Simulated opioid choice linked to opioid use disorder severity among veterans with chronic pain: Initial validation of a novel paradigm

Abstract

Background:

Modeling addictive behavior among individuals with, or at risk for, opioid use disorder (OUD) in a way that is accurate, ethical, and reproducible presents a pressing concern. OUD risk is elevated among people with chronic pain on long-term opioid therapy (LTOT).

Objectives:

To provide initial validation of a novel opioid preference task as an index of OUD and its symptomatology among veterans prescribed opioids for chronic pain, a population at high risk for poor opioid-related outcomes. The relative ease by which such a paradigm can be implemented and performed in clinical contexts, including enabling investigation of opioid reinforcement and drug-seeking behavior while avoiding ethical pitfalls associated with direct drug administration, could make this task an attractive approach for potentially tracking OUD symptoms.

Methods:

We studied 87 veterans (74 male, 13 female) on LTOT for chronic pain – 33 of whom had OUD diagnoses. Participants completed a picture-viewing choice task to assess preference for viewing opioid-related images in comparison with standardized pleasant, unpleasant, neutral, and blank images. Opioid-related choice, measured by vigor of button pressing, was tested for association with OUD severity (measured by symptom counts), as well as craving and anhedonia.

Results:

Choice for opioid-related images was positively correlated with OUD severity (i.e., number of DSM-5 measured OUD symptoms) (r=0.38, p<0.001), particularly among those meeting diagnostic criteria for OUD (r=0.47, p=0.006). Neither craving nor anhedonia correlated with opioid-related choice.

Conclusions:

Our results provide initial validation for a new opioid picture-choice paradigm in patients with chronic pain.

INTRODUCTION

The opioid crisis remains a formidable challenge today and has in fact escalated since the onset of the COVID-19 pandemic (1). The prevalence of heroin use alone has doubled since 2002 with approximately 15,000 related deaths reported in 2018 (2). While illicit use of opioids comprises one arm of the crisis, misuse of prescription opioids among individuals with chronic pain is also an urgent concern. Indeed, misuse of prescription opioids has increased more than threefold over the past 20 years (3), and opioids are now among the most commonly misused drugs in the U.S. (4), particularly among individuals suffering from chronic pain. Approximately one quarter of people prescribed long-term opioid therapy (LTOT) for chronic pain may exhibit aberrant drug-related behaviors that mark the transition from opioid use to misuse (5), including unauthorized dose escalation or self-medication of negative affect with opioids (6). Excess use of prescription opioids in vulnerable populations can culminate in opioid use disorder (OUD), characterized by loss of control over opioid use and severe functional impairment due to opioid use. Approximately 10% of individuals taking opioids for chronic pain have been found to develop OUD (3).

According to several prominent theories of OUD specifically – and substance use disorder (SUD) more generally – vulnerable individuals compulsively pursue drugs and drug-related cues at the expense of natural reinforcers (7–11). Individuals with OUD may rely more heavily on addictive drugs to self-medicate physical and emotional pain, fostering a persistent state of hedonic and allostatic dysregulation, which manifests as increased sensitivity to pain (i.e., hyperalgesia) and stress in combination with reduced pleasure from primary reinforcers (12). At the neural level, abnormalities in brain reward/antireward and hormonal systems reinforce drug use to compensate for hedonic deficits (13). For this reason, compulsive use of opioids is associated with a reduction in the motivational significance of natural, non-drug rewards relative to drug rewards (14–19). As specified in the DSM-5, six of the 11 criteria for OUD involve preference for behaviors that promote opioid procurement and consumption over alternative reinforcers, including food and fulfilling social interaction (20).

The enhanced motivation and engagement in drug-related behaviors at the expense of hedonic rewards is well-captured by the drug-choice procedure, in which individuals choose between drug and non-drug reinforcers (e.g., food or money). This procedure provides an elegant and robust model for probing the severity of addiction, and has been deployed in both preclinical studies and studies of actively-using humans (21–23). Yet, implementing drug-choice models among treatment-seeking or abstaining humans presents key ethical challenges. The identification and validation of cost-effective, easy-to-implement behavioral models that characterize this shift toward opioid reinforcement and away from hedonic rewards (but that remain safe for use in vulnerable clinical populations) is a crucial scientific, therapeutic, and epidemiological goal.

