Neuroeconomic Predictors of Smoking Cessation Outcomes: A Preliminary Study of Delay Discounting in Treatment-Seeking Adult Smokers

Abstract

Large proportions of smokers are unsuccessful in evidence-based smoking cessation treatment and identifying prognostic predictors may inform improvements in treatment. Steep discounting of delayed rewards (delay discounting) is a robust predictor of poor smoking cessation outcome, but the underlying neural predictors have not been investigated. Forty-one treatment-seeking adult smokers completed a functional magnetic resonance imaging (fMRI) delay discounting paradigm prior to initiating a 9-week smoking cessation treatment protocol. Behavioral performance significantly predicted treatment outcomes (verified 7-day abstinence, n = 18; relapse, n = 23). Participants in the relapse group exhibited smaller area under the curve (d = 1.10) and smaller AUC was correlated with fewer days to smoking relapse (r = .56, p<.001) Neural correlates of discounting included medial and dorsolateral prefrontal cortex, posterior cingulate, precuneus, and anterior insula, and interactions between choice type and relapse status were present for the dorsolateral prefrontal cortex, precuneus, and the striatum. This initial investigation implicates differential neural activity in regions associated with frontal executive and default mode activity, as well as motivational circuits. Larger samples are needed to improve the resolution in identifying the neural underpinnings linking steep delay discounting to smoking cessation.

1. Introduction

Tobacco use remains the leading cause of preventable death worldwide (Camenga and Klein, 2016; Perez-Warnisher et al., 2018). Many chronic cigarette smokers (>55%) endorse numerous attempts at smoking cessation; however, few (4-5%) are successful in maintaining long-term abstinence (García-Rodríguez et al., 2013; Patnode et al., 2021). Improving treatment success rates necessitates a better understanding of mechanisms underlying smoking relapse, at both the behavioral and neurobiological level. One promising mechanism is delayed reward discounting (DRD), a behavioral economic indicator of how much a person values or devalues a rewarding stimulus depending on how much it is delayed over time (Madden and Bickel, 2010). Specifically, DRD tasks are typically used to measure the rate at which an individual devalues rewards as a function delay duration. Evidence from numerous studies suggests DRD is a robust and reliable behavioral approach to understand maladaptive decision-making, with some authors arguing that steeper DRD is a transdiagnostic or trans-disease process across a range of health disorders (Amlung et al., 2019; Bickel et al., 2012; Lempert et al., 2019); however, see Bailey et al. (2021) for an opposing argument. Steeper rates of DRD have been reported in individuals who regularly use substances compared to non-substance using controls, with two meta-analyses supporting this conclusion (Amlung et al., 2017; MacKillop et al., 2011). DRD measured prior to treatment has also been shown to relate to smoking cessation treatment outcomes in a number of individual studies (e.g., MacKillop & Kahler, 2009; Sheffer et al., 2012) and collectively across studies in a recent systematic review (Syan et al., 2021).

Neuroeconomics research has offered insights into the neural correlates of behavioral economic decision-making, with most of these studies using functional magnetic resonance imaging (fMRI). This growing literature includes fMRI studies of DRD in healthy participants (e.g., Ballard & Knutson, 2009; Bickel et al., 2009; Kable & Glimcher, 2007; McClure et al., 2004), which have been synthesized in previous meta-analyses (Schüller et al., 2019; Wesley and Bickel, 2014). Studies have also included people who use various substances, including alcohol (Amlung et al., 2012; Boettiger et al., 2007; Claus et al., 2011), methamphetamine (Hoffman et al., 2008; Monterosso et al., 2007) and cocaine (Meade et al., 2011). These studies have reported significant blood oxygen-level dependent (BOLD) activation differences between substance use and control groups in brain regions implicated in reward, information processing, and decision-making, including the striatum, posterior parietal cortex, anterior insula, and dorsolateral prefrontal cortex (DLPFC).

With respect to nicotine dependence, three studies have examined brain activity associated with DRD decision making. MacKillop et al. (2012) conducted the first study to characterize BOLD responses during a DRD task for both monetary and cigarette rewards in a sample of 13 cigarette smokers. Increased BOLD activation was observed in many of the same regions as previous fMRI studies, including medial prefrontal cortex, DLPFC, cingulate cortex, and anterior insula. Kobiella et al. (2014) subsequently examined acute nicotine effects on BOLD activation during monetary DRD in 27 nicotine-dependent smokers and 33 non-smokers. Smokers showed less activation in precuneus and striatum compared to non-smokers. Furthermore, similar effects on brain activation were observed in both groups following a single dose of nicotine, suggesting activation differences between smokers and non-smokers are not exclusively due to acute pharmacological effects of nicotine. Finally, Clewett et al. (2014) compared functional connectivity underlying DRD choices in 39 smokers and 33 non-smokers. The smokers exhibited greater functional connectivity between a left fronto-parietal network and left anterior insular cortex, and the degree of functional connectivity between these regions was correlated with differences in discounting.

