Elsevier

Drug and Alcohol Dependence

Volume 152, 1 July 2015, Pages 93-101
Drug and Alcohol Dependence

Individualized relapse prediction: Personality measures and striatal and insular activity during reward-processing robustly predict relapse

https://doi.org/10.1016/j.drugalcdep.2015.04.018Get rights and content

Highlights

  • We did fMRI on abstinent methamphetamine-dependent individuals and determined who relapsed.

  • We used a robust classification technique called random forest to generate individual-level predictions.

  • The random forest model was consistent with a standard linear model.

  • Our models performed well, with specificity, sensitivity and ROC AUC around 0.7.

  • Our results suggest that neuroimaging can be developed to predict individual clinical outcomes.

Abstract

Background

Nearly half of individuals with substance use disorders relapse in the year after treatment. A diagnostic tool to help clinicians make decisions regarding treatment does not exist for psychiatric conditions. Identifying individuals with high risk for relapse to substance use following abstinence has profound clinical consequences. This study aimed to develop neuroimaging as a robust tool to predict relapse.

Methods

68 methamphetamine-dependent adults (15 female) were recruited from 28-day inpatient treatment. During treatment, participants completed a functional MRI scan that examined brain activation during reward processing. Patients were followed 1 year later to assess abstinence. We examined brain activation during reward processing between relapsing and abstaining individuals and employed three random forest prediction models (clinical and personality measures, neuroimaging measures, a combined model) to generate predictions for each participant regarding their relapse likelihood.

Results

18 individuals relapsed. There were significant group by reward–size interactions for neural activation in the left insula and right striatum for rewards. Abstaining individuals showed increased activation for large, risky relative to small, safe rewards, whereas relapsing individuals failed to show differential activation between reward types. All three random forest models yielded good test characteristics such that a positive test for relapse yielded a likelihood ratio 2.63, whereas a negative test had a likelihood ratio of 0.48.

Conclusions

These findings suggest that neuroimaging can be developed in combination with other measures as an instrument to predict relapse, advancing tools providers can use to make decisions about individualized treatment of substance use disorders.

Introduction

Relapse is a vexing problem in addictive disorders and, typically, only 40–60% of individuals with addictive disorders are able to maintain abstinence for more than a year after initiating treatment (Hunt et al., 1971, McLellan et al., 2000). Since numerous studies have suggested that treatment can lower relapse rates (Baker et al., 2001, Irvin et al., 1999, Kosten and O’Connor, 2003, Lancaster et al., 2006, Schmitz et al., 2001), identifying treatment-seeking patients at greatest risk of relapse could help clinicians to appropriate more resources to those individuals to more effectively reduce relapse rates. Previous studies have shown that demographic (e.g., lower socioeconomic status; Mclellan et al., 1994), social (e.g., lack of family support; National Institute of Drug Abuse, 1999), and neuroimaging measures (Janes et al., 2010, Paulus et al., 2005; e.g., failure to show differential activation during risky and safe decisions; Gowin et al., 2014a), can indicate relapse likelihood. More recent investigations have used machine learning techniques to predict individual outcomes (Connor et al., 2007, Weinstein et al., 2009). To date, few such studies have used brain imaging measures and have focused on making individually specific predictions. There is some indication that the combination of imaging and sophisticated analytic approaches may provide sufficient prediction accuracy that would allow one to develop prognostic tests of relapse. Such tests could aid a clinician in providing a patient-specific risk assessment that could be used to objectively communicate risk to the patient or change the course of treatment to reduce risk status.

One proposed marker of substance use disorders (SUDs), including methamphetamine dependence (MD; May et al., 2013, Schouw et al., 2013, Stewart et al., 2014), is altered neural response of the limbic reward system (Koob, 2013, Volkow and Fowler, 2000). There are two prominent hypotheses on how the response changes: individuals with SUDs may have hyper- or hypo-activation in response to rewarding stimuli, reflecting either enhanced incentive salience or reward deficiency, respectively. The incentive salience hypothesis derives from evidence that repeated pairing of a cue with a rewarding substance leads to enhanced dopaminergic responding, and drug-craving, when shown the cue (Berridge, 2012). The reward deficiency hypothesis derives from evidence that individuals with SUDs have impaired function of the dopamine reward system, and thus have lower response to rewards such as food, and may use substances to enhance dopamine signaling (Blum et al., 2012). A recent review suggests that the presence of drug cues may modulate reward circuitry activation, where drug cues enhance reward circuitry activation relative to controls, but natural rewards produce lower levels of activity (Leyton and Vezina, 2013, Limbrick-Oldfield et al., 2013). Corroborating this, several studies using monetary or food rewards have shown that individuals with SUDs relative to controls show decreased activation in the striatum, amygdala and insula when viewing or receiving rewards (Ihssen et al., 2011, Jia et al., 2011, Konova et al., 2012, Peters et al., 2011). The ability to stimulate reward circuitry through natural rewards may diminish the desire to stimulate it through substance use, potentially reducing the risk of relapse. It remains unclear whether processing of non-drug rewards during early abstinence can distinguish between individuals who will relapse or remain abstinent.

