Changes in Loss Sensitivity During Treatment in Concurrent Disorders Inpatients: A Computational Model Approach to Assessing Risky Decision-Making

Abstract

Background: Recent studies have employed computational modeling to characterize deficits in aspects of decision-making not otherwise detected using traditional behavioral task outcomes. While prospect utility-based modeling has shown to differentiate decision-making patterns between users of different drugs, its relevance in the context of treatment has yet to be examined. This study investigated model-based decision-making as it relates to treatment outcome in inpatients with co-occurring mental health and substance use disorders.

Methods: 50 patients (Mage = 38.5, SD = 11.4; 16F) completed the Cambridge Gambling Task (CGT) within 2 weeks of admission (baseline) and 6 months into treatment (follow-up), and 50 controls (Mage = 31.9, SD = 10.0; 25F) completed CGT under a single outpatient session. We evaluated 4 traditional CGT outputs and 5 decisional processes derived from the Cumulative Model. Psychiatric diagnoses and discharge data were retrieved from patient health records.

Results: Groups were similar in age, sex, and premorbid IQ. Differences in years of education were included as covariates across all group comparisons. All patients had ≥1 mental health diagnosis, with 80% having >1 substance use disorder. On the CGT, patients showed greater Deliberation Time and Delay Aversion than controls. Estimated model parameters revealed higher Delayed Reward Discounting, and lower Probability Distortion and Loss Sensitivity in patients relative to controls. From baseline to follow-up, patients (n = 24) showed a decrease in model-derived Loss Sensitivity and Color Choice Bias. Lastly, poorer Quality of Decision-Making and Choice Consistency, and greater Color Choice Bias independently predicted higher likelihood of treatment dropout, while none were significant in relation to treatment length of stay.

Conclusion: This is the first study to assess a computational model of decision-making in the context of treatment for concurrent disorders. Patients were more impulsive and slower to deliberate choice than controls. While both traditional and computational outcomes predicted treatment adherence in patients, findings suggest computational methods are able to capture treatment-sensitive aspects of decision-making not accessible via traditional methods. Further research is needed to confirm findings as well as investigate the relationship between model-based decision-making and post-treatment outcomes.

Keywords: concurrent disorders; decision-making; drug use; impulsivity; mental health; treatment outcome.

Figure 1

Trial Schematic of the Cambridge Gambling Task. Ten boxes are displayed varying in proportions of red to blue. Respondents guess which color hides a yellow token. Participants are then prompted to select a bet, with bet amounts appearing in either ascending or descending order. If respondents guess correctly or incorrectly, the selected bet amount is either added or subtracted to their total points.

Figure 2

Group differences in mean scores on behavioral and computational CGT measures. Error bars show standard error. *p < 0.05, **p < 0.01.

Figure 3

Cumulative model parameter estimates in patients from baseline to 6 months into treatment, showing decreases in Color Choice Bias and Loss Sensitivity over time. Error bars represent standard error. *p < 0.05, and **p < 0.01.