The neuroeconomics studies reviewed above offered initial evidence of the neural correlates of DRD in people who smoke cigarettes and have characterized differences between smokers and non-smokers. However, no fMRI studies to date have investigated if activation patterns during DRD decisions made prior to smoking cessation treatment are associated with treatment outcomes. Precedent for studying DRD-related brain responses as treatment predictors is provided by a study by Elton et al. (2019) who examined whether BOLD activation during a during a DRD task were associated with substance use disorder treatment outcomes in 33 adolescent alcohol and/or cannabis users. Greater pre-intervention engagement of reward, salience, and default mode networks predicted poorer outcomes. Using a similar rationale, the current study sought to identify neural correlates of DRD in adult smokers prior to treatment and determine if BOLD responses were linked to outcomes in a subsequent 9-week smoking cessation intervention. We predicted steeper DRD would be a negative prognostic factor of treatment outcomes, with smokers who relapsed during treatment exhibiting significantly steeper DRD compared to smokers who were successful in their quit attempt. We also predicted DRD choices would be associated with significant BOLD activation in a network of regions previously implicated in DRD, including DLPFC and medial PFC, posterior parietal cortex, anterior insula, and striatum. We expected BOLD activation during DRD decisions would relate to treatment outcomes; however, given this was the first neuroeconomics study to examine the neural correlates of smoking relapse, we did not make explicit directional hypotheses about the association between DRD-related brain activity and relapse.

2. Methods

2.1. Participants

Participants were treatment-seeking adult smokers who reported smoking at least 10 cigarettes per day and reported a motivation to quit of at least 5 on a scale of 1 to 10. Participants were required to be right-handed, 18-65 years old, with no major medical, neurological, or psychiatric disorders. Additional exclusion criteria included: receiving smoking cessation treatment in the past 90 days, consuming more than five alcoholic beverages per day; greater than weekly marijuana use; greater than monthly use of any other psychoactive drugs; MRI contraindications; and estimated IQ of less than 70 based on the Wechsler Test of Adult Reading (Holdnack, 2001; Whitney et al., 2010). Participants were paid a total of $126 for 5 experimental sessions but were not paid for participating in treatment, which was provided at no cost.

A total of 52 participants completed the MRI scan. Of these, 11 were excluded for excessive missing trials (n = 6; >25% missing), invalid responding (n = 2; >33% of invalid control trials), excessive movement (n = 2; >20% of TRs flagged in motion censoring), or loss of MRI data (n = 1). The final sample (n = 41) identified as White (68.3%), Black or African American (26.8%), mixed race (2.4%), or another race (2.4%); 2.4% identified as Hispanic. Mean age was 39.95 (SD = 12.05), median income was $30,000 (interquartile range = <$15,000-$44,999). The sample reported smoking, on average, 22.43 (SD = 11.77) cigarettes/day and participants were moderately nicotine dependent (Fagerström Test for Nicotine Dependence [FTND]; M = 4.93, SD = 2.52).

2.2. Procedures

All procedures were approved by the University of Georgia Institutional Review Board. Following an initial telephone screening, eligible participants were invited for an intake assessment to confirm eligibility and smoking characteristics. Subsequently, eligible participants were scheduled for a 90-minute MRI scan consisting of an anatomical scan, three imaging runs of the DRD paradigm, and other fMRI paradigms not included in this report. Following these assessments, participants completed a nine-week smoking cessation treatment protocol combining nicotine replacement therapy and weekly in-person counseling sessions. Specifically, participants receive eight-weeks of nicotine replacement therapy (4 weeks @ 21 mg, 2 weeks @ 14 mg, and 2 weeks @ 7 mg) in conjunction with behavioral counseling comprising motivational interviewing and relapse prevention strategies. Complete details of the treatment protocol are provided in Owens et al., (2018).

2.3. Measures

A demographic form was administered to assess participant’s age, gender, race, ethnicity, education level, and income. Impulsive personality traits were assessed using the short version of the UPPS-P Impulsiveness Scale (Cyders et al., 2014), yielding mean item ratings on a 4-point Likert scale for five subscales: negative urgency, lack of premeditation, lack of perseverance, sensation seeking, and positive urgency. One participant in the relapse group did not complete the UPPS-P scale. The Fagerström Test of Nicotine Dependence (FTND; Heatherton, Kozlowski, Frecker, & Fagerström, 1991). This six-item questionnaire is a widely used measure of nicotine dependence and has been shown to have good reliability and validity (Piper et al., 2006). Participants completed a retrospective timeline follow-back (TLFB) calendar at each weekly treatment session to indicate the number of cigarettes smoked each day since the previous session (Robinson et al., 2014). Exhaled carbon monoxide (in parts per million; COppm) was measured using a PiCO+ Smokerlyzer (Bedfont Scientific Ltd., Rochester, UK) to verify reported abstinence during each treatment session and research visit.

