Behavioral Economic Demand and Delay Discounting Are Differentially Associated with Cigarette Dependence and Use in Adolescents

Abstract

Introduction:

The Reinforcer Pathology Model describes how two behavioral economic processes, increased sensitivity to immediate rewards (delay discounting) and excessive reward derived from a substance (demand), both contribute to problematic patterns of substance use. In a novel application of this model, the current cross-sectional study examined how these distinct processes relate to different facets of cigarette use in adolescents.

Methods:

Adolescent daily cigarette smokers ages 15 to 19 (Mean age 17.7, N = 50) completed a laboratory assessment of demand using a Cigarette Purchase Task for their usual brand cigarettes and an adjusting-amount delay discounting task. Demand was conceptualized as two factors (Amplitude and Persistence) and delay discounting was calculated as Area Under the Curve (log AUC). The two factors of demand and discounting AUC were included as statistical predictors of level of cigarette dependence and average number of cigarettes smoked per day in linear regression models.

Results:

Amplitude of demand was marginally significant predictor (p = .06) of cigarettes smoked per day whereas neither Persistence of demand nor delay discounting significantly predicted this outcome. Both Amplitude of demand and delay discounting, but not Persistence, were associated with level of cigarette dependence. The effects of amplitude of demand and delay discounting on cigarette dependence or use did not significantly interact.

Conclusions:

Results of this study suggest that amplitude of cigarette demand may be a risk factor for both cigarette consumption and dependence, while discounting – a known risk factor for cigarette initiation – may relate specifically to level of dependence.

1. INTRODUCTION

Cigarette smoking by adolescents has been reduced to its lowest rate in decades, with only 9% of adolescents reporting daily smoking in 2016¹. Declines in youth smoking prevalence affected by tobacco control interventions such as educational mass media campaigns appear to be primarily accounted for by declines in initiation versus smoking cessation². Therefore, despite reduced prevalence, 5.6 million current underage smokers are projected to die prematurely of tobacco related disease³. Furthermore, the rise in prevalence of e-cigarette use has been shown to be associated with later cigarette initiation^4,5; thus, it is possible that these reductions in initiation may be reversed in the future. Given these sobering statistics, research on mechanisms of smoking behavior in youth continues to be important.

The Reinforcer Pathology Model describes how two behavioral economic processes, increased sensitivity to immediate rewards (delay discounting) and over-valuation of drug reinforcement (demand; which may be assessed via drug purchase tasks), jointly increase problematic patterns of substance use⁶. The model proposes that those with substance use disorders may continually experience the combined effects of high substance demand and steep discounting of delayed rewards. This manifests as an elevated substance value paired with a preference for receiving and using the substance of choice immediately⁷. Together, elevated levels of substance demand and steep delay discounting are theorized to be persistent markers contributing to the progression of use and the development of substance use disorders⁶. However, how these distinct mechanisms relate to different facets of cigarette use in adolescents has not been established. Individuals vary in their phenotypic presentations of these factors, such that a given individual may have high demand for cigarettes but low delay discounting overall; while another may have high discounting overall but only moderate demand for a specific substance. Likewise, cigarette consumption and dependence represent related but conceptually distinct constructs⁸. Consumption is a feature of dependence, but dependence is also characterized by physical tolerance, abstinence-induced symptoms of withdrawal, and use despite environmental restrictions⁹. The Reinforcer Pathology model suggests that distinct facets of addiction may relate differentially to demand and discounting. Specifically, demand is posited to be a recursive process, which may change as demand is assessed, and demand often changes as exposure increases¹⁰. In contrast, delay discounting is theorized to be a trans-disease process that marks a trait-like individual difference factor which predisposes certain individuals to initiate and maintain drug use^6,11.

The extent to which discounting and demand may be differentially related to drug consumption and dependence has recently begun to be explored. In two studies of adults using cannabis, frequency of consumption was specifically significantly associated with demand for cannabis, while cannabis dependence was uniquely associated with discounting^12,13, indicating that these two related processes may differentially relate to severity of dependence and frequency of use. Demand and discounting have also been explored in adolescent smokers. For example, adolescent smokers, even those who reported just having started to experiment with cigarettes, were found to discount delayed rewards more steeply than their never-smoking peers^14–16. Other studies suggest that steep discounting is a heritable risk factor for smoking initiation^17,18. Among adolescents, demand for cigarettes has been correlated with both cigarettes smoked per day (CPD) and dependence severity in smokers^19,20, indicating that demand may measure a broader facet of cigarette use in adolescents who smoke.