To that end, we developed a ‘simulated’ drug-choice paradigm: rather than making choices for active drug self-administration, participants make choices to view drug-related images, in comparison with hedonic, pleasant images (24). In our prior work, simulated drug-choice procedures have correlated with and predicted core drug-relevant outcomes among misusers of cocaine (24–26), methamphetamine (27), and more recently, prescription opioids. With regard to the latter, simulated drug-choice behavior differentiated misusers from those who took their opioids as prescribed, especially when patients further lacked self-awareness into the extent of their drug-related choices (28).

The current study provided a conceptual replication of our prior results which delineated relationships between opioid misuse and opioid pill choice (28), but with several key differences to advance theory and research in this field. First, we studied a new cohort of 87 chronic pain patients, this time comprising a sample of U.S. military veterans. Second, unlike the prior study which used a probabilistic drug-choice task with uncertain choice-outcome contingencies, here we used a novel task variant with more explicit task contingencies that models vigor of responding as an index of motivation. Here, we aimed to provide initial validation of this novel explicit choice paradigm as a means of detecting OUD symptoms among patients with chronic pain. Third, we focused here on OUD severity as main predictor variable of interest, rather than OUD diagnosis per se. The aim was to provide a first step toward testing whether the drug-choice paradigm is able to monitor and track OUD severity, rather than linkage to a simple yes/no diagnosis. Fourth, we tested the contributions of three theoretically meaningful constructs that were expected to correlate with the extent of such opioid-related choice: craving, anhedonia, and OUD symptom counts. Craving is a core, widely-investigated feature of addiction (29, 30), precipitating opioid use among individuals with OUD (31). In prior work, chronic pain patients with higher levels of craving were found to be increasingly preoccupied with their next opioid dose and were more likely to inappropriately self-medicate with opioids (32, 33). Anhedonia is common among individuals with SUD (34), thought to result from the modulatory effects of recurrent intoxication on the brain’s reward set point (35). Compared with healthy controls, individuals with OUD demonstrate reduced responsiveness to pleasant images and increased responsiveness for drug images (36). Prescription opioid misuse has been shown to exacerbate anhedonia among individuals with chronic pain (8), which may subsequently perpetuate opioid misuse (37).

The following hypotheses guided our research: First, we hypothesized that, among a sample of chronic pain patients prescribed opioids, a higher severity of OUD, indexed as the number of DSM-5 OUD symptoms, would be associated with (1A) higher choice to view opioid pill images and (1B) lower choice to view pleasant images. Second, we hypothesized that opioid pill choice would be further modulated by (2A) craving and (2B) anhedonia, such that individual differences on these variables would track with the extent of opioid-related choice, above and beyond the influence of OUD symptom counts.

METHODS

Participants

We studied 87 participants recruited from the Veterans Affairs Hospital in Utah, all of whom were receiving LTOT for chronic pain (see Table 1 for demographics and pain conditions). Participants provided written informed consent in accordance with the local Institutional Review Board. The most common ailment was chronic low back pain, though many participants reported multiple sources of pain (Table 1). Psychiatric screening determined that 33 participants met DSM-5 criteria for OUD (symptom count: 4.12 ± 2.10), whereas 54 participants were prescribed opioids but did not meet criteria for OUD and therefore served as active controls for this study. A small number of individuals in the sample (7/87) were receiving medication-assisted therapy (MAT) at the time of their involvement in the study, including methadone, buprenorphine, naloxone, or naltrexone. The effects of MAT are accounted for in our statistical analyses, below.

Table 1.

Demographics and clinical features of the study participants.

Measure	OUD (+) Dx (N = 33)	OUD (−) Dx (N = 54)	Statistical test
Gender and Age
Female, N	3	10	χ2 = 1.50, p = .472
Age (M ± SD)	56.7±11.4	60.4±9.4	t = 1.61, p = .111
Race/Ethnicity			χ2 = 3.85, p = .427
Caucasian, N	27	44
Latin American, N	2	0
African American, N	2	5
Native American, N	1	3
Other, N	1	2
Pain type			χ2 = 9.21, p = .101
Back, N	23	32
Upper body, N	0	3
Lower body, N	4	11
Arthritis, N	0	4
Neuropathy, N	2	3
Other, N	4	1
OUD Severity			--
No OUD (0-1 symptoms), N	0	54
Mild (2-3 symptoms), N	16	0
Moderate (4-5 symptoms), N	8	0
Severe (≥6 symptoms), N	9	0
Clinical Features (M± SD)
COMM (Misuse) (6)	21.06±8.57	11.06±6.34	t = −6.24, p < .001
Duration of opioid use (mo)	104.88±83.89	126.62±103.15	t = 1.02, p = .313
Opioid dose (Avg Morphine Equiv Per day)	93.15±177.98	136.60±318.12	t = 0.72, p = .475
BPI (Avg pain, 5.1) (67)	5.64±1.52	5.41±1.54	t = −0.68, p = .500
Craving (week)	37.73±29.59	11.56±14.67	t = −5.46, p < .001