2.4. Functional neuroimaging task and protocol

2.4.1. fMRI delayed reward discounting paradigm.

Participants completed three runs of an event-related DRD task (Amlung et al., 2012) consisting of dichotomous choices between a larger delayed reward (LDR; $100 available after 1 day, 1 week, 1 month, 6 months, or 1 year) and a smaller immediate reward (SIR; $10, $20, $30, $40, $50, $60, $70, $80, $90, or $99 available that day). Control items involved choices between two different immediately available monetary amounts. In total, 120 active and 21 control items were assessed using the same pseudorandomized trial sequence across all participants (Table S1). All items were included at least twice and 20 items which demonstrated the highest rate of variability of responding in a previous study (Amlung et al., 2012) were administered three times.

As depicted in Figure 1, the two monetary alternatives were presented for up to 6000 ms (choice epoch) and changed to a post-response screen upon participant response. A jittered inter-stimulus interval (ISI) screen identical to the post-response screen was displayed in-between items (ISI mean = 3000 ms.; range 1000-7000 ms.). Left-right position of the immediate option was counterbalanced across trials. Three fMRI imaging runs lasted between 7.2 and 7.4 minutes (total paradigm duration = 22 minutes). Stimuli were programmed in E-Prime 2.0 (Psychology Software Tools, Inc.; Sharpsburg, PA) and were presented using MRI-compatible goggles. Participant responses were recorded with the right hand using a MRI-compatible response box.

Figure 1. Delay Discounting Task Schematic.

Each trial on the delay discounting task included a choice phase in which participants were presented with dichotomous choices between two rewards. For active trials, the choice was between a smaller-immediate reward available today and a larger-later reward available after a delay (see Table S1 for trial list). For control trials, the choice was between a smaller and a larger reward, both available today. After making a response with the MRI compatible button box, the numeric values were replaced with Xs for the post-response and interstimulus interval phases. RT = response time.

2.4.2. MRI scanning protocol.

Imaging data was collected on a General Electric 16-channel fixed-site Signa HDx 3.0 Tesla MRI scanner. Structural images were acquired for anatomical reference using a high-resolution T1 SPGR sequence (voxel size = 1mm³, a field of view = 256² mm, matrix = 256², slice thickness = 1mm) with sufficient contiguous axial slices for whole-brain coverage. Functional imaging was conducted using T2* echo planar imaging (EPI) with a single-shot gradient echo pulse sequence (TR = 2500 ms, TE = 25 ms, field of view = 225² mm, matrix = 64², voxel size = 3.5 mm³, with 40 contiguous 3.5 mm slices collected axially). Three dummy samples preceded each functional scan.

2.6. Data analyses

2.6.1. Behavioral analyses.

Behavioral data from all three runs were aggregated and indifference points for each of the five delays were calculated based on each participant’s array of responses using procedures used in previous research (Amlung and MacKillop, 2011). The metric of DRD used was area under the curve (AUC), which does not make assumptions about the form of the underlying data, eliminating the risk of model-fitting error (Myerson et al., 2001). Smaller AUC values reflect steeper discounting of delayed rewards.

For modeling fMRI BOLD responses, individual trials on the DRD task were coded based on choice difficulty, using a similar strategy employed in previous studies (Amlung et al., 2012; Monterosso et al., 2007). Specifically, choices were coded as easy or hard based on their proximity to the participants indifference points. For each delay interval the four choices with dollar values for the immediate option nearest to the indifference point were classified as hard choices (i.e., two choices with dollar values larger, two choices with dollar values smaller). All other choices were classified as easy choices.

Subjects were divided into relapse and non-relapse groups based on treatment success criterion of 7-day point prevalence abstinence at the end of treatment. In other words, participants reporting smoking 1 or more cigarette per day were categorized into the relapse group. Participants who dropped out of treatment (Benowitz et al., 2002) or had exhaled CO equal to or greater than 10 parts per million were considered to have relapsed as recommended by the Society for Research on Nicotine and Tobacco (Monterosso et al., 2007).

2.6.2. Imaging analyses.

Image preprocessing and first-level regression analyses were conducted in Analysis of Functional Neuroimages (AFNI) (Cox, 1996) software using a standard preprocessing pipeline. Raw DICOM images were assembled into AFNI datasets using the dcm2niix_afni program. Next, 3dToutcount was used to identify volumes with more than 10% outliers, and these were subsequently censored in the individual subject regression analyses. Preprocessing included correction for slice acquisition timing, volume registration to the middle volume of the run closest in time to the anatomical image acquisition, spatial smoothing using a 3.5mm full-width half-maximum (FWHM) Gaussian filter, masking of non-brain voxels, and scaling to percent signal change from mean signal intensity per run. Motion parameter files from the volume registration were censored using a temporal derivative approach and an inter-TR threshold of 0.3 (see https://afni.nimh.nih.gov/pub/dist/doc/program_help/1d_tool.py.html). The motion censor file was merged with the outlier censor file and included in the individual subject regression model. For each subject, general linear modeling was completed in 3dDeconvolve including the following regressors: three regressors relating to DRD choice type (easy, hard, control); one regressor for invalid control trials; one regressor for missing trials with no behavioral response; six nuisance regressors to account for motion (x, y, z, roll, pitch, yaw); and linear, quadratic and cubic trends. The resulting datasets were transformed into Talairach space (Talairach and Tournoux, 1988) (Talairach and Tournoux, 1988).