Taken as a whole, the literature suggests that delay discounting is a risk factor for smoking initiation in adolescents, and that demand is associated with heavier use (greater CPD) and higher levels of cigarette dependence; however, the unique and overlapping associations of demand and delay discounting have not been concurrently explored in the context of a single sample of young smokers. Thus, we aimed to test hypotheses based on the Reinforcer Pathology model in a sample of adolescent smokers, including whether indicators of demand and delay discounting would be differentially associated with cigarette dependence severity and level of daily cigarette use. We hypothesized that greater Amplitude of demand would be associated with both greater dependence and greater smoking, and that steeper discounting would be associated with greater dependence but not greater smoking. We also examined whether Persistence of demand was associated with dependence and smoking, but did not have specific hypotheses. Based on prior studies with cannabis, we hypothesized that demand and discounting would not significantly interact in the models.

2. METHODS

2.1. Participants.

Participants were recruited from the community using online advertisements, advertisements on local buses, and via in-person recruiting sessions at local high schools. Following a phone screen to determine initial eligibility, 50 adolescent (15-19 years old, inclusive) daily smokers attended an in-person baseline session to collect information for a 4-session laboratory study of reduced nicotine content cigarettes²¹. All measures were collected at that baseline session, which took about 2 hours to complete. Inclusion criteria were smoking at least 1 cigarette a day for the past 6 months, not being pregnant, not currently seeking treatment for smoking, and not having current suicidal ideation or a history of suicide attempts. Participants were excluded if they reported daily use of alcohol or other drugs (except marijuana, where daily use was acceptable for inclusion). No formal diagnostic criteria for nicotine dependence needed to be met for inclusion. Further information on the sample can be found in the parent study ²² and information on demand indices can be found elsewhere²³. All procedures were approved by the Brown University Research Protections Office prior to implementation.

2.2. Measures.

2.2.1. Demographics.

Age, gender, race and ethnic background, disposable income, and tobacco use history variables were assessed using a demographics questionnaire. Gender was included as a covariate as gender differences in adolescent cigarette use exist²⁴. Smoking duration (current age in years minus age of smoking first whole cigarette in years) was also included as a covariate as this has been related to dependence symptoms^25–26. Breath carbon monoxide (CO) was also tested using a Smokerlyzer ED50 CO meter (Bedfont Instruments) as a measure of recent cigarette smoking.

2.2.2. Disposable Income.

Disposable income was calculated as the sum of the monetary values entered for 3 questions: “How much money do your parents give you each week?”, “How much money do you earn at your job each week?”, and “How much money do you get from other sources each week?”, less the amount stated for “How much money do you give back to your parents each week?” Income was included as a covariate as it can affect demand²⁷.

2.2.3. Cigarettes per day.

Using a Time Line Follow-Back interview, which has been validated for use in adolescents²⁸, participants reported on their cigarettes smoked per day for the last 30 days. Average number of cigarettes smoked per day (CPD) was calculated as the total number of cigarettes smoked divided by 30.

2.2.4. Modified Fagerström Tolerance Questionnaire.

This measure of cigarette dependence has been validated for use in adolescents and is adapted from the Fagerström Tolerance Questionnaire²⁵. The mFTQ has been validated for use in adolescent smokers, and includes a categorical item measuring cigarettes per day (less than 1 per day, 1-15 per day, 16-25 per day, or over 26 per day). The questionnaire includes 6 further questions querying whether the adolescent inhales their cigarettes, how soon they smoke upon waking, which cigarette they would least like to give up, difficulty not smoking where it is not allowed, smoking while ill, and smoking in the morning. The scale ranges from 2 to 8, with a 4 representing moderate dependence.

2.2.5. Cigarette Purchase Task.

To determine demand for their usual brand cigarettes, adolescents were asked how many cigarettes they would purchase in a 24-hour period across a range of prices per cigarette²⁹. Participants are asked to imagine that they have no other access to sources of cigarettes other than the ones presented, and that they could not save cigarettes for later use.

