Consistency of Biomarkers of Exposure in the Population Assessment of Tobacco and Health Study from Samples Taken 3 to 5 Years Apart

Abstract

Introduction

Urinary biomarkers are useful in characterizing exposure to harmful and potentially harmful constituents of tobacco products and linking exposure to health outcomes. However, the consistency/reproducibility of many urinary biomarkers over long periods is unknown.

Methods

Among people who exclusively used cigarettes in the Population Assessment of Tobacco and Health Study Waves 1, 2, 4, and 5 (ranging from 746 to 1361 subjects), we used weighted models to estimate variance components and intra-class correlation coefficients (ICC) for 15 biomarkers of exposure for urine samples collected 3–5 years apart, creatinine-only-adjusted and also adjusted for demographic and behavioral predictors.

Results

In models adjusted only for creatinine, ICC values of biomarkers ranged from 0.41 (95% confidence interval (CI): 0.32, 0.49) (N‐acetyl‐S‐(2‐carbamoylethyl)‐L‐cysteine) to 0.73 (95% CI: 0.65, 0.81) (4-(methylnitrosamino)-1-(3-pyridyl)-1-butanol), varying within each chemical class. For models adjusted for predictors, associations between biomarkers and predictors were similar for samples collected 3–5 years and 1 year apart. Predictor-adjusted ICCs for samples collected 3–5 years apart ranged from 0.29 (95% CI: 0.17, 0.40) (N‐Acetyl‐S‐(2‐carbamoylethyl)‐L‐cysteine) to 0.63 (95% CI: 0.56, 0.69) (N‐Acetyl‐S‐(2‐hydroxyethyl)‐L‐cysteine) and appeared not different from those for samples collected 1 year apart.

Conclusions

Even for 3 or 5 years between urine sample collection, unadjusted biomarkers of exposure showed fair to excellent reproducibility. Similar consistency between 1 year and 3–5 years between collections was found when including predictors in the model.

Implications.

These biomarkers may be useful to characterize long-term exposures to harmful and potentially harmful constituents from cigarettes with different characteristics for those who smoke cigarettes exclusively.

Introduction

Urinary exposure biomarkers can be useful in characterizing exposure to harmful and potentially harmful constituents (HPHCs) of tobacco products, examining the impact of changes in constituent yields, and linking exposure to health outcomes.^1–3 However, urinary biomarkers of exposure to different HPHCs have a wide range of half-lives from a few hours to many years.^4,5 The concentrations of most HPHC exposure biomarkers in spot urine samples reflect recent exposures due to their short half-lives. Other biomarkers including cadmium, and, to some extent, 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanol (NNAL) are longer lived and may reflect cumulative exposures over a longer time period.⁶

Most adverse health outcomes from tobacco product use take many years to manifest and are likely impacted by aggregate exposure to HPHCs over a long time period. Coronary heart disease is associated with total years of cigarette smoking, cigarettes smoked per day and degree of inhalation.⁷ The risk of death from chronic obstructive pulmonary disease is associated with overall lifetime exposure to cigarette smoke.⁸ Thus, it is important to be able to understand the association between long-term exposure and adverse health outcomes. One way to accomplish this would be to measure biomarkers repeatedly over a lifetime of use. However, this is expensive and impractical in most cases. An alternative approach is to monitor variables related to product use behavior and exposure, include their influence when characterizing exposure, and use a limited number of biomarker measurements to estimate exposure over longer periods of time. However, it is uncertain how well urinary biomarkers in a few spot urine samples can reliably characterize long-term exposure to non-persistent chemicals in cigarette smoke and predict adverse health outcomes that may result from extended periods of exposure. Understanding the long-term reproducibility or consistency of these biomarkers is crucial to understanding the utility of these biomarkers as measures of HPHC exposure from tobacco products with different characteristics, an important consideration in decision-making by tobacco product regulators.

The Population Assessment of Tobacco and Health (PATH) Study uses a longitudinal cohort design to evaluate how changes in product use behavior over time alter the factors that may influence the product design-product use behavior-exposure-health impact continuum.⁹ The longitudinal design of the PATH Study and the repeated collection of biological samples provides a unique opportunity to compare the consistency of biomarkers of exposure over different time periods, which has not been adequately addressed in previous studies.

In an earlier paper, we evaluated the consistency of select urinary biomarkers of HPHC exposure in those who smoked exclusively over 1 year in the PATH Study.⁴ However, the consistency of these biomarkers over periods longer than 1 year is unknown. The goal of the current study is to describe the consistency of these urinary biomarkers over 3–5 years using data from the PATH Study Waves 1, 4, and 5 and compare this to our previous findings over 1 year to better understand how the time interval between urine sample collection can influence the reliability of biomarkers of exposure, especially for repeat measurements for the same subject.

Methods

Study Population

The PATH Study is a nationally representative, longitudinal cohort study of approximately 46,000 adults and youth representative of the civilian, non-institutionalized population in the United States, aged 12 years and older. It began in 2013 as a collaboration between the National Institute on Drug Abuse of the U.S. National Institutes of Health and the Center for Tobacco Products of the U.S. Food and Drug Administration, administered by Westat, and uses Audio Computer-Assisted Self-Interviewing for data collection in English or Spanish.^9,10 The PATH Study was approved by the Westat Institutional Review Board (Organization number IORG0000410, IRB number 00000695) with all adult participants providing written informed consent.