Note. OUD = Opioid Use Disorder, COMM = Current Opioid Misuse Measure, BPI = Brief Pain Inventory

Explicit Choice Task

Participants completed a choice paradigm adapted from our prior work (24–27), used for the first time in this study. Three categories of images – pleasant (e.g., smiling babies), unpleasant (e.g., spiders), and neutral (e.g., kitchen utensils) – were selected from the International Affective Image System (IAPS); images of opioid pills were selected from freely available online repositories and were matched to the IAPS images on size and the presence of human faces. A fifth category, blank (black) screens was also included in the task, which provided a comparison of the respective pictorial stimuli with non-stimuli (i.e., choosing nothing). To guard against the potential elicitation of severe drug craving, the task was administered in a clinical setting by trained staff with oversight from the Principal Investigator. By the end of the experimental protocol, if craving were to remain at problematic levels that would have had actionable consequences, research staff were trained to administer progressive muscle relaxation as a means of de-escalating craving. If craving remained high after this procedure, a staff psychologist or clinical social worker was available to treat the patient with CBT or motivational interviewing, and triage for further care as needed. In this sample, craving levels were manageable and transient, and problematic craving levels were not reached.

On each trial, two images from these respective categories were presented side-by-side, and participants used continuous button-pressing for choosing one image over the other (Figure 1). A button press for one image enlarged this image to fit the entire screen, removing the non-chosen image from the display. Continued button pressing was required to maintain the image on the screen for the trial duration; otherwise, without continuous button-pressing, the side-by-side image display would reappear. Participants could toggle between images as desired. However, whatever the choice pattern, each trial with two given images always lasted 5000 ms; after the time had elapsed a new pair of side-by-side images would appear to start the next trial. Images were pseudorandomly paired, such that (1) a given trial never contained two images from the same category, (2) each image category was presented on a total of 28 trials, and (3) each image category appeared on the left and right side of the screen 14 times.

Figure 1.

During the explicit [image] choice task, two images were presented side by side from different categories (i.e., pleasant, unpleasant, neutral, blank, pill; 28 total images presented per category; pleasant and pill are pictured above). Selecting an image via button pressing enlarged that chosen image to fill the screen (bottom) and removed the unchosen image. Continuous button pressing kept the chosen image on the screen for the trial duration. Toggling between images was allowed. Each trial, regardless of the choice pattern, lasted for five seconds. The entire task contained 70 trials, each with two pictures from which to choose.

Data processing for the task occurred as follows (38): during each trial, we calculated which of the two pictures had the higher number of button presses, and then recorded that picture category as the “choice” for that trial. For example, if during a trial there were 10 button presses for a pill image and 8 button presses for a pleasant image, then the score for the trial was scored as ‘1’ for pill and ‘0’ for pleasant. We then summed the total number of these trial choices separately for each picture category (pleasant, unpleasant, neutral, pill, and blank) across the 70 trials that comprised the task (note that for trials where equal number of presses occurred for both available picture types, the trial was scored as 0.5 for both picture categories). We note that the task was designed such that all study participants always performed the same number of trials (N=70). Given that each participant’s total score is the same as all other participants, the group effect in analysis is necessarily zero (and thus not reported below). An advantage of this approach to data processing is that our results cannot be explained by individual/group differences in the vigor of responding or other related confounds. Instead, significant results reflect how individuals allocate their behavior for those 70 trials.

OUD Severity

Per DSM-5, OUD severity was determined by corresponding number of OUD symptoms, an important index of impairment that can be used to guide treatment decisions (39). Mild OUD severity is indicated by the presence of 2-3 symptoms, moderate OUD severity is indicated by the presence of 4-5 symptoms, and severe OUD is indicated by the presence of ≥6 symptoms.

