortality has increased.^11,32,33 Increasing risk in both sexes may partly be attributed to changes in cigarette design and composition in the past 50 years.^{11,12,34–38} For example, filtered cigarettes, which were introduced in the 1950s, are associated with deeper inhalation and smaller particle size of smoke, which may increase the deposition of tobacco carcinogens throughout the airway.^{12,36,37,39,40} In addition, cigarette content of tobacco-specific nitrosamines increased between 17% and 73% from 1978 to 1995 when measured under standard FTC smoking conditions.^11,12,37 Furthermore, these changes may explain the shift in smoking-related lung cancer incidence from squamous cell to adenocarcinoma.^{11,12,34–38} However, while smoking-related lung cancer has increased, average sales-weighted tar yield decreased by 44% between 1968 and 1998²² (See Document ID yqpk0154, page 10). Harris⁴¹ provided a possible explanation for this disparity when he reported weak associations between FTC tar rating and tobacco-specific nitrosamines (r² = 0.38 and 0.76 for NNN and NNK, respectively; both p < 0.01). In spite of this disadvantage, the cumulative tar index allowed us to simultaneously account for changes in tar yield and smoking behavior (e.g., cigarettes per day) over time.

This population-based case-control study included histologically confirmed cases, relatively good statistical power for the main cancer types, and information from both a lifetime exposure questionnaire and longitudinal federal government reports. The first limitation to this study was that machine-testing of cigarette yields has been shown to underestimate smoking exposures.¹² Smokers compensate their breathing to achieve a steady nicotine dose.¹² In addition, cigarettes have been engineered to produce misleadingly lower yields under machine smoking conditions, by such means as perforated filters which smokers cover with their fingers.^11,12 Therefore, we most likely underestimated the cumulative tar index and biased the odds ratios nondifferentially toward the null. Second, the fact that we observed null associations with cancer in the lowest tertile of tar exposure/pack-years was likely due to insufficient power to detect small associations. Third, residual confounding by unmeasured human papillomavirus (HPV) infection may have biased the associations with UADT cancer, especially squamous cell carcinoma. The direction of bias is difficult to evaluate because the reports on the association between tobacco smoking and HPV infection have been inconsistent,^42–45 although our reported associations appear to be biased toward the null. Fourth, adjusting the tar-cancer associations for pack-years may have biased the associations toward the null. Smokers who switch to lower-tar brands likely increase their smoking intensity (compensation), and adjusting for an intermediate variable generally biases estimates toward the null.^12,46 Fifth, selection bias may have occurred if tobacco exposure (measured as pack-years or cumulative tar) was associated with participation differentially for eligible cases and controls. Ten percent of eligible UADT cancer cases and 25% of eligible lung cancer cases died before they could be interviewed. Selective-survival bias could have occurred because smoking is associated with shorter survival time for these cancers.^47,48 Therefore, we would expect a downward bias in OR estimation in this scenario that nonparticipation was selectively greater in more highly exposed cases. Given these limitations, this study does not support the claim of a null or inverse association between “low exposure” to tobacco smoke and risk of these cancer types.

In conclusion, our study suggests that cumulative tar exposure is associated with cancer risk and is associated with small and large cell lung cancer after adjusting for pack-years. The Family Smoking Prevention and Tobacco Control Act of 2009 (United States) requires tobacco product manufacturers and importers to report harmful and potentially harmful constituents (HPHC’s) to the FDA, including 93 carcinogens, toxicants and addictive substances measured from a machine smoking regimen.⁴⁹ Although machine smoking protocols have limited ability to reflect real exposure to smokers, researchers have found that they “may be the limit of current scientific assessment of differences between brands that can be used for regulatory assessment of product toxicity.”¹⁰ It is possible that novel exposure measures incorporating smoking duration and intensity, as well as constituent levels by tobacco product could help to identify people at high risk for cancer who would benefit from screening and/or tobacco cessation intervention. The present study suggests that cumulative tar, a crude estimate of total smoke constituent exposure, may improve exposure assessment and risk estimation particularly for small cell and large cell lung subtypes. Biomarkers of tobacco smoke constituents should also continue to be identified to improve cancer risk assessment.⁵⁰ Public health messages should meanwhile focus on abstaining from all tobacco products, regardless of tar content.^11,12 However, tobacco products should also be strictly regulated to deliver lower doses of carcinogens, in terms of total particulate matter or specific harmful constituents such as tobacco-specific nitrosamines.³⁸

What’s new?

Tobacco use is the leading preventable risk factor for cancer mortality worldwide, but emission differences between individual tobacco products are often not taken into account in exposure estimates. Here, the authors focused on this and estimated cumulative cigarette tar exposure from 39 government reports for participants of a case-control study. This cumulative tar exposure was associated with lung cancer –especially small and large cell subtypes—even after adjusting for pack-years. The data highlight the need to incorporate the composition of tobacco carcinogens in lifetime smoking exposure to improve lung cancer risk estimation.

Abbreviations

CL

confidence limit

CSI

comprehensive smoking index

FTC

US Federal Trade Commission

IQR

interquartile range

OR

odds ratio

PL

semi-Bayes posterior limit

ROC-AUC

area under the curve of the receiver operating characteristic

SBOR

semi-Bayes odds ratio

UADT

upper aerodigestive tract Additional Supporting Information may be found in the online version of this article