Predictors of Tobacco and Alcohol Co-Use From Ages 15 to 32: The Amsterdam Growth and Health Longitudinal Study

Abstract

Tobacco and alcohol are often used in tandem over time, but specific predictors of course and patterns of course over time need explication. We examined differences in alcohol and tobacco course among an adolescent population as they transitioned into young adulthood across a 17-year period. Data came from participants (n = 303 for ages 15–21, n = 196 for ages 21 to 32; 52% female and 54% female, respectively) enrolled in the Amsterdam Growth and Health Longitudinal Study, an epidemiologic investigation examining disease across the life span. We utilized parallel latent growth modeling to assess the impact of sex, personality traits, cholesterol, blood pressure, and body mass index (BMI), on initial status and linear change over time in course of tobacco and alcohol. Females reported less alcohol use at adolescent baseline (β = −21.79), less increase during adolescence (β = −7.92, p < .05), slower decrease during young adulthood (β = 4.67, p < .05), and more rapid decline in tobacco use during young adulthood (β = −70.85, p <.05), relative to males. Alcohol and tobacco use baseline status’ and change over time were all significantly associated with one another during both adolescence and young adulthood (p < .05; aside from alcohol baseline and slope during young adulthood). Effects of BMI, cholesterol, blood pressure, and personality traits were also observed on tobacco and alcohol course. In light of the strong, but sex dependent relationship between alcohol and tobacco course, particularly from ages 15 to 21, prevention efforts to curb heavy alcohol and tobacco use should consider targeting course taking into account biological sex and other notable covariates.

Alcohol and tobacco smoking have detrimental health and economic effects around the world. Alcohol consumption and tobacco smoking are a particular concern in the young population. Youth who drink alcohol are significantly more likely to experience legal, social, and school problems (Miller, Naimi, Brewer, & Jones, 2007). Drinking youth are also at a higher risk for cognitive problems (e.g., impaired memory, impaired growth and integrity of brain structures), multiple forms of injuries, as well as suicide and homicide (Health & Services, 2007). Tobacco smoking adolescents experience negative short-term health effects, including reduced fitness and lung capacity, decreased maximum vital capacity, frequent upper respiratory infections, as well as long-term health effects including higher risk for cardiovascular disease, cancer, and premature death (Health & Services, 2014). Despite these adverse health consequences and the legal constraints on accessibility, alcohol use and tobacco smoking are still prevalent during adolescence with significant adolescent minorities in Europe and the United States reporting regular use of both substances (Abuse, 2016; Central Bureau of Statistics, 2016; Kann et al., 2016).

In the Netherlands, 13 of every 100 adolescents aged 12–17 report daily cigarette smoking (Central Bureau of Statistics, 2016). Prevalence is even higher (33%) for young adults aged 18–24. Alcohol use is also very prevalent, with 69% of 15- to 16 year-old Dutch adolescents reporting regular drinking, and 8% of Dutch adolescents under the age of 13 reporting alcohol intoxication in the past 30 days (Central Bureau of Statistics, 2016). Gender differences are noted in the Netherlands, with adolescent girls generally reporting less tobacco smoking and alcohol use relative to adolescent boys (Central Bureau of Statistics, 2016; Verhagen, Uitenbroek, Schreuders, El Messaoudi, & De Kroon, 2015). In the United States, the most recent National Survey on Drug Use and Health, and the Youth Risk Behavior Survey found that 33% of youth aged 12 to 20 years reported drinking some alcohol in the past 30 days with 13% to 18% of those binge drinking (Abuse, 2016; Kann et al., 2016). As of 2015, about 9 of every 100 U.S. high school students (9.3%) reported tobacco smoking cigarettes in the past 30 days (Abuse, 2016; Singh et al., 2016). Overall, U.S. adolescent boys report slightly higher tobacco smoking and alcohol use relative to adolescent girls, however there is significant variability in tobacco smoking and alcohol use rates depending on adolescent age range and race (Abuse, 2016; Kann et al., 2016). In the case that current tobacco smoking rates continue, an estimated 5.6 million Americans under the age of 18 are at risk of dying prematurely from a smoking-related disease (Health & Services, 2014).