Additional Study Measures

In addition to OUD severity, we also examined self-reported craving and anhedonia, which are individual difference measures with a priori theoretical relevance to opioid pill choice.

Craving

Self-reported opioid craving was assessed using visual analogue scales, with response options ranging from 0-100, based on items from Wasan et al. (2012; (33)). Weekly craving and current (“right now”) craving were both assessed; however, because of their very high correlation (r=0.93, thus statistically indistinguishable), we only report results from the weekly craving measure in the analyses below.

Anhedonia

We assessed the Snaith-Hamilton Pleasure Scale (SHAPS), concentrating on its anhedonia subscale. The SHAPS is a 14-item self-report that assess hedonic capacity. Responses were made using a 4-point Likert-type scale ranging from ‘definitely agree,’ to ‘strongly disagree.’ A higher total score indicates greater anhedonia.

Statistical Analyses

A repeated measures ANCOVA with 5 levels (Picture Type: pleasant, unpleasant, neutral, pill, blank), with OUD severity included in the model as a between-subjects covariate, examined differences in image choices as a function of OUD symptom Severity. We then tested the ANCOVA for robustness while including the presence/absence of MAT therapy. Our planned analyses of interest were to examine correlations between OUD severity with pill- and pleasant choice. Such relationships between choice and OUD symptom counts were verified with Poisson regression. Then, for correlations that reached significance, we performed the following supplemental analyses as needed. First, supplemental correlations were conducted between choice and symptom counts after splitting the data by OUD diagnosis, with the rationale that a monitoring function of the task for opioid symptomatology was expected to be more relevant (i.e., predictive) among individuals with an OUD diagnosis. Second, for any correlations that may have contained outliers, we repeated the correlation using a non-parametric approach to test for robustness.

The main ANCOVA analysis was followed by additional mixed ANCOVAs, this time including into the model the individual difference measures of interest (craving, anhedonia) each entered one at a time. In these models, craving and anhedonia were each tested for interaction with choice behavior, above and beyond the effect of OUD severity. Thus, each of these ANCOVAs included two interaction terms: the original interaction between Picture Choice and OUD Severity, and the interaction between Picture Choice and the relevant individual difference variable. As with the analyses with OUD severity, significant interactions with covariates were followed by correlation analyses to localize the effect. Finally, specifically for the relationships that reached significance, our last analytical step was to control for the presence/absence of MAT therapy and average pain scores on the Brief Pain Inventory (BPI) in two final ANCOVAs. In total, because three separate ANCOVAs were tested, we specified a Bonferroni-corrected significance level of p=0.0167 (p=0.05/3). Following a significant omnibus effect in the ANCOVA using this corrected threshold, the standard p<0.05 threshold was used for follow-up and supplemental analyses, in order to ensure a clear and comprehensive understanding of the data.

RESULTS

The repeated measures ANCOVA with OUD Severity revealed a main effect of Picture Type [F(4,82)=110.88, p<0.001]: all participants selected pleasant images most frequently (M=22.83, SD=3.59), followed respectively by neutral images (M=19.34, SD=3.06) and blank images (M=12.69, SD= 3.39) (Bonferroni-corrected pairwise comparisons: all p<0.001), and finally by pill images (M=8.27, SD=4.34) and unpleasant images (M=6.87, SD=5.06); choices for the latter two image types did not differ (Bonferroni-corrected pairwise comparison: p=0.60). This main effect was qualified by a significant Picture Type × OUD Severity interaction [F(4,82)=3.73, p=0.008], and this interaction remained significant after controlling for the presence/absence of MAT therapy [F(4,81)=3.71, p=0.008] and BPI average pain [F(4,81)=4.31, p=0.003). Planned follow-up analyses showed that this interaction was driven by a positive correlation between OUD Severity (number of DSM-5 symptoms) and opioid picture choice (r=0.38, p<0.001), but lack of correlation with other four picture types (rs<|0.19|, ps>0.079) (Figure 2). The association between opioid-related choice and higher OUD severity was confirmed with Poisson regression (b=0.07, SE =0.01, p<0.001) (Figure 3). Additional analyses split by OUD diagnosis indicated that, as expected, the relationship between opioid picture choice and OUD symptom count was driven by those participants who met criteria for OUD diagnosis (r=0.47, p=0.006); there was no correlation among those who did not meet criteria for OUD (r=−0.01, p=0.93). Non-parametric (Spearman) correlations similarly showed that pill trial choice was still correlated with OUD symptoms among individuals with OUD (i.e. the OUD group) (ρ=0.45, p=0.008), but not among individuals without OUD (ρ=−0.00, p=0.99), suggesting that the association is not attributable to a potentially extreme score seen in Figure 3. Taken together, results indicate that greater OUD severity was uniquely associated with opioid pill choice.