Group-level analyses used the beta weights generated from the general linear model analyses and included two phases. First, an a priori region of interest (ROI) analysis was conducted based on coordinates from our study that used a similar DRD task and analytic strategy (Amlung et al., 2012). Ten a priori ROIs were chosen, and spherical ROIs were drawn around the center of mass coordinates for each ROI (see Table S2 for coordinates and Figure S1 for anatomical locations of the ROIs). The size of individual ROI spheres was chosen to approximate the size (# of voxels) in Amlung et al., (2012). Second, a data-driven disjunction (Boolean “OR”) mask approach was implemented based on methods from previous neuroeconomics studies (Amlung et al., 2012; Ballard and Knutson, 2009; MacKillop et al., 2012). Specifically, voxels exhibiting significant BOLD response in any of the three trial types (easy, hard, control) were identified at a minimum voxel-level threshold of p<.0001 and cluster extent of 8 voxels. The p value was subsequently decreased to break apart large clusters into discrete regions (see Table S3 in Supplementary Materials). Analysis of variance (ANOVA) models were then conducted in SPSS 28 (IBM; Armonk, NY) using mean BOLD signal in each ROI as the dependent variable. In these models, relapse group (relapse vs. non-relapse) was included as a between-subjects factor, and choice type (easy, hard control) was included as a within-subjects factor; group x choice type interactions were also examined. Results are reported at an uncorrected p value (p < .05) along with false discovery rate (FDR) (Benjamini and Hochberg, 1995) of q < .05 to correct for multiple comparisons in these analyses.

3. Results

3.1. Behavioral results

Treatment outcome data indicated 90.2% of participants attended at least one treatment session (M = 5.0, SD = 3.5 session attended), and participants averaged 24.7 (SD = 23.3) days until relapsing to daily smoking. Therapist adherence to the treatment protocol was 96.2% and adherence to motivational interviewing spirit across all sessions was 98.0%. At the 9-week follow-up, 23 participants relapsed and 18 did not relapse. Groups were significantly different with respect to intake FTND total (p < .01) and cigarettes/day (p < .05), but did not significantly differ with respect to age, gender, race, income, or expired carbon monoxide at intake (ps > .05). Groups did not significantly differ for any of the UPPS-P subscales (ps > .41; see Table 1).

Table 1.

Participant Characteristics

	No-relapse	Relapse
	M (SD)/%	M (SD)/%
N	18	23
Gender	66.7% Male	69.6% Male
Age	36.72 (11.93)	42.63 (11.55)
Race	73.7% White; 21.1% Black; 5.3% Other	60.9% White; 34.8% Black; 4.3% Mixed Race
Income1	$30,000-$44,999	$15,000-$29,999
FTND (intake)	3.56 (2.15)	5.92 (2.28)**
Cigarettes/day (intake)	17.78 (8.21)	25.71 (12.64)*
COppm (intake)	19.89 (10.99)	23.50 (11.09)
Days to relapse	--	5.65 (11.24)
AUC	0.539 (0.284)	0.285 (0.177)
UPPS-P subscales
Negative urgency	2.05 (0.58)	2.15 (0.59)2
Lack of premeditation	2.05 (0.36)	1.98 (0.45)2
Lack of perseverance	2.01 (0.41)	1.89 (0.41)2
Sensation seeking	2.79 (0.69)	2.95 (0.58)2
Positive urgency	1.63 (0.57)	1.74 (0.62)2

Note:

¹Median values reported for income;

²n = 22;

AUC = area under discounting curve; FTND = Fagerström Test for Nicotine Dependence; COppm = Expired breath carbon monoxide in parts per million.

^*p < .05

^**p < .01.

Delay discounting curves are presented in Figure 2. Compared to the no-relapse group, the relapse group had a significantly smaller AUC [F(1,39) = 12.58, p = .001, η²_p = .24], reflecting steeper discounting. This difference in delay discounting remained significant after statistically controlling for FTND at intake [F(1,38) = 4.18, p = .048, η²_p = .10] and cigarettes per day at intake at intake [F(1,38) = 7.47, p = .009, η²_p = .16]. Smaller AUC was significantly correlated with fewer days to relapse [r = .56, p < .001], higher intake FTND [r = −.52, p < .001], and higher intake cigarettes per day [r = −.40, p < .05]. There was no significant main effect of Gender [F(1,37) = 0.65, p = .57] or Gender × Relapse Group interaction [F(1,37) = 0.18, p = .67] for AUC.

