In this paper, we used two approaches to examine predictors of relapse to methamphetamine use following abstinence: a general linear model and a machine-learning algorithm called random forest. As reported in Sections 3.2 and 3.4.2 and Figs. 1 and 2, both methods identified striatal activity during reward processing as a predictor of relapse. Individuals who remained abstinent showed greater activity during large relative to small rewards, whereas those individuals who relapsed showed the opposite effect. In addition to identifying neural processing, the random forest model also identified personality metrics, such as depression and sensation seeking ratings, as valuable predictors of relapse (see Section 3.4.1). The random forest procedure also involved testing the accuracy of the predictors for identifying participants who would relapse. We recently learned that the validation scheme we used with the random forest procedure was biased toward over-estimating model accuracy because we did not test it on a separate sample. Thus, the accuracy estimates we make of 74%, 72%, and 75% for the clinical, fMRI, and combined models, respectively (see Table 3 and Fig. 3) are likely 5–20% higher than one would expect them to be in an independent sample. Importantly, the random forest model identified personality variables, and two streams of evidence converged to suggest that striatal reward processing may be an important variable in determining relapse likelihood. While these findings are not incorrect, we believe we should have pointed out as a significant limitation of our finding that determining the accuracy will require future testing of our model with a separate, independent sample.

Background

Cocaine use disorders (CUDs) have been associated with increased risk-taking behavior. Neuroimaging studies have suggested that altered activity in reward and decision-making circuitry may underlie cocaine user's heightened risk-taking. It remains unclear if this behavior is driven by greater reward salience, lack of appreciation of danger, or another deficit in risk-related processing.

Methods

Twenty-nine CUD participants and forty healthy comparison participants completed the Risky Gains Task during a functional magnetic resonance imaging scan. During the Risky Gains Task, participants choose between a safe option for a small, guaranteed monetary reward and risky options with larger rewards but also the chance to lose money. Frequency of risky choice overall and following a win versus a loss were compared. Neural activity during the decision and outcome phase were examined using linear mixed effects models.

Results

Although the groups did not differ in overall risk-taking frequency, the CUD group chose a risky option more often following a loss. Neuroimaging analyses revealed that the comparison group showed increasing activity in the bilateral ventral striatum as they chose higher-value, risky options, but the CUD group failed to show this increase. During the outcome phase, the CUD group showed a greater decrease in bilateral striatal activity relative to the comparison group when losing the large amount, and this response was correlated with risk-taking frequency after a loss.

Conclusions

The brains of CUD individuals are hypersensitive to losses, leading to increased risk-taking behaviors, and this may help explain why these individuals take drugs despite aversive outcomes.

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.

There is some evidence that neuroimaging can be used to predict relapse among abstinent methamphetamine-dependent (MD) individuals. However, it remains unclear what cognitive and neural processes contribute to relapse. This investigation examined whether insula activation during risk-taking decisions-a process shown to be disrupted in MD-is able to predict susceptibility for relapse. Sixty-eight MD enrolled in a treatment program during early abstinence completed a risk-taking task during functional magnetic resonance imaging. Sixty-three of the sixty-eight individuals were followed up 1 year after the study. Of these, 18 MD reported relapse. The 45 abstinent MD showed patterns of insula activation during risky decisions that resembled those found in prior studies of healthy controls, consisting of lower insula activation during safe decisions paired with higher activation during risky decisions. In contrast, the 18 relapsed MD showed similar insula activation during safe and risky decisions. An increase in one standard deviation in the difference in insula activation between risky and safe choices was associated with a 0.34 odds ratio for relapse at any given time. A median split of insula activation (difference between risky and safe) showed that individuals in the bottom half were two times more likely to relapse. In addition, a model that included several other brain regions increased prediction accuracy compared with insula-based model alone. These results suggest that failure to differentially activate the insula as a function of risk is a part of an altered risk-processing network associated with an increased susceptibility to relapse.

Objective

Relapse rates are consistently high for stimulant user disorders. In order to obtain prognostic information about individuals in treatment, machine learning models have been applied to neuroimaging and clinical data. Yet few efforts have been made to test these models in independent samples or show that they can outperform linear models. In this exploratory study, we examine whether machine learning models relative to linear models provide greater predictive accuracy and less overfitting.

Method

This longitudinal study included 63 methamphetamine-dependent (training sample) and 29 cocaine-dependent (test sample) individuals who completed an MRI scan during residential treatment. Linear and machine learning models predicting relapse at a one-year follow up that were previously developed in the methamphetamine-dependent sample using neuroimaging and clinical variables were applied to the cocaine-dependent sample. Receiver operating characteristic analysis was used to assess performance using area under the curve (AUC) as the primary outcome.

Results

Twelve individuals in the cocaine-dependent sample remained abstinent, and 17 relapsed. The linear models produced more accurate prediction in the training sample than the machine learning models but showed reduced performance in the testing sample, with AUC decreasing by 0.18. The machine learning models produced similar predictive performance in the training and test samples, with AUC changing by 0.03. In the test sample, neither the linear nor the machine learning model predicted relapse at rates above chance.

Conclusions

Although machine learning algorithms may have advantages, in this study neither model's performance was sufficient to be clinically useful. In order to improve predictive models, stronger predictor variables and larger samples are needed.

Aims

To determine if methamphetamine-dependent (MD) individuals exhibit behavioral or neural processing differences in risk-taking relative to healthy comparison participants (CTL).

Design

This was a cross-sectional study comparing two groups' behavior on a risk-taking task and neural processing as assessed using functional magnetic resonance imaging (fMRI).

Settings

The study was conducted in an in-patient treatment center and a research fMRI facility in the United States.

Participants

Sixty-eight recently abstinent MD individuals recruited from a treatment program and 40 CTL recruited from the community completed the study.

Measurements

The study assessed risk-taking behavior (overall and post-loss) using the Risky Gains Task (RGT), sensation-seeking, impulsivity and blood-oxygenation-level-dependent activation in the brain during the decision phase of the RGT.

Findings

Relative to CTL, MD displayed decreased activation in the bilateral rostral anterior cingulate cortex (ACC) and greater activation in the left insula across risky and safe decisions (P < 0.05). Right mid-insula activation among CTL did not vary between risky and safe decisions, but among MD it was higher during risky relative to safe decisions (P < 0.05). Among MD, lower activation in the right rostral ACC (r = -0.39, P < 0.01) and higher activation in the right mid-insula (r = 0.35, P < 0.01) during risky decisions were linked to a higher likelihood of choosing a risky option following a loss.

Conclusions

Methamphetamine-dependent individuals show disrupted risk-related processing in both anterior cingulate and insula, brain areas that have been implicated in cognitive control and interoceptive processing. Attenuated neural processing of risky options may lead to risk-taking despite experiencing negative consequences.