While alcohol use and tobacco smoking can lead to negative health outcomes in their own right, they often develop simultaneously, with tobacco smoking being particularly common amid youth treated for alcohol disorders (Myers & Kelly, 2006). There is mounting evidence of the deleterious health consequences that are the result of alcohol and tobacco comorbid use disorders (Castellsagué et al., 1999; Hurley, Taylor, & Tizabi, 2012; Hurt et al., 1996; Koob, 2011; Pelucchi, Gallus, Garavello, Bosetti, & La Vecchia, 2006). For example, use of alcohol and nicotine together increases risk for developing cancers of the digestive and upper respiratory tracts, mouth, throat, esophagus, and larynx (Bagnardi, Blangiardo, Vecchia, & Corrao, 2001; Talamini et al., 1998; Taylor & Rehm, 2006). Preliminary observations have also shown that tobacco’s cardiovascular effects may exacerbate alcohol-induced cerebrovascular pathology (Aberle, Privratsky, Burd, & Ren, 2003; Iida, Iida, Fujiwara, & Dohi, 2003; Mukamal, 2006).

Tobacco and alcohol both exert reinforcing effects in part by increasing dopamine release through the mesolimbic pathway and into the nucleus accumbens (Söderpalm, Ericson, Olausson, Blomqvist, & Engel, 2000) by stimulating nicotinic acetylcholine receptors (Chatterjee & Bartlett, 2010; Gianoulakis, 2004; Nocente et al., 2013; Koob & Bloom, 1988). Importantly, there is a “cross-tolerance” effect of tobacco and alcohol when used together, which leads to higher levels of each drug needed to experience similar levels of reward over time (Burch, de Fiebre, Marks, & Collins, 1988; Collins, Burch, de Fiebre, & Marks, 1988; Collins, Romm, Selvaag, Turner, & Marks, 1993; Collins, Wilkins, Slobe, Cao, & Bullock, 1996; de Fiebre & Collins, 1993). At the same time, course may mitigate short-term deleterious consequences that could further expedite a path toward coaddiction, for example, smoking a cigarette to stimulate oneself when experiencing the sedative effects of high levels of alcohol consumption.

The primary aim of this investigation was to examine differences in a longitudinal, parallel change of alcohol and tobacco course among an adolescent population as they transitioned into young adulthood (i.e., ages of 15 to 32 years old). Given previous evidence on the importance of understanding sex differences in alcohol consumption and tobacco smoking (Faeh, Viswanathan, Chiolero, Warren, & Bovet, 2006; Lotrean, Kremers, Ionut, & de Vries, 2009; Lotrean, Mesters, Ionut, & de Vries, 2009; Weitzman & Chen, 2005; Wetzels, Kremers, Vitória, & de Vries, 2003), we examined the impact of sex specifically. A parallel change model of alcohol use and tobacco smoking has not been examined in this way, nor with respect to sex as a key covariate along with other relevant predictors of course over time, such as personality traits, and physical health measures. Personality appears to play a key role in explaining tobacco smoking and alcohol drinking behavior in adolescence and young adulthood (Heinrich et al., 2016; Yóñez, Leiva, Estela, & Čukić, 2017). For example, higher levels of neuroticism and extraversion are associated with higher risk of initiating smoking or being a current smoker in early adolescence (Yáñez et al., 2017). Moreover, personality is a strong driver of alcohol use initiation in adolescence (Heinrich et al., 2016), particularity impulsivity, extraversion, and sensation seeking traits. Controlling for general health is also vital when examining longitudinal tobacco smoking and alcohol use in adolescence and young adulthood, as physical measures such as BMI, blood pressure, and lipid levels are important confounders for these negative health behaviors (Jerez & Coviello, 1998; Oesterle et al., 2004; Wakabayashi, 2009).

Much of the previous research in this area is limited to regression methodologies (e.g., logistic regression) and limited longitudinal data (J. S. Brook, Lee, Rubenstone, Brook, & Finch, 2014; Faeh et al., 2006; Lotrean, Kremers, et al., 2009; Lotrean, Mesters, et al., 2009; Weitzman & Chen, 2005; Wetzels et al., 2003), while this study utilizes parallel latent growth modeling, a technique that allows us to examine patterns of change, and within-person variation in this change for alcohol and tobacco course. This technique also allows us to evaluate the association of initial status (i.e., intercept) with pattern of change (i.e., slope) in alcohol and tobacco course, as well as their association with time-invariant covariates (Preacher, 2008). In addition, while several studies examined sex differences in tobacco smoking and alcohol use in adolescents utilizing some form of longitudinal data (Brook et al., 2014; Lotrean, Kremers, et al., 2009; Lotrean, Mesters, et al., 2009; Wetzels et al., 2003), this study is unique in that it follows participants over a 17-year period (Wijnstok, Hoekstra, van Mechelen, Kemper, & Twisk, 2013).These data provide a unique perspective on the changing relationship between alcohol and tobacco course throughout two segments of the life span, and what predicts patterns of change in these two different segments of development.