Baseline interviews and biospecimen collection were conducted at Wave 1 in 2013/2014. Follow-up interviews have been conducted almost annually since Wave 1. At Wave 1, a stratified probability sample of 11,522 adults who completed the Wave 1 Adult Interview and provided a urine specimen was selected for biospecimen laboratory analyses. The sample was selected to ensure respondents represented diverse tobacco product use patterns, including users of multiple tobacco products and never users of any tobacco product. These sampled adults constitute the Wave 1 Biomarker Core. Members of the Wave 1 Biomarker Core were asked for urine specimens in each follow-up wave after completion of the interview for that wave. We refer to Table 1 in the Biomarker Restricted Use Files User Guide and the Laboratory File Inventory for more details (https://doi.org/10.3886/Series606). Additional information about the design of the PATH Study and biomarker data are available online.^10,11 Three analytic samples were used. Sample 1 included data from Wave 1 (September 12, 2013–December 15, 2014) and the Wave 1 cohort at Wave 2 (October 23, 2014–October 30, 2015) to examine the reproducibility of all biomarkers over approximately 1 year. Sample 2 included data from Wave 1 and the Wave 1 cohort at Wave 5 (December 1, 2018–November 30, 2019) to examine the reproducibility of nicotine, tobacco-specific nitrosamine (TSNA), volatile organic compound (VOC), and metal exposure biomarkers over approximately 5 years. Sample 3 included data from Wave 1 and the Wave 1 cohort at Wave 4 (December 1, 2016–January 3, 2018) to examine the reproducibility of polycyclic aromatic hydrocarbon (PAH) exposure biomarkers over approximately 3 years. Sample 3 was used for PAH exposure biomarkers because PAH metabolite concentrations were not available for Wave 5 at the time of this analysis. We refer to Sample 1 as the “1-year follow-up sample,” Sample 2 as the “5-year follow-up sample,” and Sample 3 as the “3-year follow-up sample.” Among households that were screened, the overall weighted response rate for the Adult Interview ranged from 69.5% to 83.2% for the first five waves of the PATH Study.¹² For each sample, we included participants who smoked cigarettes exclusively at both time points, defined as those who reported currently using cigarettes and not traditional cigars, filtered cigars, cigarillos, e-cigarettes (Wave 1)/e-products (Waves 2, 4, and 5), pipes, hookah, snus, or dissolvable tobacco. Further, we required participants to have valid sample weights and complete data at both time points on biomarkers and core predictors of biomarker concentrations (age, race, sex, number of cigarettes smoked in past 30 days, favorite cigarette brand, time since last cigarette use, and urine creatinine). Observations with urine creatinine concentrations <10 mg/dL or >370 mg/dL were excluded. Flow charts describing the construction of the three analytical samples are included in Supplementary Figures S1–S3.

Predictors of Biomarkers of Exposure

Previous studies have reported the influence of demographics, use behavior and sample collection timing on levels of biomarkers of exposure.^13–15 Demographic variables were obtained from the PATH Study Restricted Use File (RUF) and include categorical age (18–24, 25–34, 35–54, and 55 years and older), sex (male/female), and race (White race alone, Black race alone, Asian race alone, and Other race including multiracial). Behavioral predictors include number of cigarettes smoked in the past 30 days, time since last cigarette use, change in favorite cigarette brand, and cannabis use. Cannabis use was only included in the model when this use behavior had been previously shown to influence specific tobacco biomarkers of exposure.⁴ Within each analytic sample, change in favorite cigarette brand (yes/no) was derived based upon participants’ self-reported favorite cigarette brand at each time point. Interested readers can find further details on the questionnaires, the calculation of use behavior variables, and procedures used for correcting values of age, number of cigarettes smoked in past 30 days, and time since last cigarette use at the PATH Study website¹⁰ and in our previous publications.^4,16

Biomarkers of Exposure

Biomarker data used in this analysis are from the PATH Study Biomarker Restricted Use Files (BRUF). A list of precursor compounds, urinary biomarkers, and limits of detection are provided in Table 1. Urinary biomarkers of nicotine exposure, cotinine (COTT) and trans-3’-hydroxycotinine (HCTT), were measured by isotope-dilution liquid chromatography tandem mass spectrometry (LC/MS/MS) after hydrolysis of the glucuronides, and their concentrations were the sums of both the free and bound forms. The molar sum of COTT and HCTT was calculated as total nicotine equivalents (TNE2).¹⁷ The TSNA urinary metabolite 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanol (NNAL) was analyzed by isotope-dilution liquid chromatography–tandem mass spectrometry using procedures as previously described.^18,19 The PAH biomarkers, 2‐hydroxyfluorene (2-FLU) and 2-naphthol (2-NAP), were measured in urine using enzymatic hydrolysis, online solid phase extraction, and high-performance liquid chromatography isotope-dilution tandem mass spectrometry, as described previously.^20,21 Urinary concentrations of the VOC metabolites, including N‐acetyl‐S‐(2‐hydroxypropyl)‐L‐cysteine (2HPMA), N‐acetyl‐S‐(2‐carboxyethyl)‐L‐cysteine (2CoEMA), N‐acetyl‐S‐(2‐carbamoylethyl)‐L‐cysteine (2CaEMA), N‐acetyl‐S‐(2‐hydroxyethyl)‐L‐cysteine (2HEMA), 3,4-methylhippuric acid (34MHA), N‐acetyl‐S‐(N‐methylcarbamoyl)‐L‐cysteine (MCaMA), and N‐acetyl‐S‐(2‐cyanoethyl)‐L‐cysteine (2CyEMA), were measured in urine using ultrahigh performance liquid chromatography coupled with electrospray ionization tandem mass spectrometry with some modification to a published procedure.²² Cadmium (Cd) and lead (Pb) were measured in urine specimens using inductively coupled plasma universal cell technology mass spectrometry after a simple dilution sample preparation step, allowing for selective determination.^23,24 In Waves 1 and 2, Cd was measured as ¹¹⁴Cd. In Wave 5, Cd was measured as ¹¹¹Cd. In all waves of the PATH Study, Pb was measured as the sum of ²⁰⁶Pb, ²⁰⁷Pb, and ²⁰⁸Pb. Concentrations less than the LOD were imputed using the limit of detection/√2.²⁵ Detection rates (percent above the detection limit) were above 99% for all biomarkers except 2HEMA (83%–87%) and cadmium (95%–96%). Urine creatinine was measured by an enzymatic assay on a commercial automated clinical chemistry analyzer with a detection limit of 1.1 mg/dL and was used to account for urine dilution.