In a previous study, we examined early-abstinent MD during the decision phase of a risk-taking task and showed that a lack of differentiation between safe and risky options distinguished individuals who would relapse (Gowin et al., 2014a). That study attempted to identify processing differences between individuals who relapse versus abstain (i.e., disrupted risky decision-making). Here, we use data from the same sample to focus on a different question: can neuroimaging be developed as a practical prediction tool to identify individuals at risk of relapse? Improving diagnosis of SUDs is a critical issue to the field (Volkow and Baler, 2013). To address this question, we use a novel statistical model to determine how well neuroimaging can be used to predict clinical outcomes in combination with clinical, demographic and behavioral measures. We wanted to address a problem in neuroimaging prediction models identified by Whelan and Garavan (2013); they showed that the failure of neuroimaging studies to use out-of-sample data disposes them to inflate prediction estimates. We reduced the risk of inflated estimates by using random forest (Breiman, 2001), a robust model that employs a training and testing set. We hypothesized that those individuals who have the greatest difficulty in differentially processing levels of reward, i.e., the neural activation difference to small versus large rewards, might be at greatest risk for relapse. Moreover, we aimed to examine whether a machine learning approach, i.e., random forest, using neural activation during reward could be used to develop a robust test to assess relapse risk of individual participants. Support for the hypotheses and evidence for a robust test would integrate reward-processing dysfunctions with a practically useful tool that would make a significant contribution to addiction medicine.

Section snippets

Sample

Sixty-eight (fifteen female) MD individuals were recruited through 28-day inpatient drug treatment programs at the Veterans Affairs San Diego Healthcare System and Scripps Green Hospital (La Jolla, CA). Both treatment programs employ 12-step models, daily education and exercise, and require participants to attend Narcotics Anonymous meetings. All participants completed the 28-day program and consented to participate in a clinical interview, a brain scan, and a follow-up phone interview one year

Group characteristics

There were no significant differences between the group that relapsed and the group that remained abstinent in demographics or drug use history (p > 0.05; Table 1). Mean number of days to relapse among MD who relapsed was 175.1 (SEM = 29.3). A chi-squared analysis showed that the relapse group (21%) had a higher prevalence of current marijuana dependence relative to the abstinent group (4%), but the small number of participants meeting this criteria (N = 6) precludes a subgroup analysis. MD who

Discussion

We examined whether brain activation during reward-processing can accurately predict which methamphetamine-dependent individuals will relapse in the year following treatment. Our results suggest that the degree to which the striatum differentially processes large, risky versus small, safe rewards is a robust predictor of relapse. In particular, those individuals who show brain activation that fails to differentiate reward magnitudes relapsed sooner. The present results complement our previous

Role of funding sources

This work was supported by grants from the National Institute on Drug Abuse (R01-DA016663, P20-DA027834, R01-DA027797, and R01-DA018307 as well as a VA Merit Grant to Martin Paulus). Sponsors played no role in the design, conduct of the study, collection, management, analysis, and interpretation of the data; or with preparation, review, or approval of the manuscript.

Contributors

Authors MW, SFT and MPP designed and implemented the study. Authors JLG, TMB and MPP analyzed the data. All authors contributed to and have approved the final manuscript.

Conflict of interest

All authors declare that they have no conflicts of interest.

Acknowledgments

Martin Paulus had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. We would like to thank Dr. F. Berger, T. Flagan, H. Donovan, D. Leland, M. Mortezaei and B. Friedrich for assistance and support during data acquisition. We would like to thank Drs. Jennifer Stewart and Anna Konova for critical reviews of the manuscript.

References (58)

  • E.H. Limbrick-Oldfield et al.

    Fronto-striatal dysregulation in drug addiction and pathological gambling: consistent inconsistencies?

    NeuroImage Clin.

    (2013)
  • A.C. May et al.

    Methamphetamine dependent individuals show attenuated brain response to pleasant interoceptive stimuli

    Drug Alcohol Depend.

    (2013)
  • R.R. McCrae et al.

    A contemplated revision of the NEO Five-Factor Inventory

    Personal. Individ. Differ.

    (2004)
  • M.P. Paulus et al.