Method

Sample

In 1976, individuals were initially recruited from two secondary schools in the Netherlands for a prospective, epidemiologic investigation examining natural growth and disease across the life span: the Amsterdam Growth and Health Longitudinal Study (AGHLS; Bernaards, Kemper, Twisk, van Mechelen, & Snel, 2001; Kemper et al., 1997; Koppes, Kemper, Post, Snel, & Twisk, 2000; Koppes, Twisk, Snel, Van Mechelen, & Kemper, 2000). Once primary school started and the investigators waited to begin collecting data at the end of primary school, students were not available for data collection for reasons including moving away or moving to a different school. By the time students completed primary school, several participants were no longer available for data collection. Previous papers have examined this cohort for selective dropout and have not found that there is evidence of this potential problem (Hoekstra, Barbosa-Leiker, Koppes, & Twisk, 2011; Hoekstra, Barbosa-Leiker, & Twisk, 2013; Twisk & de Vente, 2002; van de Laar et al., 2012). Moreover, inclusion of a broad set of covariates, and examining the outcomes over time, significantly improves the ability of our estimation procedure that uses maximum likelihood to appropriately handle missing data (McPherson et al., 2015; McPherson, Barbosa-Leiker, McDonell, Howell, & Roll, 2013). Data for this specific investigation come from a prospective cohort sample of participants (n = 303 for ages 15 to 21, n = 196 for ages 21 to 32; 52% female and 54% female, respectively) enrolled in AGHLS and include all participants that were still available for data collection. The sample of n = 196 for ages 21 to 32 is a subsample of the n = 303 for ages 15 to 21. Further details about the AGHLS cohort are provided elsewhere (Wijnstok et al., 2013).

Data Collection

Collection of data has evolved over time, but most visits included a standardized, full battery of self-report questionnaires and blood and urine specimens collected for analysis at each study visit, with some extra assessments performed that were project-specific. As of 2006, over the course of 35 years, 10 rounds of measurement had been gathered in this observational, longitudinal cohort study (Wijnstok et al., 2013). For this investigation, we were able to make use of seven measurement occasions due to data completeness across the outcomes and covariates we needed to include and also because the alcohol use and tobacco smoking levels were very low prior to the cohort turning 15 years old. Given our interest in examining the role of sex, a time-invariant covariate, we wanted to compare its ability to predict change over time relative to other time-invariant covariates, such as baseline body mass index (BMI), personality characteristics, blood pressure and lipid levels.

Approval for this study was obtained from the Medical Ethics Committee of the Vrije Universiteit Amsterdam University Medical Centre and written informed consent was obtained from each participant, as well as from the parent or guardian (during adolescence).

Measurement

Outcomes.

Measurements of tobacco and alcohol use were gathered using self-report measures. Between the ages of 13 and 16 years, participants shared information about tobacco smoking during a confidential interview and from ages 21 onward through comprehensive self-report questionnaires. Alcohol use was measured with a crosscheck dietary history interview covering the month prior to the interview. The average amount and number of alcoholic beverages consumed were converted to grams per week of consumption. Alcoholic beverages >16 g per 100 g of the beverage were considered as “spirits”, and beverages with <16 g of alcohol content (almost 95% true wine, 5% fortified wine, 1% others) as well as beverages not reported as “beer” were classified as “wine”. While a distinction can be made between “non-users” [participants consuming <10 g of alcohol (approximately one unit) per week], light drinkers (10 to 50 g/week), moderate drinkers (50 to 100 g/week), and heavier drinkers (>100grams/week; Koppes, Kemper et al., 2000), we treated alcohol use as a continuous variable. At each measurement year, participants also filled out a survey on their cigarette use, own-rolled tobacco, cigars/cigarillos, and pipe tobacco. Tobacco smoking was expressed in total grams of tobacco smoked per week (1 cigarette = 1 g, 1 package of own-rolled tobacco = 40g, 1 cigar/cigarillo = 3 g, 1 package of pipe tobacco = 50g). While participants could be classified as nonsmokers (<1 g tobacco per day), light smokers (1 to 10 g of tobacco per), or moderate to heavy smokers (≥10 g tobacco per day; CLAIRE M Bernaards, Twisk, Van Mechelen, Snel, & Kemper, 2003), we treated tobacco use as a continuous variable.