Table 1.

Constituents, Biomarkers of Exposure, and Limits of Detection

Constituent	Biomarker	Detection Limits
Nicotine	Cotinine (COTT)	0.030 ng/mL
	trans-3’-Hydroxycotinine (HCTT)	0.030 ng/mL
	Total Nicotine Equivalents (TNE2)	Not applicable
4-(methylnitrosoamino)-1-(3-pyridyl)-1-butanone (NNK)	4-(methylnitrosamino)-1-(3-pyridyl)-1-butanal (NNAL)	0.0006 ng/mL
Naphthalene	2-Naphthol (2-NAP)	0.09 ng/mL
Fluorene	2‐Hydroxyfluorene (2‐FLU)	0.008 ng/mL
Acrylamide	N‐Acetyl‐S‐(2‐carbamoylethyl)‐L‐cysteine (2CaEMA)	2.20 ng/mL
Acrylonitrile	N‐Acetyl‐S‐(2‐cyanoethyl)‐L‐cysteine (2CyEMA)	0.5 ng/mL
Xylene	3,4-Methylhippuric acid (34MHA)	8.00 ng/mL
Acrolein	N‐Acetyl‐S‐(2‐carboxyethyl)‐L‐cysteine (2CoEMA)	6.96 ng/mL
N,N‐Dimethylformamide /isocyanates	N‐Acetyl‐S‐(N‐methylcarbamoyl)‐L‐cysteine (MCaMA)	6.26 ng/mL
Acrylonitrile, Vinyl chloride, Ethylene oxide	N‐Acetyl‐S‐(2‐hydroxyethyl)‐L‐cysteine (2HEMA)	0.79 ng/mL
Propylene Oxide	N‐Acetyl‐S‐(2‐hydroxypropyl)‐L‐cysteine (2HPMA)	5.3 ng/mL
Cadmium	Cadmium	0.036 ng/mL (Waves 1 and 2) 0.055 ng/mL (Wave 5)
Lead	Lead	0.03 ng/mL (Waves 1 and 2) 0.022 ng/mL (Wave 5)

Statistical Analysis

Statistical analysis, which has been described previously,¹⁶ was conducted using SAS software version 9.4 (SAS Institute Inc., Cary, NC, USA), and adjusted for sample design using all-waves urine weights for the Wave 1 Biomarker Core to provide nationally representative estimates of persons who were established as smoking cigarettes in the U.S. civilian, non-institutionalized adult population at the time of Wave 1 and Wave 2, Wave 1 and Wave 4, or Wave 1 and Wave 5. For initial analysis of each analytic sample, we used weighted log-transformed biomarker concentrations with creatinine concentration included in the model and subject-specific random intercepts. Within-person and between-person variance and intra-class correlation coefficients (ICCs) were calculated from a random effects model (no fixed effects) to reflect the consistency of each biomarker (ICC1) for the 3-year and 5-year follow-up samples. As previously described,¹⁶ a macro was created to perform nationally representative variance estimation using balanced repeated replicate weights. Calculation of ICC1 did not include demographic and behavior factors in the model. In addition, we also used weighted mixed effects models to estimate the association between all demographic and use behavior predictors and log-transformed biomarker concentrations. ICCs based on Wave 1 and Wave 4 or Wave 5 (ICC2) were obtained from models adjusted for demographic and behavior factors and compared with ICC2 based on Wave 1 and Wave 2 to investigate whether changes occurred in predictors and consistency of each biomarker as the time gap between samples lengthened.

Data Availability

Biomarker data are available from the PATH Study Biomarker Restricted Use Files (https://doi.org/10.3886/ICPSR36840.v15). Demographic and behavioral variables are available from the PATH Study Restricted Use File (https://doi.org/10.3886/ICPSR36231.v29).

Results

The 1-year follow-up sample included 1361 participants and the 5-year follow-up sample included 746 participants. Table 2 presents the sizes and weighted percentages of the sample when it is differentiated by demographic (age, sex, and race) or behavioral (time since last smoked cigarette, change of favorite cigarette brand between waves, and cannabis use) factors. The number of cigarettes smoked in the last 30 days was treated as continuous and its weighted mean and 95% confidence interval of the mean are presented instead. Descriptive statistics for the 3-year follow-up sample are available in Supplementary Table S1. As expected and seen in Table 2, there were fewer participants aged 18–24 in Wave 5, largely as a result of participants aging into a higher category after 5 years. Also, the weighted percent of participants who primarily used a different brand in a later wave were higher as the time between waves increased with 16.2% changing brands between Waves 1 and 2, 23.8% between Waves 1 and 4, and 30.3% between Waves 1 and 5.