Figure 2. Delay discounting by smoking relapse status.

Temporal discounting curves corresponding to the overall area under the curve (AUC) values for the relapse (dashed line) and non-relapse (solid line) groups. Insert depicts corresponding data at 1- and 7-day delay lengths. Individual data points reflect mean indifference point at each delay (+/− standard error of the mean). **p < .01.

3.2. Imaging results

Summary maps depicting significant BOLD activity during easy and hard DRD trials (relative to control trials) in the total sample are shown in Figure 3. Complete montages for each choice type are provided in Figure S2 in supplementary materials. Compared to control choices, hard choices were associated with greater BOLD activation in bilateral middle frontal gyrus / dorsolateral prefrontal cortex, medial frontal gyrus / supplementary motor area, bilateral posterior parietal cortex / precuneus, bilateral anterior insula, bilateral dorsal striatum (caudate/putamen), and posterior cingulate. We also observed less BOLD activity during hard vs control choices in bilateral intraparietal lobule. Easy choices were associated with greater activation than control choices in several of these areas, but to a lesser extent, including medial frontal gyrus, posterior cingulate, and precuneus. There were no clusters of greater activation in control compared to easy choices.

Figure 3. BOLD Activation during Easy and Hard Discounting Choices.

Axial slices depicting clusters of significant BOLD activation during easy and hard choices. In both cases, the activation depicted reflects a voxel-wise contrast subtracting activation during control choices (i.e., warm colors = greater activation during easy or hard choices relative to control choices; blue colors = greater activation during control choices relative to easy or hard choices). Images thresholded at voxel-wise p < .001, minimum cluster extent of 15 contiguous voxels. Images shown in radiological convention (left = right). Abbreviations: Ant. Insula = anterior insula; Caud. Put = caudate putamen; Post cingulate = posterior cingulate; Medial FG = medial frontal gyrus; MidFG = middle frontal gyrus; DLPFC = dorsolateral prefrontal cortex.

3.2.1. A priori ROI analysis.

Mean BOLD signal was extracted from each of the 10 a priori ROIs for each participant and examined using ANOVA models. One sample t-tests confirmed that all a priori ROIs exhibited significant activation responses at the group level (ts > 3.20; ps < .002). Complete results are provided in Table 2. Significant main effects of Choice Type after FDR correction were found for 8 of 10 ROIs. As shown in Figure 4A, several of these regions exhibited greater activation during hard choices compared to the other two choice types (Patterns A and B in Figure 4A), including bilateral DLPFC and bilateral anterior insula. Other regions were engaged in both types of intertemporal choices (easy and hard) and to a lesser extent in control choices (Pattern C), including bilateral posterior cingulate cortex, medial frontal gyrus, and left precuneus. Postcentral gyrus and thalamus did not significantly differ by choice type. There were no significant main effects of Group (ps > .10) and no significant Group × Choice Type interactions (ps > .063) in the a priori ROIs.

Table 2.

Results of a priori region of interest analysis

	M.E. Choice Type			M.E. Group			Group × Choice Type
Region	F(1,39)	p	ηp2	F(1,39)	p	ηp2	F(1,39)	p	ηp2
1. Medial frontal gyrus	7.89	.001*	0.168	1.08	.306	0.027	2.08	.132	0.051
2. Posterior cingulate (R)	4.97	.009*	0.113	1.79	.189	0.044	1.41	.251	0.035
3. Posterior cingulate (L)	6.85	.002*	0.149	0.19	.666	0.005	2.41	.096	0.058
4. Postcentral gyrus (L)	0.74	.481	0.019	0.06	.814	0.001	2.32	.105	0.056
5. Thalamus	1.54	.222	0.038	0.51	.479	0.013	1.51	.227	0.037
6. Anterior insula (L)	14.48	<.001*	0.271	0.21	.648	0.005	2.19	.118	0.053
7. Anterior insula (R)	5.84	.004*	0.130	0.18	.678	0.004	2.86	.063	0.068
8. Precuneus (L)	5.45	.006*	0.123	0.31	.582	0.008	2.09	.131	0.051
9. DLPFC (R)	7.36	.001*	0.159	0.83	.367	0.021	2.11	.128	0.051
10. DLPFC (L)	13.71	<.001*	0.260	2.81	.102	0.067	1.08	.344	0.027

Note:

^*Effect significant following false discovery rate (FDR) correction;

L = left; R = right; DLPFC = Dorsolateral prefrontal cortex; TLRC = Talairach coordinates

Figure 4. Results of a priori and Empirically Defined ROI Analysis.

Bar plots depicting BOLD percent signal change by choice type for the a priori ROIs (Panel A) and empirically defined ROIs (Panel B) and In Panels A and B, bars reflect mean (+/− standard error) for easy (dark gray), hard (light gray), and control (black) choices. Notation for significant pairwise effects: A = all three choice types different; B = Hard different from easy and control; C = Easy and hard different from control; D = Hard different from control; E = Hard different from easy; NS = non-significant main effect.