Covariates.

Sex, BMI, personality characteristics, and all other covariates were collected at baseline. BMI, blood pressure, and lipid levels were used as covariates in order to control for the general health of the study participants. BMI was calculated by dividing weight in kg by height in meters and systolic blood pressure was measured manually (average of three measurements) during adolescence, and in a horizontal position using an automated device (Dinamap ProCare 100, Milwaukee, WI) during young adulthood. A trained research assistant collected the BMI data. Fasting cholesterol levels were calculated from blood drawn from the antecubital vein (Roche Diagnostics, Mannheim Germany). We also controlled for baseline personality traits as ample research shows that personality is highly predictive of tobacco smoking and alcohol use in adolescents (Brook et al., 2008; Heinrich et al., 2016; Stautz & Cooper, 2013; Yáñez et al., 2017). Five personality domains of Inadequacy, Social inadequacy, Rigidity, Recalcitrance, and Dominance were scored based on the youth version of the validated Dutch Personality Inventory (DPI; Luteijn, Starren, & Van Dijk, 1985), with higher scores indicating higher levels of the domain. Inadequacy refers to having physical complaints, anxiety, feelings of depression and insufficiency; social inadequacy refers to avoiding social contacts, and having uncomfortable feelings in social situations; rigidity reflects the need for regularity; recalcitrance reflects distrust of others, desire to criticize others, and solving problems by oneself; dominance refers to wanting to take leadership, and having confidence in one’s own abilities (Koppes et al., 2008). The DPI scale has been previously shown to have acceptable validity and reliability, with Cronbach’s alpha ranging from 0.70 to 0.87 (Luteijn et al., 1985; Uijtdewilligen et al., 2011).

Statistical Analysis

We utilized parallel latent growth modeling (McPherson et al., 2016; via maximum likelihood, a method used to assess change in two related outcomes over time) to assess the impact of several prespecified covariates on tobacco and alcohol use over time. Several of the physiological variables, such as blood pressure and cholesterol levels were included in order to control for the overall health of the participants. The covariates included in the model were the following: sex, BMI, systolic and diastolic blood pressure, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol and the 5 personality traits noted above on initial status (i.e., intercept) and linear change (i.e., slope) in tobacco and alcohol (both measured in grams/week) use over time (Muthén & Muthén, 1998–2010). This methodology allows for the simultaneous estimation of two outcomes, and for their random effects, both at initial status and linear change (or other patterns) over time, to covary, thereby allowing us to estimate the random effects’ associations. In addition, parallel latent growth modeling in this context holds constant the residual covariances between the intercept and slope random effects of alcohol and tobacco use.

We assessed the fit of the growth models with numerous robust goodness of fit indices. The chi-square statistic, which evaluates the absolute fit of the hypothesized model to the data is sensitive to sample sizes (Hu & Bentler, 1999), therefore, we used additional fit indices. The comparative fit ndex (CFI), the standardized root mean square residuals (SRMR), and the root mean square error of approximation (RMSEA). Substantial research recommends a CFI value close to or above .95, an SRMR value close to or below. 06 and an RMSEA value close to or below .08, as indicative of good model fit (Hu & Bentler, 1999; Marsh, Hau, & Wen, 2004).

We analyzed data spanning a total of 17 years across the life span, and broke up our analyses into two different time periods and thus two different models. The first period was adolescence starting at age 15 and ending at age 21. The second period was young adulthood starting at age 21 and ending at age 32. We used age 15 initial status, or baseline covariates to predict change in our outcomes, and age 21 as the baseline for the young adulthood model. The two models represent natural break points in the data given the almost complete reversal from a steady increase in substance use in adolescence to a steady decrease in substance use in young adulthood. We attempted to fit a single trajectory for the entire life span, but this introduced several convergence problems, and more importantly, made it more difficult to full a part the unique findings associated with the different periods in one’s life. We initially included the complete set of covariates listed previously, but our final models only included covariates that were significantly associated with at least one of the four random effects in each of the models run. Figure 1 is a conceptual depiction of how the two outcomes were regressed onto the final set of covariates.