Table 2.

Demographic and Behavioral Characteristics (N) of 1-Year and 5-Year Follow-Up Samples and Weighted Percentages of Each Characteristic (%)

	1-year follow-up sample (Waves 1 and 2) N (%)		5-year follow-up sample (Waves 1 and 5) N (%)
Overall	1361 (100.0)		746 (100.0)
Sex
Male	565 (44.3)		277 (42.5)
Female	796 (55.7)		469 (57.5)
Race
White alone	1047 (78.1)		568 (77.4)
Black alone	207 (16.0)		124 (16.8)
Asian alone	10 (1.1)		8 (1.5)
Other (including multiracial)	97 (4.8)		46 (4.4)
Age	Wave 1	Wave 2	Wave 1	Wave 5
18–24 years	175 (7.6)	138 (5.9)	82 (6.2)	11 (0.7)
25–34 years	294 (20.5)	292 (20.0)	171 (25.2)	152 (17.2)
35–54 years	572 (43.3)	574 (41.8)	342 (44.4)	328 (43.9)
55 years and older	320 (28.6)	357 (32.3)	151 (24.2)	255 (38.2)
Time since last smoked a cigarette
In the past hour	996 (74.7)	1018 (77.3)	563 (77.2)	572 (78.9)
Sometime today	200 (14.3)	186 (12.3)	100 (12.7)	102 (12.4)
Yesterday	103 (7.1)	104 (7.3)	48 (5.5)	45 (5.8)
Day before yesterday	22 (1.7)	13 (0.8)	14 (1.4)	12 (1.6)
Three or more days ago	40 (2.2)	40 (2.4)	21 (3.2)	15 (1.4)
Cannabis usea
Yes	216 (15.1)	269 (21.3)	132 (19.1)	182 (26.7)
No	1143 (84.9)	1090 (78.6)	613 (80.8)	564 (73.4)
Changed favorite brand between waves
No	1117 (83.8)		516 (69.7)
Yes	244 (16.2)		230 (30.3)
	Weighted Mean (95% CI)		Weighted Mean (95% CI)
Number of cigarettes smoked in the last 30 days	425.2 (397.3, 453.0)	404.7 (376.8, 432.6)	435.9 (395.7- 476.0)	430.5 (384.2, 476.8)

^a—Sample size for cannabis use, which was only included in the 2CaEMA and 2CyEMA analyses, was smaller due missing responses.

Table 3 provides the scaled between-subject variance, within-subject variance and ICC of biomarkers with only creatinine included as a covariate in the model for the 3-year and 5-year follow-up samples (ICC1). ICC1 values ranged from 0.41 for 2CaEMA to 0.73 for NNAL. For every chemical class of biomarkers (nicotine, TSNAs, PAHs, VOCs and metals), there was range of ICC1 values but at least one biomarker with an ICC1 value greater than 0.64.

Table 3.

Variance Components^a and Intra-Class Correlation Coefficients (ICC1 = between/total) for Creatinine-Adjusted Biomarkers for the 3-Year and 5-Year Follow-Up Samples^b With 95% Confidence Intervals for ICC1

Analyte	Waves	Between-subject variance	Within-subject variance	ICC1	Etemadi et al.
COTT	1 and 5	0.2458	0.1185	0.6746 [0.5860, 0.7632]	0.58
HCTT	1 and 5	0.2803	0.1298	0.6836 [0.6017, 0.7654]	0.44
TNE2	1 and 5	0.2356	0.1214	0.6600 [0.5728, 0.7472]
NNAL	1 and 5	0.1889	0.0700	0.7298 [0.6506, 0.8089]	0.62
2-NAP	1 and 4	0.0338	0.0378	0.4716 [0.3856, 0.5576]	0.43
2-FLU	1 and 4	0.0655	0.0355	0.6489 [0.5878, 0.7101]	0.31
2CaEMA	1 and 5	0.0304	0.0442	0.4075 [0.3218, 0.4932]	0.27
2CyEMA	1 and 5	0.1606	0.0686	0.7006 [0.6278, 0.7734]	0.66
34MHA	1 and 5	0.0576	0.0570	0.5027 [0.4322, 0.5732]	0.34
2CoEMA	1 and 5	0.0559	0.0376	0.5980 [0.5292, 0.6668]	0.39
MCaMA	1 and 5	0.0494	0.0289	0.6305 [0.5781, 0.6830]	0.43
2HEMA	1 and 5	0.1171	0.0574	0.6711 [0.6172, 0.7249]
2HPMA	1 and 5	0.0526	0.0614	0.4615 [0.3733, 0.5497]	0.41
Cadmium	1 and 5	0.1015	0.0406	0.7142 [0.6678, 0.7606]
Lead	1 and 5	0.0419	0.0510	0.4512 [0.3776, 0.5247]

^a—Weighted variances were estimated by dividing the variance estimates from SAS PROC MIXED by the average weights for all included participants due to the way PROC MIXED estimates variances when WEIGHT statement and REPEATED statement are specified.