2.2.6. Delay Discounting Task.

Using a computerized adjusting-amount discounting task³⁰, participants were asked to make a choice by clicking the button associated with one of two options: $100 available after a delay, or an amount of money available now. The delayed amount remained at $100 across all trials, while the 8 delay values (1 week, 2 weeks, 1 month, 4 months, 8 months, 1 year, 5 years, and 10 years) varied randomly across blocks of trials. The amount of money available now started at either $0 or $100 (selected randomly), and adjusted downward by 15% if the “now” value was chosen on that trial, and upward by the same percentage if the delayed value was chosen.

2.3. Data Analysis.

2.3.1. Area Under the Curve (AUC) Value calculation for discounting.

Data from the discounting task were analyzed using an atheoretical approach known as Area Under the Curve (AUC) to calculate individual rates and discounting³¹. The delay values were first log-transformed prior to calculating AUC to correct for skewness and ensure a normal distribution of values. After normalizing the indifference points (with respect to the delayed amount of $100) and the delay values (with respect to the highest delay, ten years), each successive pair of points along the curve was treated as the four corners of a distinct, trapezoidal area and then summed. AUC values, unlogged, can range from 1 (maximum demand/no discounting) to 0 (no demand/maximum discounting). For discounting, higher AUC values represent greater self control.

2.3.2. Factor analysis of demand indices.

Further details on relationships between specific demand indices in this sample can be found elsewhere²³. To reduce the number of demand indices included in subsequent models, exploratory factor analysis (EFA) was utilized. EFA was conducted using a principal component analysis (PCA) method of estimation with oblique (oblimin) rotation and was specified to be constrained to a two-factor solution (i.e., Persistence and Amplitude) to replicate earlier findings^19,32. The entered variables were breakpoint, P_max, O_max (log10 transformed), intensity, and 1/elasticity (cube root transformed). The factor structure (i.e., two factors) was pre-specified based on findings from prior research¹⁹. The factor structure was verified by examination of the scree plot³³. A factor loading of.40 on the pattern matrix was used as the criterion for determining if an item significantly loaded on a given factor³⁴.Demand indices were permitted to load onto multiple factors, consistent with previous studies. Factor scores were derived by use of standardized regression coefficients. The scores had a mean of 0 and a variance equal to the squared multiple correlation between the estimated factor scores and the true factor values.

2.3.2. Data analysis plan.

Purchase task data were examined for orderliness according to standard methods³⁵. No purchase task data were excluded using this method. Delay discounting data were examined for 1) If the last amount chosen at the highest delay was larger than the amount chosen at the lowest delay (i.e., insensitivity to delay increase) and 2) if indifference points were flat across all delay values. No data met these criteria. One participant had missing delay discounting data. All other predictor variables were first tested using bivariate correlations. Then, as in previous papers testing the RP model^12,13, both the Amplitude and Persistence factors of demand, and Discounting log AUC were tested as independent statistical predictors of dependence and average cigarettes per day (CPD) using linear regression methods (SPPS version 24, IBM). Smoking duration, gender, and disposable income were included as covariates in all models. Significance test thresholds were set at 0.05. Finally, linear regression models including all predictors, covariates, and a term expressing interaction between Amplitude of demand and delay discounting predicting both CPD and dependence were tested.

3. RESULTS

3.1. Participant Characteristics.

Participants (n = 50) were 17.7 (SD=1.0) years old on average, 50% female, 47.5% White, and had an average baseline CO value of 11.2 ppm (SD=7.2). Participants smoked 8.2 (SD= 4.5) CPD on average and reported an average dependence score of 4.2 (SD=1.5; range = 2–8), indicating moderate cigarette dependence.

3.2. Factor Analysis.

The first three Eigenvalues were 3.36, 0.90, and 0.41. Examination of the scree plot indicated that a two-factor structure was the best solution. Table 1 provides the significant loadings of the five CPT demand indices on each of the rotated factors. A two-factor solution accounted for 85% of the variance and was determined to be the best fitting structure. The ‘Persistence’ factor, Factor 1 (67.2% variance) comprised elasticity, O_max, P_max, and breakpoint from the cigarette purchase task. This factor reflects insensitivity to increases in price of cigarettes. The ‘Amplitude’ factor, Factor 2 (17.9% variance) comprised intensity and elasticity, and is reflective of the total level of demand that would be demonstrated in unrestricted conditions. The two factors were significantly correlated (r = .37, p < .01).