Figure 1.

Parallel latent growth model of alcohol and tobacco use across two lifetime segments (Note: BMI = body mass index; HDL = high density lipoprotein; BP = blood pressure; Growth factors only shown, not individual time points).

Missing data were addressed using maximum likelihood, a method consistent with current expert recommendations (Enders, 2010; Little & Rubin, 2002; McPherson, Barbosa-Leiker, Burns, Howell, & Roll, 2012; McPherson et al., 2013). All statistical analyses were conducted using Mplus 7.4 (Muthén & Muthén, 1998–2010) and SPSS 24 (IBM Corp. 2017). We used an alpha threshold of 0.05 to determine statistical significance and included 95% confidence intervals around our unstandardized results reported.

Results

Adolescence Life Period

Model fit was excellent for the adolescent life period. The chi-square statistics was not statistically significant, and all of the subjective fit indices met recommended criteria (CFI = 0.995, SRMR = 0.019, and RMSEA = 0.027 [90% CI 0.000, 0.060]), suggesting that modeling linear change was appropriate within this age group. We observed significant sex differences such that females exhibited less alcohol use at adolescent baseline (age 15: β = −21.79, p < .05) and less increase in alcohol use during adolescence (β = −7.92, p < .05; see Table 1). Figure 2 depicts patterns of alcohol use and tobacco smoking during adolescence, broken up across sex. In addition, those with a higher BMI exhibited higher initial levels of alcohol use (age 15: β = 5.50, p < .05), while those higher on the Rigidity and Social Inadequacy personality traits had lower initial levels of alcohol use (age 15 Rigidity: β = –0.72, p < .05; age 15 Social Inadequacy: β = −1.24, p < .05). In addition, systolic blood pressure was a significant predictor of linear change for alcohol use at adolescence (β = −0.22, p < .05). While there were no direct significant differences across sex on tobacco smoking, those with higher BMI exhibited higher initial levels of tobacco smoking at adolescent baseline (age 15: β = 3.14, p < .05), and those higher on the Social Inadequacy trait exhibited lower initial levels of tobacco smoking at adolescent baseline (age 15: β = −0.43, p < .05). Systolic blood pressure was also a significant predictor of tobacco smoking linear change at adolescence (β = −0.19, p < .05). There were no other statistically significant covariate associations with alcohol and tobacco smoking course over time during adolescence.

Table 1.

Parallel Random Linear Growth Model of Alcohol Use and Tobacco Smoking Ages 15 to 21 in the Amsterdam Growth and Health Longitudinal Study

Covariate	Outcome	Unstandardized estimate (β)	95% CI	p
Sex →	SM intercept	−3.592	[−8.858, 1.674]	.181
BMI →	SM intercept	3.143	[1.994, 4.292]	.000
Rigidity →	SM intercept	−.003	[−.305, .299]	.984
Social inadequacy →	SM intercept	−.433	[−.817, −.050]	.027
HDL Cholesterol →	SM intercept	5.581	[−4.489, 15.652]	.277
Systolic BP →	SM intercept	−.222	[−.469, .026]	.079
Sex →	AL intercept	−21.793	[−33.814, −9.773]	.000
BMI →	AL intercept	5.499	[2.883, 8.116]	.000
Rigidity →	AL intercept	−.715	[−1.404, −.026]	.042
Social inadequacy →	AL intercept	−1.241	[−2.115, −.368]	.005
HDL cholesterol →	AL intercept	18.145	[−4.767, 41.056]	.121
Systolic BP →	AL intercept	.367	[−.198, .932]	.203
Sex →	SM slope	−3.114	[−7.053, .826]	.121
BMI →	SM slope	.213	[−.574, 1.000]	.596
Rigidity →	SM slope	−.166	[−.397, .064]	.157
Social inadequacy →	SM slope	−.136	[−.418, .145]	.343
HDL cholesterol →	SM slope	.520	[−7.486, 8.525]	.899
Systolic BP →	SM slope	−.215	[−.399, −.032]	.021
Sex →	AL slope	−7.922	[−11.972, −3.872]	.000
BMI →	AL slope	.139	[−.675, .954]	.737
Rigidity →	AL slope	−.032	[−.268, .205]	.792
Social inadequacy →	AL slope	.156	[−.132, .443]	.288
HDL cholesterol →	AL slope	−1.799	[−9.950, 6.352]	.665
Systolic BP →	AL slope	−.189	[−.377, −.001]	.049