^b—Sample sizes for the 3-year, and 5-year follow-up samples were 952, and 746, respectively. For 2CaEMA and 2CyEMA, one participant was excluded from the 1-year and 5-year analysis due to incomplete data in cannabis use.

^c—ICC calculated on a 5-year follow-up of the Golestan Cohort Study for cigarette smokers by Etemadi, A., et al. Urinary Biomarkers of Carcinogenic Exposure among Cigarette, Waterpipe, and Smokeless Tobacco Users and Never Users of Tobacco in the Golestan Cohort Study. Cancer Epidemiol Biomarkers Prev. 2019 Feb;28(2):337–347 are listed for comparison.

HCTT=trans-3’-Hydroxycotinine; COTT = Cotinine; TNE2 = Total nicotine equivalents, the molar sum of COTT and HCTT; NNAL = 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanol; 2-NAP = 2-Naphthol; 2-FLU = 2‐Hydroxyfluorene; 2HEMA = N‐Acetyl‐S‐(2‐hydroxyethyl)‐L‐cysteine; 2CoEMA = N‐Acetyl‐S‐(2‐carboxyethyl)‐L‐cysteine; 2CyEMA = N‐Acetyl‐S‐(2‐cyanoethyl)‐L‐cysteine; MCaMA = N‐Acetyl‐S‐(N‐methylcarbamoyl)‐L‐cysteine; 2CaEMA = N‐Acetyl‐S‐(2‐carbamoylethyl)‐L‐cysteine; 2HPMA = N‐Acetyl‐S‐(2‐hydroxypropyl)‐L‐cysteine; 34MHA = 3,4-Methylhippuric acid.

Table 4 provides, as an example, the associations between demographic and behavioral predictors and log-transformed urinary cotinine for the 1- and 5-year follow-up samples. Supplementary Tables S2–S15 provide the results for the other biomarkers of exposure. In most cases the associations between predictors and each biomarker were similar between the 3- or 5-year follow-up samples and the 1-year follow-up sample. For most biomarkers, the beta coefficients for age and race generally had the same sign across analytic samples. In some cases, the beta coefficients’ p-values became non-significant in the 3- or 5-year follow-up samples. The associations with sex were also similar across samples. However, we note that, compared to the 1-year follow-up sample, the association between sex and 2CoEMA became stronger and statistically significant in the 5-year follow-up sample (Supplementary Table S10). While the associations with time since last smoked a cigarette were overall similar across analytic samples, for several VOC biomarkers (Supplementary Tables S7, S11, S12 and S15) the beta coefficients for some categories of “time since last smoked a cigarette” were attenuated and no longer statistically significant in the 5-year compared to the 1-year follow-up sample.

Table 4.

Associations Between Demographic and Behavioral Predictors and Log-Transformed Urinary Cotinine (COTT) (ng/mL) in the 1-Year and 5-Year Follow-Up Samples as Estimated by Weighted Linear Mixed Effects Models

	1-year follow-up sample Wave 1–Wave 2		5-year follow-up sample Wave 1–Wave 5
	Estimate (95% CI)	P-value	Estimate (95% CI)	P-value
Intercept	1.45 (1.10, 1.80)	<0.00001	1.21 (0.505, 1.92)	0.00098
Age
18–24 years	−0.167 (−0.264, −0.071)	0.00086	−0.0986 (−0.194, −0.0033)	0.043
25–34 years	−0.0692 (−0.134, −0.00460)	0.036	−0.0238 (−0.0953, 0.0476)	0.51
35–54 years	−0.0409 (−0.0858, 0.00403)	0.074	−0.0192 (−0.0645, 0.0261)	0.40
55 years and older	Ref.		Ref.
Sex
Male	−0.0703 (−0.112, −0.0287)	0.0011	−0.0809 (−0.136, −0.0252)	0.0048
Female	Ref.		Ref.
Race
White alone	Ref.		Ref.
Black alone	−0.113 (−0.184, −0.0413)	0.0023	−0.108 (−0.185, −0.0302)	0.0070
Asian alone	−0.599 (−1.01, −0.184)	0.0051	−0.175 (−0.505, 0.156)	0.30
Other, including multiracial	0.0272 (−0.0532, 0.108)	0.50	−0.118 (−0.369, 0.132)	0.35
Urinary creatinine	0.00264 (0.00239, 0.00288)	<0.00001	0.00254 (0.00217, 0.00291)	<0.00001
Time since last smoked cigarette
In the past hour	1.62 (1.27, 1.97)	<0.00001	1.84 (1.12, 2.55)	<0.00001
Sometime today	1.55 (1.21, 1.89)	<0.00001	1.81 (1.11, 2.52)	<0.00001
Yesterday	1.25 (0.925, 1.58)	<0.00001	1.52 (0.836, 2.20)	0.00003
Day before yesterday	0.809 (0.471, 1.15)	<0.00001	0.690 (0.0794, 1.30)	0.027
Three or more days ago	Ref.		Ref.
Number of cigarettes smoked in the last 30 days	0.00042 (0.00034, 0.00050)	<0.00001	0.00045 (0.00033, 0.00057)	<0.00001
Changed favorite brand between waves
No	Ref.		Ref.
Yes	−0.0652 (−0.123, −0.00725)	0.028	−0.0344 (−0.106, 0.0369)	0.34