Table 1.

Factor Loadings from Exploratory Factor Analysis of the Cigarette Purchase Task

Indices	Factor 1 Persistence	Factor 2 Amplitude
Intensity	−.04	.97
Omax	.70	.39
Pmax	1.0	−.22
Breakpoint	.89	.05
Elasticity (1/α)	.57	.47
Eigenvalue	3.4	.90
% Variance	67.2	17.9

3.3. Correlations.

Correlations between predictor variables and outcome variables can be found in Table 2. The association between Amplitude of demand and discounting log AUC (r = −.22) was large but not statistically significant (p = .12). The correlation between Persistence of demand and discounting log AUC was moderate (r = −0.14) and also not statistically significant. Amplitude was significantly positively correlated with both cigarette dependence and average CPD, and discounting AUC was significantly negatively correlated with dependence such that steeper discounting was associated with greater dependence. Log discounting AUC was not significantly correlated with CPD in this sample. Greater smoking duration was significantly positively correlated with cigarettes smoked per day. Dependence and cigarettes smoked per day were significantly positively correlated.

Table 2.

Correlations between predictor variables and outcome variables (N=50).

	1	2	3	4	5	6	7	8
1. Gender (1= Male, 2 =Female)	1	−.01	.13	−.07	−.16	.09	−.05	−.05
2. Smoking Duration (years)		1	.16	.34*	−.00	.26	.69	−.01
3. Cigarette Dependence (mFTQ score)			1	.46**	.03	.48**	.30**	−.52**
4. Average Cigarettes Per Day				1	.20	.42**	.24	−.27
5. Disposable Income					1	.04	.64	−.03
6. Amplitude of Demand+						1	.37**	−.22
7. Persistence of Demand							1	−.14
8. Discounting log AUC+								1

⁺Cube root transformed.

^*p<.04

^**p<.01

AUC= Area Under the Curve.

3.4. Statistical Predictors of Cigarette Dependence.

Multiple linear regression models (summarized in Table 3) analyzed the extent to which both factors of demand and discounting log AUC were associated with cigarette dependence. The overall model was significant (F = 5.423, p < .0001; R² = .437). Both greater Amplitude of demand (sr²=.11, p = .02) and steeper discounting log AUC (sr²=.23, p = .001) were significantly associated with greater dependence. Neither Persistence of demand nor any of the included covariates were significant predictors of dependence. Next, we tested the same linear regression model with an interaction term between Amplitude of demand and discounting log AUC to determine if a relationship between these two constructs was associated with dependence; however, this term was not significant (p = .14). Further results of this model are not reported.

Table3.

Multiple regression models predicting baseline nicotine dependence (top) and average cigarettes smoked per day at baseline (bottom).

Outcome: Cigarette Dependence
Predictor	B	β	SE	95% CI	P Value	sr2
Amplitude of Demand	.50	.31	.21	(.07, .93)	.02**	.11
Persistence of Demand	.19	.11	.20	(−.21, .59)	.35	.02
Discounting log AUC	−1.3	−.42	.37	(−2.1, −.59)	.00**	.23
Smoking Duration	.05	.07	.08	(−.12, .22)	.55	.00
Gender	.30	.09	.37	(−.46, 1.0)	.43	.01
Disposable Income	.00	.03	.00	(−.00, .00)	.78	.00
Outcome: Average Cigarettes Per Day (CPD)
Predictor	B	β	SE	95% CI	P Value	sr2
Amplitude of Demand	1.2	.27	.66	(−.08, 2.6)	.06*	.07
Persistence of Demand	.47	.10	.63	(−.79, 1.7)	.14	.01
Discounting log AUC	−1.7	−.19	1.1	(−4.1, .60)	.14	.05
Smoking Duration	.54	.26	.27	(.00, 1.0)	.04**	.08
Gender	−.61	−.06	1.1	(−3.0, 1.7)	.60	.00
Disposable Income	.00	.19	.00	−(.00, .01)	.14	.05

Note: AUC-Area Under the Curve.

Double asterisks** indicates statistical significance at p < .05, a single asterisk* denotes significance at p < .10.