    Increased activation in the right insula during risk-taking decision making is related to harm avoidance and neuroticism

    Neuroimage

    (2003)
  • A. Pierucci-Lagha et al.

    Diagnostic reliability of the Semi-structured Assessment for Drug Dependence and Alcoholism (SSADDA)

    Drug Alcohol Depend.

    (2005)
  • J.M. Schmitz et al.

    Naltrexone and relapse prevention treatment for cocaine-dependent patients

    Addict. Behav.

    (2001)
  • M.L. Schouw et al.

    Dopaminergic dysfunction in abstinent dexamphetamine users: results from a pharmacological fMRI study using a reward anticipation task and a methylphenidate challenge

    Drug Alcohol Depend.

    (2013)
  • M. Zuckerman

    Item revisions in the Sensation Seeking Scale form V (SSS-V)

    Personal. Individ. Differ.

    (1996)
  • A. Baker et al.

    Randomized controlled trial of brief cognitive-behavioural interventions among regular users of amphetamine

    Addiction

    (2001)
  • T.M. Ball et al.

    Single-subject anxiety treatment outcome prediction using functional neuroimaging

    Neuropsychopharmacology

    (2013)
  • A.T. Beck et al.

    An inventory for measuring depression

    Arch. Gen. Psychiatry

    (1961)
  • K.C. Berridge

    From prediction error to incentive salience: mesolimbic computation of reward motivation

    Eur. J. Neurosci.

    (2012)
  • K. Blum et al.

    “Liking” and “wanting” linked to Reward Deficiency Syndrome (RDS): hypothesizing differential responsivity in brain reward circuitry

    Curr. Pharm. Des.

    (2012)
  • L. Breiman

    Random forests

    Mach. Learn.

    (2001)
  • L. Breiman et al.

    Random Forests

    (2001)
  • J. Brown et al.

    Self-reports by alcohol and drug abuse inpatients: factors affecting reliability and validity

    Br. J. Addict.

    (1992)
  • K.K. Bucholz et al.

    A new, semi-structured psychiatric interview for use in genetic linkage studies: a report on the reliability of the SSAGA

    J. Stud. Alcohol

    (1994)
  • C.R. Cloninger

    A systematic method for clinical description and classification of personality variants. A proposal

    Arch. Gen. Psychiatry

    (1987)
  • J.P. Connor et al.

    The application of machine learning techniques as an adjunct to clinical decision making in alcohol dependence treatment

    Subst. Use Misuse

    (2007)
  • Cited by (53)

    • A review of functional brain differences predicting relapse in substance use disorder: Actionable targets for new methods of noninvasive brain stimulation

      2022, Neuroscience and Biobehavioral Reviews
      Citation Excerpt :

      In later trials of the rock/paper/scissors task in CUD (Stewart et al., 2014b), and during stop errors in the stop-signal task in CUD (Luo et al., 2013), less activation was predictive of a greater likelihood of relapse. Similarly, in patients with MUD during the rock/paper/scissor task, less activation during wining, tying, and losing outcomes combined and during tying outcomes alone was associated with a greater chance of relapse (Stewart et al., 2014a), while a smaller difference between activation patterns during small, safe rewards and large, risky rewards in the risky gains task was associated with a greater chance of return to methamphetamine use among persons with MUD over 48 weeks (Gowin et al., 2015). In contrast, abstainers in a combined stimulant disorder population had a larger negative BOLD response in the ACC during presentation of the distractor stimuli of the oddball task (Clark et al., 2014),

    • Insular Cortical circuits

      2022, Neurocircuitry of Addiction
    • Disturbances across whole brain networks during reward anticipation in an abstinent addiction population

      2020, NeuroImage: Clinical
      Citation Excerpt :

      Reward processing is mediated by a collection of subcortical and prefrontal cortical regions (Haber and Knutson, 2010; Knutson et al., 2001; O'Doherty et al., 2001) that are connected to form a complex neural network encoding various types of rewarding stimuli (Belin and Everitt, 2008; Draganski et al., 2008; Haber and Knutson, 2010). Substance addiction disorders are associated with disturbances within this reward network during the processing of non-drug rewards (Balodis and Potenza, 2015; Hommer et al., 2011; Just et al., 2019; Koob and Le Moal, 2005; Luijten et al., 2017; Wrase et al., 2007), and which are associated with drug relapse during abstinence (Gowin et al., 2015; Stewart et al., 2014). Most studies in addiction populations, however, only probe regional differences in brain functioning, and do not attempt to elucidate features of global connectivity across networks that respond during certain psychological processes, such as reward.

    View all citing articles on Scopus

    Supplementary material can be found by accessing the online version of this paper at http://dx.doi.org/10.1016/j.drugalcdep.2015.04.018.

    View full text