Table 5 provides the scaled between-subject variance, within-subject variance and ICC2 as estimated from the adjusted mixed effects models. In almost all cases (14 out of 15), between-subject variance was lower in the 3- or 5-year follow-up samples compared to the 1-year follow-up sample. The only exception was 2HEMA, although its between-sample difference was only 3.5%. Across all biomarkers, the average difference in between-subject variance comparing analytic samples with 3- or 5-year follow-up to 1-year follow-up was −14.8%. In almost all cases (12 out of 15), within-subject variance was higher in the 3- or 5-year follow-up samples compared to the 1-year follow-up sample. The three exceptions were 2-FLU, MCaMA, and Cd. The average difference in within-subject variance between samples with longer follow-up and the 1-year follow-up sample was 11.1%. Consistent with the generally lower between-subject variance and higher within-subject variance in the 3- or 5-year follow-up samples compared to the 1-year follow-up sample, in all cases, ICC2 was lower in the 3- or 5-year follow-up samples compared to the 1-year follow-up sample. The average difference in ICC2 between the 3- or 5-year follow-up samples and the 1-year follow-up sample was −13.1% and ranged from −35% for 2CaEMA to −0.6% for 2-FLU. For all biomarkers, the 95% confidence limits for ICC2 determined for the 1-year follow-up sample overlapped with the 3- or 5-year follow-up samples.

Table 5.

Variance Components^a and Intra-Class Correlation Coefficients (ICC2 = between/total) Estimated From the Mixed Models^b for the 1-Year, 3-Year, and 5-Year Follow-Up Samples^c With 95% Confidence Intervals for ICC2

	Between-subject variance	Within-subject variance	ICC2
COTT
Waves 1 and 2	0.0947	0.0735	0.5631 [0.4735, 0.6526]
Waves 1 and 5	0.0804	0.0942	0.4606 [0.3215, 0.5998]
HCTT
Waves 1 and 2	0.1452	0.0846	0.6320 [0.5606, 0.7033]
Waves 1 and 5	0.1258	0.0988	0.5601 [0.4479, 0.6722]
TNE2
Waves 1 and 2	0.1005	0.0759	0.5697 [0.4855, 0.6359]
Waves 1 and 5	0.0822	0.0934	0.4681 [0.3372, 0.5990]
NNAL
Waves 1 and 2	0.0879	0.0526	0.6256 [0.5581, 0.6930]
Waves 1 and 5	0.0781	0.0678	0.5356 [0.3917, 0.6794]
2-NAP
Waves 1 and 2	0.0170	0.0319	0.3478 [0.2653, 0.4304]
Waves 1 and 4	0.0168	0.0375	0.3087 [0.2090, 0.4085]
2-FLU
Waves 1 and 2	0.0278	0.0366	0.4311 [0.3461, 0.5160]
Waves 1 and 4	0.0257	0.0342	0.4283 [0.3460, 0.5105]
2CaEMA
Waves 1 and 2	0.0307	0.0388	0.4413 [0.3778, 0.5048]
Waves 1 and 5	0.0180	0.0446	0.2871 [0.1741, 0.4002]
2CyEMA
Waves 1 and 2	0.0640	0.0602	0.5152 [0.4417, 0.5888]
Waves 1 and 5	0.0571	0.0623	0.4785 [0.3669, 0.5900]
34MHA
Waves 1 and 2	0.0315	0.0515	0.3800 [0.3049, 0.4551]
Waves 1 and 5	0.0234	0.0544	0.3012 [0.2008, 0.4015]
2CoEMA
Waves 1 and 2	0.0475	0.0343	0.5809 [0.5198, 0.6420]
Waves 1 and 5	0.0398	0.0353	0.5300 [0.4398, 0.6201]
MCaMA
Waves 1 and 2	0.0258	0.0311	0.4535 [0.3752, 0.5318]
Waves 1 and 5	0.0200	0.0269	0.4270 [0.3378, 0.5162]
2HEMA
Waves 1 and 2	0.0878	0.0505	0.6350 [0.5906, 0.6794]
Waves 1 and 5	0.0909	0.0539	0.6277 [0.5633, 0.6921]
2HPMA
Waves 1 and 2	0.0387	0.0557	0.4104 [0.3444, 0.4764]
Waves 1 and 5	0.0383	0.0588	0.3943 [0.3105, 0.4780]
Cadmium
Waves 1 and 2	0.0513	0.0426	0.5465 [0.4744, 0.6186]
Waves 1 and 5	0.0367	0.0413	0.4705 [0.3998, 0.5410]
Lead
Waves 1 and 2	0.0433	0.0390	0.5261 [0.4590, 0.5932]
Waves 1 and 5	0.0332	0.0529	0.3857 [0.3020, 0.4694]

^a—Weighted variances were estimated by dividing the variance estimates from SAS PROC MIXED by the average weights for all included participants due to the way PROC MIXED estimates variances when WEIGHT statement and REPEATED statement are specified.

^b—Models for all biomarkers were adjusted for age, sex, race, urine creatinine, time since last smoked cigarette, number of cigarettes smoked in the last 30 days, and change in favorite cigarette brand. Models for 2CaEMA and 2CYEMA were additionally adjusted for cannabis use.

^c—Sample sizes for the 1-year, 3-year, and 5-year follow-up samples were 1361, 952, and 746, respectively. For 2CaEMA and 2CyEMA, one participant was excluded from the 1-year and 5-year analysis due to incomplete data in cannabis use.