3.2.2. Empirical ROI analysis.

A total of 20 empirical ROIs were identified in the disjunction mask. The anatomical location of these ROIs is depicted in Figure S3, and the coordinates and cluster sizes are provided in Table S3. Similar to the a priori analyses above, we examined differential BOLD activity as a function of choice type and relapse group using a series of mixed ANOVAs with FDR correction for multiple comparisons. Complete results of these analyses are presented in Table 3. Significant main effects of Choice Type were found for 15/20 ROIs and all effects remained significant following FDR correction. Pairwise contrasts between the three choice types revealed five patterns, as indicated in Figure 4B. Three of the regions exhibited significant differences between all three choice types with the same pattern (hard > easy > control), including medial frontal gyrus, left DLPFC, and posterior cingulate (Pattern A). Six regions had greater activation during hard compared to easy and control choices, with the latter two not significantly differing (Pattern B; right middle frontal gyrus/DLPFC, bilateral insula, right precuneus, and left caudate). Two regions were significantly more active during the two intertemporal choice types (easy and hard) compared to control (Pattern C; thalamus and left precuneus). Three regions differed between hard and control trials only (Pattern D; middle frontal gyrus/orbitofrontal gyrus, right precuneus, and right caudate). Finally, two regions differed only between easy and hard choices (Pattern E), including a cluster spanning the precentral gyrus/middle insula and left posterior insula/superior temporal gyrus (the only region exhibiting significant task-related deactivation in the empirically defined ROIs).

Table 3.

Results of Empirically Defined ROI Analysis

	M.E. Choice Type			M.E. Group			Group × Choice Type
Region	F(2,78)	p	ηp2	F(1,39)	p	ηp2	F(2,78)	p	ηp2
Medial frontal gyrus / SMA	20.33	<.001*	0.343	0.62	.436	0.016	2.16	.122	0.052
Thalamus (L)	3.58	.032*	0.084	0.93	.340	0.023	1.50	.231	0.037
Thalamus (R)	2.16	.122	0.052	1.62	.211	0.040	2.05	.136	0.050
Precuneus-1 (R)	4.71	.012*	0.108	1.15	.289	0.029	1.48	.233	0.037
MFG / DLPFC (L)	13.70	<.001*	0.260	0.79	.380	0.020	1.44	.244	0.036
Caudate / Putamen (R)	3.57	.033*	0.084	0.67	.419	0.017	3.48	.036	0.082
Posterior cingulate	11.85	<.001*	0.233	0.84	.366	0.021	0.73	.487	0.018
Putamen (L)	1.86	.162	0.046	0.06	.815	0.001	3.67	.030	0.086
Inferior parietal lobule (L)	0.52	.597	0.013	0.00	.969	0.000	2.19	.119	0.053
Precuneus (L)	7.01	.002*	0.152	0.76	.388	0.019	2.06	.134	0.050
Anterior insula	25.75	<.001*	0.398	0.30	.589	0.008	1.33	.270	0.033
Anterior insula (R)	19.31	<.001*	0.331	0.19	.662	0.005	1.82	.168	0.045
Posterior insula / STG (L)	4.67	.012*	0.107	0.16	.691	0.004	0.18	.837	0.005
Precuneus-2 (R)	5.56	.006*	0.125	0.20	.659	0.005	4.17	.019	0.097
Parahippocampal gyrus / MTG (R)	0.99	.378	0.025	0.74	.395	0.019	0.75	.475	0.019
Caudate (L)	8.67	<.001*	0.182	1.01	.322	0.025	2.80	.067	0.067
Middle Insula / precentral gyrus (L)	4.27	.017*	0.099	0.04	.851	0.001	2.20	.118	0.053
MFG / DLPFC (R)	23.40	<.001*	0.375	0.02	.893	0.000	3.27	.043	0.077
MFG / BA6 (R)	11.96	<.001*	0.235	2.36	.133	0.057	1.63	.203	0.040
MFG / OFC (R)	3.72	.029*	0.087	1.74	.195	0.043	0.98	.381	0.024

Note:

^*Effect significant following false discovery rate (FDR) correction;

L = left; R = right; BA = Brodmann area; DLPFC = Dorsolateral prefrontal cortex; MFG = Middle frontal gyrus; MTG = Middle temporal gyrus; OFC = Orbitofrontal cortex; STG = Superior temporal gyrus; SMA = Supplementary motor area

We also found Group × Choice Type interactions in four regions at a nominal significance level of p < .05; however, these effects were not significant after applying FDR correction. These regions included right DLPFC, right precuneus, right caudate, and left putamen. As shown in Figure 5, it appears these interactions are largely attributed to reduced activation during control trials in the relapse group compared to no-relapse group. An exception is right precuneus, where the non-relapse group exhibited significantly greater activation during hard choices compared to the other choice types but did not significantly differ between the three choice types in the relapse group. Finally, there were no statistically significant main effects of group (ps >.133) among the empirical ROIs.