Figure 2.

Correlation magnitude between OUD severity and image choice. There was a significant positive correlation between OUD severity and opioid pill image choice (p < .05, indicated by asterisk), but lack of significant correlation with the other four image categories.

Figure 3.

Scatterplot of the significant association between OUD severity (symptom count) and opioid pill choice across the sample. Poisson regression statistics are as follows: b = .072, SE = .0129, p < .001, t = 5.58.

We then performed two more ANCOVAs while including the two individual difference covariates: (1) weekly opioid craving and (2) anhedonia, each entered separately and sequentially into the models, with OUD Severity already accounted for. In these models, neither covariate showed significant moderation with Picture Type. For opioid craving, the main effect of Picture Type was still significant [F(4,80)=101.00, p<0.001], but the Picture Type × Craving interaction was not [F(4,80)=0.59, p=0.670] [and neither was the Picture Type × OUD Severity interaction: F(4,80)=0.54, p=0.54], likely due to observed high overlap between craving and symptoms (r=0.73, p<0.001). For anhedonia, likewise, the main effect of Picture Type was still significant [F(4,79)=14.75, p<0.001], but the Picture Type × Anhedonia interaction was not [F(4,79)=0.96, p=0.433] [though here the Picture Type × OUD Severity remained significant: F(4,79)=3.75, p=0.008]. For both variables, given the nonsignificant interactions, follow-up analyses were not performed.

DISCUSSION

In the present study, 87 veterans who were prescribed opioids due to chronic pain completed a picture choice task as a proxy measure of behavioral preference for opioid pills over non-drug salient alternatives (i.e., pleasant, unpleasant, and neutral images; and blank screens). We have used variations of this paradigm in multiple studies and addictions spanning over a decade (24–27, 38, 40, 41), including in opioid misusers (28). Other investigators and laboratories have used similar paradigms with similar success (42–44). The simulated drug-choice task echoes (21, 45), but provides an ethical alternative to, classical drug-choice paradigms such that it can be used in OUD and opioid-using patients (23). Our choice paradigm extends highly robust paradigms assessing passive attention bias or cue-reactivity tasks (46–51), in that our task requires participants to make a choice for viewing images, providing high face validity with respect to the behavior of interest (i.e. the decision to take drugs). In our study, we predicted that patients who had higher OUD severity would demonstrate an enhanced choice for the pill-related images (and reduced choice for the pleasant images), compared with patients with fewer (or no) symptoms of OUD. We also predicted that select individual difference variables, encompassing opioid craving and anhedonia, would modulate opioid-related choice behavior.

Consistent with Hypothesis 1A, we found evidence of significant association between OUD severity and enhanced opioid-related choice behavior, such that individuals with a more severe disease phenotype chose more opioid images for viewing than those with less severe disease. These results expand upon findings from our previous study in opioid-treated pain patients (28). First, we studied a new cohort of participants – U.S. military veterans – a population at a uniquely high risk for OUD and other substance use disorders due to trauma and associated psychiatric symptoms (52). Second, we employed a task with different choice-outcome contingencies: whereas our prior work used a probabilistic choice task with uncertain contingencies, this study used an explicit choice task with certain contingencies that also required rigorous, continuous responding on each trial. From a behavioral economic perspective, probabilistic choice may be thought to reflect ‘intensity’ (preference for opioid images over alternative images when there is no cost associated with the choice), whereas the current task employing explicit choice may be thought to reflect breakpoint [O_max: preference for opioid images over alternative images when there is a cost of effort expenditure; e.g., (38, 53)]. Choice-task findings in OUD also extend our prior work in stimulant users, who likewise select with high frequency drug-related images for viewing in correlation with core addiction-relevant symptoms and neurobiological abnormalities (24–27, 38). The relative ease by which this paradigm can be implemented and performed in clinical contexts, including its ethical viability, makes simulated drug-choice an attractive approach for potentially predicting future OUD outcomes, including relapse, adherence to prescribed medications, and the development of promising new treatments that aim to shift the balance away from drug reinforcement and toward other hedonic rewards (23). Although we did not find evidence in this study for a relationship between OUD severity and pleasant choice, in contrast to Hypothesis 1B, future studies should continue to evaluate the fundamental interplay between opioids and pleasant reinforcers in the context of OUD and opioid misuse.