HCTT=trans-3’-Hydroxycotinine; COTT = Cotinine; TNE2 = Total nicotine equivalents, the molar sum of COTT and HCTT; NNAL = 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanol; 2-NAP = 2-Naphthol; 2-FLU = 2‐Hydroxyfluorene; 2HEMA = N‐Acetyl‐S‐(2‐hydroxyethyl)‐L‐cysteine; 2CoEMA = N‐Acetyl‐S‐(2‐carboxyethyl)‐L‐cysteine; 2CyEMA = N‐Acetyl‐S‐(2‐cyanoethyl)‐L‐cysteine; MCaMA = N‐Acetyl‐S‐(N‐methylcarbamoyl)‐L‐cysteine; 2CaEMA = N‐Acetyl‐S‐(2‐carbamoylethyl)‐L‐cysteine; 2HPMA = N‐Acetyl‐S‐(2‐hydroxypropyl)‐L‐cysteine; 34MHA = 3,4-Methylhippuric acid.

Discussion

ICC1 for samples collected 3- and 5-years apart in the PATH Study ranged from 0.41 to 0.73 with at least one biomarker in every chemical class greater than 0.64. Ciccheti²⁶ previously suggested guidelines for interpreting ICC results: < 0.40 = poor, 0.40 to 0.59 = fair, 0.60 to 0.74 = good, and 0.75 to 1.0 = excellent. Our results indicate that among people who exclusively smoked cigarettes in the PATH Study, even for 3 or 5 years between sample collections, ICC1 for all of these biomarkers falls within the fair to good range with at least one biomarker in each class in the good range. In all cases ICC1 determined in our study with data from the PATH Study were higher than the previous work by Etemadi et al.²⁷ (see Table 2) who evaluated ICC values for some tobacco exposure biomarkers taken 5 years apart. This may be due to the well-controlled study design, well-defined study population and larger number of participants of the PATH Study, as well as the populations studied and product use behavioral differences.

In order to evaluate the association of changes in cigarette design characteristics with biomarkers of exposure, the contribution of use behavior to the variation in biomarkers of exposure should be addressed.²⁸ In this study utilizing repeated biomarkers of exposure collected at different time intervals, we found that associations between predictors and biomarkers of exposure were overall similar regardless of time elapsed between sample collection. Differences in beta coefficients or statistical significance, where they occurred, are likely due to a limited number of participants in certain categories of covariates (e.g., Asian race) or a decrease in the number of subjects in each age category as observed with the aging up of participants in the lowest age category in Wave 1.

As for the consistency of biomarkers after adjusting for demographics and behavior predictors, we found that compared to a 1-year follow-up sample period, in all cases, ICC2 was lower when samples were collected 3 or 5 years apart. However, ICC2 for samples taken 3 or 5 years apart were on average only 13.1% lower than ICC2 for samples taken 1 year apart. In addition, for all biomarkers, the confidence intervals for ICC2 for the 1-year follow-up sample and the 3- or 5-year follow-up sample overlapped. We have previously shown¹⁶ that including demographics and use behavior covariates in the model reduces the between-subject variance significantly more than the within-subject variance leading to lower calculated ICC. Our results indicate that among people who exclusively smoked cigarettes, even for 3 or 5 years between sample collections, when the contributions of the demographic and use behavior predictors were removed as in this study, the ICC2 for most of these biomarkers in those who exclusively smoked cigarettes long-term still fell within the fair range. These biomarkers could be valuable tools to assess differences in HPHC exposures among those who regularly smoke cigarettes and use/switch between cigarettes with different characteristics (e.g., filter ventilation, menthol flavoring, nicotine concentration). This paper does not fully address the changes in product design or brands/subbrands used by those who smoke cigarettes. The next step will be to examine differences in biomarker concentrations between participants who smoked cigarette subbrands that changed characteristics between waves²⁹ or changes in biomarker concentrations among participants who switched between different cigarette brands that have different characteristics.

ICC is a function of the between-subject variance and the within-subject variance. Increases in within-subject variance and decreases in between-subject variance over time can both contribute to lower ICC in the samples with longer follow-up. Our analysis of PATH Study data determined that adjusted within-subject weighted variance was commonly higher for the 3- or 5-year follow-up samples compared to the 1-year follow-up samples. As individuals age, their individual use behavior may change, influenced by numerous factors including concern about their health. It would be expected that the variation in individual use behavior and characteristics of the tobacco product they use and, therefore, variation in exposure would increase as the time between sampling increases. People are more likely to change use behavior and switch to use of products that have different characteristics over longer periods of time compared to shorter periods of time. The results of this study support this premise, since, as the time between waves increased, the percent of people who changed products between waves also increased. In addition, urine collection conditions and lab measurement variations may contribute to inconsistency of biomarkers taken from the same subject over longer time periods. It is important to understand this variation over time and how much the complexity of changing use behavior influences the ability to derive useful data from repeated measurements of biomarkers. This will allow us to better understand how product characteristic changes particularly influence exposure to HPHCs and therefore adverse health outcomes.

In contrast, the between-subject weighted variance was usually lower for the 3- or 5-year follow-up samples compared to the 1-year follow-up samples. While the total tobacco market has been changing drastically with the introduction of new products like electronic nicotine delivery systems, heated tobacco products, and nicotine pouches,³⁰ partially due to FDA regulation which has limited the introduction of cigarettes with different characteristics, basic cigarette design of the most commonly used products has not changed dramatically over the past decade. While there will always be substantial variation in biomarker levels across the population of people who use tobacco products due to differences in puffing, inhalation, metabolism, unsmoked butt length and blocking of ventilation holes,^28,31 if the market is stable, it is expected that this between-person distribution of exposure would be fairly constant.