3.5. Statistical Predictors of Average CPD.

Multiple linear regression models (summarized in Table 3) analyzed the extent to which both factors of demand and discounting log AUC were associated with average cigarettes smoked per day. The overall model was significant (F = 3.510, p .007; R² = .334). Greater Amplitude of demand was marginally significantly associated with greater CPD (sr²=.07, p = .06). Greater smoking duration also significantly associated with CPD (sr²=.08, p =.04). Neither Persistence of demand, discounting log AUC, nor the remaining covariates were significant statistical predictors. Next, we tested the same linear regression model with an interaction term between Amplitude of demand and discounting log AUC to determine if a relationship between these two constructs was associated with average cigarettes smoked per day; however, this term was not significant (p = .46). Further results of this model are not reported.

4. DISCUSSION

The results of this study demonstrate that behavioral economic indices of cigarette demand and delay discounting, each of which contribute to a Reinforcer Pathology conceptualization of addiction, are both associated with facets of adolescent cigarette smoking. However, while demand for cigarettes was associated with both cigarette dependence severity and cigarette consumption in this sample, discounting was statistically predictive only of dependence severity. Smoking duration was also significantly associated with cigarettes smoked per day; though it was not significantly correlated nor did it statistically predict heavier dependence in this sample. As the sample age was relatively restricted, smoking duration necessarily reflects earlier initiation. It is possible that those who reached the whole-cigarette smoking milestone at a younger age are on a heavier-smoking trajectory, as it is known that younger age of initiation is associated with heavier smoking³⁶.

Delay discounting appears to operate as a trans-substance process^37,38, as discounting taps into a general preference for immediate versus delayed rewards that are not necessarily specific to cigarettes. Our finding that discounting was related to dependence level comports with data suggesting that discounting is predictive of poorer treatment outcomes in adolescents and dependence severity in adults^39,40. In addition to discounting, demand Amplitude was also associated with dependence. This indicates that in this sample, in contrast to findings with adults using cannabis^12,13, discounting was not a unique statistical predictor of dependence; but instead shared predictive power with demand. This finding indicates that in these adolescents, both processes are operating concurrently to contribute to greater levels of dependence, with both increased preferences for immediate reward and reinforcement from smoking contributing to the overall construct of dependence on cigarettes. Finally, delay discounting did not statistically predict consumption, while greater demand amplitude was associated with a higher level of consumption. Given that experience using the drug is associated with drug reinforcement, it is intuitive that demand is more closely related to consumption level than discounting.

Overall, the current results demonstrate the utility of using the Reinforcer Pathology model to distinguish between different facets of cigarette use in adolescent smokers. However, limitations of the study exist. First, the AUC metric may not be the ideal measure of discounting, as it may be insensitive to extreme outliers and to random responding; and the longest delay values may influence AUC values disproportionately⁴¹. However, it has the benefit of being atheoretical, unlike conceptions of discounting that rely on theoretical models of the processes underlying self-control³¹. Furthermore, the current study is cross-sectional in nature and has a relatively small sample size, as the original study was not powered for this secondary analysis. Future research should replicate these findings in a larger sample, and further establish the relative contributions of these dual processes on smoking in youth. In order to do this, longitudinal studies of the etiology of behavioral economic processes, and the temporal patterning of discounting and demand processes on smoking milestones are needed.

As behavioral mechanisms of youth smoking are uncovered, future prevention and intervention efforts may isolate and target high-risk components of Reinforcer Pathology on which to intervene. Several recent studies have shown that demand measures can both predict⁴² and moderate⁴³ treatment outcomes. However, this work has not yet been extended to adolescent smokers. As adolescent smokers may be both more amenable to changes in their smoking trajectory given that behavior may be more malleable during this developmental stage in which other smoking influences are in flux, yet at the same time tend to respond poorly to current forms of treatment⁴⁴, novel ways to understand smoking maintenance and progression in this population is crucial. Behavioral economics as a framework suggests treatment modalities, such as those that identify and support alternative, nonsmoking sources of reinforcement. Furthermore, delay discounting as a robust trans-disease process may be another target for intervention in this still-developing population^38,45. Such cross-substance behavioral factors may show promise for intervening on tobacco use beyond cigarettes in at-risk youth.

HIGHLIGHTS.

• The Reinforcer Pathology model has not yet been applied to adolescent smoking

• Higher demand, but not discounting, was associated with greater cigarettes smoked

• Both demand and discounting were significantly associated with greater dependence

• This model holds promise for studying cigarette smoking in youth