Figure 5. Relapse Group by Choice Type Interactions.

Regions exhibiting significnat nteractions between group and choice type in empirically defined ROIs. Bars reflect mean (+/− standard error) for easy, hard, and control choices; checkered bars reflect the no relapse group. Anatomical locations of ROIs depicted on axial slices using a standard Talairach brain, presented in radiological convention (left = right).

4. Discussion

This study was the first to use a neuroeconomic approach to investigate differential neural processing during DRD decision making and smoking cessation treatment outcomes in adult cigarette smokers. Steep DRD is robustly associated with worse treatment outcomes across behavioral studies (Syan et al., 2021), but no prior study had examined whether brain activation supporting those decisions similarly predicts treatment outcomes. Another goal of the study was to extend findings of a growing number of neuroeconomics studies examining the neural correlates of DRD in smokers. Because this is the first study to examine neural correlates of DRD as a predictor of treatment outcomes and the sample size is relatively modest, particularly for comparisons based on relapse status, these findings should be considered preliminary and interpreted with caution.

The behavioral results of this study supported our hypotheses: participants who were ultimately unsuccessful in stopping smoking at 8-weeks showed substantially steeper discounting at baseline compared to those participants who maintained smoking abstinence. This is consistent with results of the systematic review by Syan et al. (2021). Interestingly, relapse and non-relapse groups did not significantly differ on five domains of impulsive personality traits as assessed by the UPPS-P questionnaire. This suggests that the participants who were unable to maintain abstinence were not characterized as impulsive in general sense but instead showed a specific difference in DRD and immediate reward preference. This is also consistent with the multidimensional nature of impulsivity, with DRD and impulsive personality traits representing generally distinct domains (MacKillop et al., 2016). Taken together, the behavioral findings confirm that the clinical utility of steep DRD is not limited to differentiating people who use substances from controls, but increased preference for smaller-immediate rewards also meaningfully differentiates subgroups of smokers based on relapse status.

The unique contributions of our neuroimaging results generally fall into two categories: 1) further characterizing the neural correlates of DRD choices in cigarette smokers, and 2) comparing neural activation based on smoking relapse status. In the first case, the fMRI results revealed significant differences by choice type (i.e., greater BOLD activation during hard relative to easy choices) in several brain regions in prefrontal cortex, posterior parietal cortex, anterior insula, and striatum, among others. This pattern is largely consistent with previous neuroeconomics studies in healthy participants (Carter et al., 2010; Schüller et al., 2019; Wesley and Bickel, 2014) and the limited number of studies in substance use samples (Owens et al., 2019). To date, only a few fMRI studies have investigated DRD in smokers (Clewett et al., 2014; Kobiella et al., 2014; MacKillop et al., 2012). In a within-subjects study of smokers, MacKillop et al. (2012) found choices for delayed rewards compared to immediate rewards activated several of the same brain areas as the current study, including medial prefrontal cortex, anterior insula, middle frontal gyrus, and cingulate gyrus. Clewett et al. (2014) subsequently demonstrated greater functional coupling between anterior insula and the left fronto-parietal network in smokers compared to non-smokers, again pointing to a key role for the anterior insula in DRD among smokers (see Naqvi & Bechara, 2010 for a review of the role of the insula in addiction). Together with the current findings, these studies confirm that more difficult (or potentially conflicting) DRD choices based on the similarity in subjective value of the choice options are associated with greater recruitment of fronto-striatal and fronto-parietal brain regions involved in executive function and decision-making (DLPFC); conflict processing and intention (medial PFC/anterior cingulate), and prospective thought (posterior parietal cortex).

Beyond characterizing the neural correlates of DRD, our study is the first to investigate differences in DRD-related neural processing based on smoking cessation outcomes, revealing nuanced findings. To start, in both a priori and empirical ROIs, there were no main effects of relapse status, suggesting overall activation levels across choice types did not differ as a function of treatment outcome. In interaction analyses, selective differences were only present in empirical ROIs. Specifically, the interactions implicated DLPFC, precuneus, caudate, and putamen. The effects were of modest effect size and would not survive stringent type 1 error correction. The DLPFC has been identified as the key cognitive control region activated during DRD decisions in a meta-analysis by Wesley and Bickel (2014), and this region was identified in a systematic review by Owens et al. (2019) as the region for which activation levels during DRD most consistently differentiated individuals with substance use disorders from controls. Likewise, the striatum (i.e., caudate and putamen) and precuneus have been identified as key reward and prospection regions involved in DRD decision making (Carter et al., 2010), and networks involving these regions consistently differentiate individuals with substance use disorders from controls (Owens et al., 2019). Decreased functional connectivity between the striatum and DLPFC predicted smoking lapse during a 12-hour period in male smokers (Yuan et al., 2018), and negative functional coupling between DLPFC and striatum is positively associated with craving scores following cigarette cue exposure (Yuan et al., 2017). Together with the current results, these findings suggest DLPFC and striatum are centrally involved in smoking relapse and related neurocognitive processes such as craving and DRD.