Note. Single-headed arrows (i.e., →) represent specified regression paths; p < .05 = bold text. Sex = 0 = males, 1 = females. SM = smoking; AL = alcohol; BMI = body mass index; HDL = high density lipoprotein; BP = blood pressure.

Figure 2.

Sex differences in alcohol use and tobacco smoking trajectories among 15 to 21 year olds in the Amsterdam Growth and Health Longitudinal Study (n = 303).

Young Adulthood Life Period

Model fit was excellent for the young adulthood life period. The chi-square statistics was not statistically significant, and all of the subjective fit indices met recommended criteria (CFI = 0.998, SRMR = 0.056, and RMSEA = 0.009 [90% CI 0.000, 0.051]), suggesting that modeling linear change was appropriate within this age group. We observed significant sex differences such that females exhibited a steeper decline in tobacco smoking during young adulthood (β = −70.85, p < .05) and less decrease in alcohol use during young adulthood (β = 4.67, p < .05; see Table 2). However, as shown in Figure 3, females demonstrated lower mean levels of alcohol use and tobacco smoking at age 21. Those higher on the Rigidity personality trait demonstrated a steeper decline in tobacco smoking decrease during young adulthood (β = −2.24, p < .05). Those with higher high density lipoprotein (HDL) cholesterol, and systolic blood pressure, exhibited lower initial levels of tobacco smoking at young adulthood baseline (age 21 HDL: β = −49.40, p < .05; age 21 systolic blood pressure: β = −1.33, p < .05). Systolic blood pressure was also significantly associated with higher level of alcohol use at young adulthood baseline (β = 0.10, p < .05). There were no other statistically significant covariate associations with alcohol and tobacco smoking course over time during young adulthood.

Table 2.

Parallel Random Linear Growth Model of Alcohol Use and Tobacco Smoking Ages 21 to 32 in the Amsterdam Growth and Health Longitudinal Study

Covariate	Outcome	Unstandardized estimate (β)	95% CI	p
Sex →	SM intercept	− 16.126	[−43.949, 11.697]	.256
Rigidity →	SM intercept	− 1.383	[−2.935, .169]	.081
HDL cholesterol →	SM intercept	−49.401	[−93.852, −4.949]	.029
Systolic BP →	SM intercept	−1.326	[−2.319, −.334]	.009
Sex →	AL intercept	1.157	[−1.406, 3.721]	.376
Rigidity →	AL intercept	.050	[−.078, .178]	.442
HDL cholesterol →	AL intercept	3.685	[−.363, 7.732]	.074
Systolic BP →	AL intercept	.103	[.014, .193]	.024
Sex →	SM slope	−70.845	[−100.522, −41.168]	.000
Rigidity →	SM slope	− 2.238	[−3.509, −.968]	.001
HDL cholesterol →	SM slope	37.954	[−2.152,78.060]	.064
Systolic BP →	SM slope	−.700	[−1.723, .323]	.180
Sex →	AL slope	4.673	[.488, 8.858]	.029
Rigidity →	AL slope	.034	[−.144, .211]	.711
HDL cholesterol →	AL slope	−2.811	[−7.888, 2.266]	.278
Systolic BP →	AL slope	.040	[−.095, .176]	.561

Note. Single-headed arrows (i.e., →) represent specified regression paths; p < .05 = bold text. Sex = 0 = males, 1 = females. SM = smoking; AL = alcohol; BMI = body mass index; HDL = high density lipoprotein; BP = blood pressure.

Figure 3.

Sex differences in alcohol use and tobacco smoking trajectories among 21- to 32-year-old adults in the Amsterdam Growth and Health Longitudinal Study (n = 196).