Strengths and Weaknesses

We leveraged longitudinal data in the PATH Study to examine HPHC exposure biomarker consistency over periods longer than 1 year. The PATH Study provides a unique strength for evaluating the impact of tobacco use on adverse health since it follows the same subjects over a long time. This well-designed longitudinal study allows for the repeated collection of biomarkers along with important predictors including demographics and product use behavior, enabling research into questions not answerable by most other investigations.

Body mass index was not included as a predictor in this analysis because there was a substantial number of missing values in waves 2, 4, and 5. We determined that the statistical power achieved by including more subjects outweighed the additional insight that would be provided by including body mass index in the analysis. However, it may be an important factor in variation of biomarkers as the body weight of subjects varies from one wave to another. In addition, we did not include socioeconomic status as a predictor of biomarkers of exposure. Lastly, our algorithm to estimate nationally representative weighted variance and ICCs required balanced data. However, not all participants had complete data in all waves. To preserve relatively large sample sizes, we used only two repeated measures per person (baseline and end of follow-up) to evaluate biomarker consistency.

Even for the case of 3 or 5 years between urine sample collection, creatinine-adjusted biomarkers of exposure to HPHCs among those who exclusively smoked cigarettes in the PATH Study exhibited suitable reproducibility with and without accounting for select demographic and use behavior variables, suggesting they could be useful tools for assessing HPHC exposures from cigarettes for those who smoke. In particular, these biomarkers may be useful to assess the impact of different cigarette design characteristics, such as high and low tip ventilation, tobacco nicotine concentration, or menthol and nonmenthol flavoring, on chemical exposures among exclusive cigarette smokers who switch between cigarettes with different product characteristics. Because product characteristics and tobacco use behavior are more likely to change over longer periods of time, this finding has important implications for further studies evaluating the impact of these parameters on biomarkers of exposure. Investigators should consider the time frame of their study and which biomarkers are most useful in addressing the specific question under study. Biomarkers with short half-lives may be advantageous for evaluating the immediate impact of exposure. Conversely, longer half-life biomarkers may be valuable for assessing the impact of long-term repeated exposure. The decision will depend on the hypothesis of the investigation and must be made on a case-by-case basis by the investigator.

Funding

This manuscript is supported with Federal funds from the National Institute on Drug Abuse (NIDA), National Institutes of Health (NIH), and the Center for Tobacco Products (CTP), Food and Drug Administration (FDA), Department of Health and Human Services, under contract to Westat (contract nos. HHSN271201100027C and HHSN271201600001C) and through interagency agreements between NIH NIDA and FDA CTP and between FDA CTP and the Centers for Disease Control and Prevention and a collaboration between NIH NIDA and the National Cancer Institute, the National Institute on Minority Health and Health Disparities, the National Heart, Lung, and Blood Institute, the National Institute on Alcohol Abuse and Alcoholism, the National Institute of Environmental Health Sciences, the Office of Behavioral and Social Sciences Research, the Office of Disease Prevention, the National Institute of Diabetes and Digestive and Kidney Diseases, and the National Library of Medicine. Dr. Kimmel was substantially involved in providing scientific expertise and providing scientific management for contract nos. HHSN271201100027C and HHSN271201600001C.

Declaration of Interests

David L. Ashley has received funds for work done for the World Health Organization Tobacco Free Initiative, as a consultant for Pfizer, as an employee of Cherokee National Operational Systems and as an independent contractor for McKing Consulting. No other authors report potential conflicts of interest.

Disclaimer

The views and opinions expressed in this manuscript are those of the authors only and do not necessarily represent the views, official policy or position of the U.S. Department of Health and Human Services or any of its affiliated institutions or agencies. Use of trade names is for identification only and does not imply endorsement by CDC, the U.S. Public Health Service, or the U.S. Department of Health and Human Services or any of its affiliated institutions or agencies.

Data availability

Data are available in a public, open-access repository, the National Addiction and HIV Data Archive: https://doi.org/10.3886/Series606

Author contributions

David L. Ashley (Conceptualization [Equal], Data curation [Equal], Formal analysis [Equal], Investigation [Equal], Validation [Equal], Writing—original draft [Equal]), Wanche Zhu (Conceptualization [Equal], Formal analysis [Equal], Investigation [Equal], Methodology [Equal], Writing—review & editing [Equal]), Deeopak Bhandari (Formal analysis [Equal], Writing—review & editing [Equal]), Lanqing Wang (Resources [Equal], Supervision [Equal], Writing—review & editing [Equal]), Jun Feng (Formal analysis [Equal], Writing—review & editing [Equal]), Yuesong Wang (Formal analysis [Equal], Writing—review & editing [Equal]), Lei Meng (Formal analysis [Equal], Writing—review & editing [Equal]), Baoyun Xia (Formal analysis [Equal], Writing—review & editing [Equal]), Jeffery M. Jarrett (Formal analysis [Equal], Writing—review & editing [Equal]), Cindy M. Chang (Conceptualization [Equal], Funding acquisition [Equal], Writing—review & editing [Equal]), Heather L. Kimmel (Conceptualization [Equal], Funding acquisition [Equal], Writing—review & editing [Equal]), and Benjamin C. Blount (Conceptualization [Equal], Resources [Equal], Supervision [Equal], Writing—review & editing [Equal])