Another complexity in these findings was that several interactions were driven by reduced activation during control trials (larger vs. smaller monetary amounts) in the relapse group compared to no-relapse group. Beyond discounting, this could reflect higher anhedonia or diminished attention, both of which have been implicated in smoking cessation (Leventhal et al., 2009; McClernon and Kollins, 2008). Finally, the results raise a methodological question, insofar as the analytic methods equate participants’ levels of choice impulsivity (i.e., comparing BOLD activity within a specific choice type), when the significant variable is scaled in absolute differences across individuals. In other words, innovations in methods may also been needed to understand these relationships. Collectively, the results implicate DRD-related neural activity in several regions with relapse, but simultaneously suggest a larger sample will be necessary to fully understand these brain-behavior relationships.

The findings have several other potential implications. Including assessments of DRD at the initiation of smoking cessation treatment may be a useful clinical marker of whether an individual is likely to be successful in treatment (Syan et al., 2021). While the long-form DRD assessment in this study was quite thorough (i.e., 120 DRD trials over 22 minutes), there are brief low-burden assessments that have been implemented in previous studies (Gray et al., 2014; Koffarnus and Bickel, 2014). The extent to which these abbreviated DRD tasks predict smoking cessation outcomes needs be evaluated in future research. For those smokers who show steeper discounting, supplemental treatment interventions could be implemented to reduce DRD, such as episodic future thinking (EFT) (Aonso-Diego et al., 2021; Patel and Amlung, 2020). Indeed, previous studies have demonstrated that experimentally reducing DRD via episodic future thinking also reduces cigarette smoking in laboratory self-administration studies (Stein et al., 2016) and self-reported cigarettes smoked during the week following EFT. Whether EFT exhibits clinical efficacy as an adjunct to best-practice treatments is an open question, but a substantial proportion of individuals are not successful in established treatments (Patnode et al., 2021).

These preliminary findings provide a foundation for several future research directions. An important priority is to replicate the current results in a larger sample of smokers which will provide greater statistical power to detect significant differences between relapse and non-relapse groups. A larger sample would also permit application of machine learning (Barenholtz et al., 2020; Bzdok and Meyer-Lindenberg, 2018; Mak et al., 2019) and other data-driven analytic techniques to improve prediction of smoking from fMRI data (e.g., Zhang et al., 2019). To this end, Coughlin et al. (2020) used machine learning to examine predictors of cessation outcomes following cognitive behavioral therapy in adult smokers. In their analyses, DRD emerged as the most important predictor of treatment outcomes in a model with 80–81% prediction accuracy. A final priority is to track smoking cessation outcomes over a longer time interval (e.g., 6 to 12 months post treatment) to examine neural correlates of sustained abstinence.

The current findings should be interpreted in the context of the study’s strengths and limitations. Strengths of the study include an established DRD paradigm with a full range of intertemporal choices across reward magnitudes and delay lengths. The analytic approach included a priori ROIs based on prior neuroeconomics studies and data-driven ROIs to provide a comprehensive analysis of brain responses. Another strength is the smoking cessation treatment protocol combining free nicotine replacement therapy and motivational interviewing with high clinician adherence. The most salient limitation of this study is the relatively modest sample size. Although the total sample size of 41 was reasonable for the whole-group analyses by choice type—as indicated by our ability to disentangle activation underlying easy, hard, and control choices—the small group sizes for comparisons by relapse status likely constrained power to detect statistically significant differences. We also examined gender effects for DRD behavioral performance, finding no significant main effects or interactions with relapse status; however, our sample size was not sufficiently powered for testing gender effects in the fMRI analyses. A larger sample in a future study would permit a thorough analysis of potential gender effects in fMRI responses. The fMRI analyses also focused task-related activation in a priori and empirically defined ROIs, but we did not examine functional connectivity or interactions between regions. Future studies with smokers who maintain abstinence or lapse following treatment would benefit from inclusion of analyses of structural or functional connectivity (e.g., (Yuan et al., 2018, 2017). Finally, the DRD assessment evaluated choices for hypothetical monetary rewards. Although several studies indicate close correspondence between discounting rates and DRD-related neural correlates between hypothetical and actual monetary rewards (Bickel et al., 2009; Johnson and Bickel, 2002; Lagorio and Madden, 2005; Madden et al., 2004), this correspondence has not been examined in smokers.

In sum, the current study advances the link between steep DRD and smoking cessation outcome, replicating the predictive behavioral relationship and examining, for the first time, the neural correlates. In doing so, it implicates several brain regions involved in cognitive control, reward, salience, and prospection, albeit revealing small effect size differences. Nonetheless, the results provide preliminary support for the viability of applying a neuroeconomic framework for understanding the determinants of success and failure in smoking cessation treatment.