Young Adulthood and Adolescence: Tobacco and Alcohol Use Associations

There were several associations between the residual random effect covariances of alcohol use and tobacco smoking (see Table 3 for the residual covariances, and Table 4 for the residual correlations), including a significant, positive relationship between amount of linear change in tobacco smoking and linear change in alcohol use during adolescence (β = 30.62, p < .05). Tobacco smoking initial status, alcohol use initial status, alcohol use change over time and tobacco smoking change over time were all significantly associated with each other during adolescence. During young adulthood, all of these relationships remained significant, except the relationship between alcohol use at initial status and alcohol use change over time (p > .05).

Table 3.

Residual Covariances of Intercepts and Slopes in the Amsterdam Growth and Health Longitudinal Study

Residual random effect covariance		Unstandardized estimate (β)	95% CI	p
Ages 15 to 21
Alcohol intercept ↔	Alcohol slope	−175.997	[−267.929, −84.066]	.000
Smoking intercept ↔	Smoking slope	49.635	[8.971, 90.298]	.017
Alcohol intercept ↔	Smoking intercept	641.898	[507.431, 776.366]	.000
Alcohol slope ↔	Smoking slope	30.624	[6.666, 54.581]	.012
Alcohol intercept ↔	Smoking slope	85.522	[8.639, 162.404]	.029
Alcohol slope ↔	Smoking intercept	−73.413	[−110.830, −35.995]	.000
Ages 21 to 32
Alcohol intercept ↔	Alcohol slope	−102.921	[−328.573, 122.730]	.371
Smoking intercept ↔	Smoking slope	−224.727	[−412.700, −36.754]	.019
Alcohol intercept ↔	Smoking intercept	2443.813	[1363.982, 3523.645]	.000
Alcohol slope ↔	Smoking slope	12.656	[1.778, 23.535]	.023
Alcohol intercept ↔	Smoking slope	−152.549	[−256.890, −48.209]	.004
Alcohol slope ↔	Smoking intercept	−130.113	[−239.249, −20.977]	.019

Note. Double–headed arrows (i.e., ↔) indicate residual covariance between random effects. p < .05 = bold text.

Table 4.

Residual Correlations of Intercepts and Slopes in the Amsterdam Growth and Health Longitudinal Study

Residual random effect correlation		Standardized estimate	(SE)	p
Ages 15 to 21
Alcohol intercept ↔	Alcohol slope	−.526	(.145)	.000
Smoking intercept ↔	Smoking slope	.259	(.123)	.035
Alcohol intercept ↔	Smoking intercept	.778	(.044)	.000
Alcohol slope ↔	Smoking slope	.394	(.176)	.025
Alcohol intercept ↔	Smoking slope	.188	(.088)	.032
Alcohol slope ↔	Smoking intercept	−.520	(.185)	.005
Ages 21 to 32
Alcohol Intercept ↔	Alcohol slope	− .542	(.260)	.037
Smoking Intercept ↔	Smoking slope	− .875	(.102)	.000
Alcohol Intercept ↔	Smoking intercept	.768	(.183)	.000
Alcohol Slope ↔	Smoking slope	.826	(.575)	.151
Alcohol Intercept ↔	Smoking slope	− .802	(.377)	.033
Alcohol Slope ↔	Smoking intercept	−.508	(.340)	.135

Note. Double-headed arrows (i.e., ↔) indicate residual correlations between random effects. p < .05 = bold text.

Discussion

The sex differences observed in this sample suggest that males, relative to females, demonstrate significantly higher rates of alcohol use and greater increase over time, especially during adolescence. In light of the strong relationship observed between alcohol and tobacco course, particularly from ages 15 to 21, our data suggest that prevention efforts to curb heavy alcohol use should consider targeting tobacco use concurrently, and earlier. However, it is also important to note that females during young adulthood showed increases in alcohol consumption and decreases on smoking. Addressing course through prevention and treatment efforts is in need of additional research to clarify how best to proceed with such work that recognizes sex differences. The trends over time and differences between males and females reported here for drinking and smoking during adolescence and early adulthood are consistent with more recent reports out of the Netherlands and the United States (Abuse, 2016; Central Bureau of Statistics, 2016; Kann et al., 2016; Verhagen et al., 2015).