We did not find support for Hypothesis 2A that craving would further modulate opioid-related choice. When we entered weeklong craving as a covariate into an ANCOVA, the interaction between craving and overall choice behavior was not significant. A relationship was initially expected given that craving is a core feature of addiction that has been well-characterized among individuals with chronic pain and found to predict treatment outcomes, including risk of relapse (54–56), which may be driven by an attentional bias toward opioid cues (8) and may be ameliorated by targeted interventions such as Mindfulness-Oriented Recovery Enhancement (MORE) (51, 57). One explanation for the null finding in our study may be related to a highly significant positive correlation between OUD severity and craving. That is, individual differences in craving do not appear to explain additional variance in opioid-related choice beyond DSM-5 symptom counts (even as the DSM-5 symptoms are far more expansive than craving). Also unexpectedly, our results also did not support Hypothesis 2B that anhedonia would modulate opioid-related choice behavior. A relationship between choice and anhedonia was anticipated given the reported elevation in anhedonia and negative affect among opioid users compared with controls (58–60), and given that improvements in positive affect have been associated with reduced risk of opioid misuse (61, 62). One potential reason for the discrepancy between prior results and our own could relate to differences in the instruments used: while we measured anhedonia using the SHAPS, Garfield and colleagues report improved outcomes with the TEPS (63). However, it is also possible that, particularly in light of the non-significant association between OUD severity and pleasant choice (above), compromised hedonic processing may figure less prominently in this particular choice phenotype than we predicted. Alternatively, chronic pain is also associated with anhedonia, and given that both groups reported comparable levels of pain severity, they may have exhibited similar deficits in choice for pleasurable stimuli (37). Future studies will need to expand upon these results, potentially investigating the interplay between craving, anhedonia, and positive and negative affective states (33, 34, 37, 55, 64–66).

The current study has several limitations. First, associations between variables were examined cross-sectionally, which limits the conclusions that can be drawn about predicting clinical outcomes. We view this study as a first step in a line of research culminating in future prospective and/or experimental investigations that confirm and extend these findings. Given our cross-sectional findings indicating that the task seems to be sensitive to the number of OUD symptoms (i.e., OUD severity), future longitudinal studies may show the task to be useful as a means of tracking changes in OUD severity over the course of treatment. If shown to have significant predictive validity, ultimately this task might be used to guide OUD treatment decisions. Second, while the investigation of drug choice in a veteran population is a unique strength of this study, we acknowledge that such a sample also presents some concerns of generalizability. Veterans not only experience chronic pain, but also multiple additional stressors (i.e., PTSD, depression), that may exert unidentified or unexpected influences on drug choice. Furthermore, given that the participants in the current study were using prescription opioids, the results may not represent drug choice among individuals using illicit opioids, such as heroin. Third, results may have been influenced by demand characteristics or socially desirable responding, wherein participants did not want to behave in ways suggesting a behavioral preoccupation with opioids. This concern was mitigated by having participants complete the task in a private laboratory context, under a NIH certificate of confidentiality, and by providing participants with clear instructions that there were no right or wrong answers on the task, and rather that they should simply choose the images they find most appealing.

In conclusion, using a picture-choice task to measure objective preference for opioid pill images in a sample of U.S. military veterans, those with higher OUD severity showed preferential choice for opioid-related stimuli. Consistent with results from our previous study, which was conducted in a community (rather than veteran) sample of prescription opioid users with chronic pain, choice for opioid images was relatively low overall, but significantly elevated among individuals with OUD symptoms compared with those who took opioids as prescribed. Finally, although opioid image choice was not significantly modulated by either self-reported craving or anhedonia, we have supported a role for our task in characterizing and potentially monitoring addictive behavior among vulnerable patients on LTOT for chronic pain. That is, although our results will need to be empirically verified in future prospective studies, current data provide an important proof-of-concept that our drug-choice task may provide an objective index of OUD severity that can be used to monitor patients and track their severity and/or treatment response into the future.