Other factors, such as adolescent levels of rigidity and social inadequacy are also important to factor into future preventions efforts. Those males who are less rigid and/or experience less social inadequacy may represent novel targets for prevention. In addition, higher BMI, and lower HDL cholesterol were associated with more alcohol use in adolescence and less smoking in young adulthood, respectively. This may reflect important differences in overall health that could be indicate that those with worse health engage in more drinking and smoking across both of these time periods. The blood pressure findings are in need of additional study since it is an unexpected finding that higher blood pressure was associated with lower levels of alcohol use and smoking. In adulthood, for example, high blood pressure is a consistent finding among those that smoking and use alcohol heavily (Aberle et al., 2003; Iida et al., 2003; Mukamal, 2006). One possible explanation is that high blood pressure at this early age is indicative of other health problems the individual in question could be experiencing, which led them to avoid smoking and drinking at an early age on the advice of their health care provider. Rather than speculate further, we note this inconsistency with previous literature which points to a need to examine this effect in other, future studies as this finding might be an isolated aberration. Nevertheless, these findings will inform future prevention efforts to reduce down-stream alcohol and nicotine coaddiction.

There is currently a dearth of existing combined treatment approaches for alcohol use and co-occurring tobacco smoking. During use of alcohol and tobacco, one drug may serve as a cue for the expectancy of use for the other (Field, Mogg, & Bradley, 2005; Sayette, Martin, Wertz, Perrott, & Peters, 2005), which can augment craving and promote relapse (Field et al., 2005; Sayette et al., 2005; Shiffman et al., 1996). In fact, use of tobacco can reestablish alcohol-seeking responses that were once extinct (Lê, 2002; Lê, Wang, Harding, Juzytsch, & Shaham, 2003). Our data suggest that this may be particularly true among males. However, there is some data suggesting that sequencing alcohol treatment prior to tobacco smoking cessation treatment can produce better outcomes (Kalman, Kahler, Garvey, & Monti, 2006; Marks, Hill, Pomerleau, Mudd, & Blow, 1997). Two consistent conclusions from studies of tobacco smoking cessation in the context of alcohol treatment have been: 1) the integration can mutually enhance patients’ ability to abstain from both drugs providing a critical window of opportunity (Kalman, Kim, DiGirolamo, Smelson, & Ziedonis, 2010), and 2) the tobacco smoking intervention must be high intensity (i.e., standalone therapy) to be effective (Bobo, McIlvain, Lando, Walker, & Leed-Kelly, 1998; Kalman et al., 2006; Kalman et al., 2010; Kohn, Tsoh, & Weisner, 2003; O’Malley et al., 2006). While alcohol treatment programs recognize the need for incorporating polysubstance abuse treatment (Kalman et al., 2010), tobacco use disorder treatment is a consistent exception, even though many alcohol addicted patients are concerned about their tobacco smoking (Rohsenow, Colby, Martin, & Monti, 2005).

In a long-running epidemiological cohort study such as this, data completeness is an ongoing concern as people drop out, move, or otherwise become unavailable for data collection. While there are gaps in that we were only able to capture 7 out of 10 possible measurement occasions, we still view these data as amenable for a clear analysis that can make an important contribution to the literature on alcohol and tobacco course. Another weakness of this investigation lies in its relatively healthy, educated and somewhat homogeneous study sample, possibly hindering generalizability to a more heterogeneous population. Nevertheless, in the case that small effects are detected in this healthy sample, these findings may suggest greater effects in a more heterogeneous study population, and therefore should not be overlooked.

In summary, our investigation revealed strong relationships between baseline alcohol and tobacco course, and linear change in alcohol and tobacco course among adolescents and young adults. Additionally, sex differences observed in this sample suggest that relative to females, males exhibit significantly higher rates of alcohol use and greater increase over time, particularly during adolescence. These data add to the growing recognition concerning the importance of alcohol and tobacco course, and strengthen our understanding of the relationship between alcohol and tobacco use over time. The public health risks associated with alcohol and tobacco use highlight the need to develop treatment programs targeting course. These findings offer further understanding of the predictors and patterns of alcohol and tobacco course and will aid in developing more effective interventions for treatment of alcohol and tobacco coaddiction.

Public Health Significance.

This study exposed strong relationships between baseline alcohol and tobacco course, and linear change in alcohol and tobacco course among adolescents and young adults. Males exhibited significantly higher rates of alcohol use and larger increase over time, particularly during adolescence. Prevention efforts to reduce heavy alcohol use should consider targeting tobacco use concurrently, and should do this earlier among males.