Time is money: Computational modelling of intertemporal decision making in human cocaine addiction

Abstract:

Introduction: Everyday decisions frequently require a compromise between waiting time and outcome value. The extent to which individuals favour smaller- immediate over larger-future rewards is measured by the delay discounting rate k. Abnormal k reflects greater waiting impulsivity and has been associated with several psychiatric disorders, most prominently drug addiction [1]. Thus, the delay discounting construct has been put forward as a research domain criterion (RDoC), which is thought to capture a key transdiagnostic decision making process underlying psychiatric symptoms [2]. However, patients characterised by increased waiting impulsivity do not consistently choose monetary rewards with greater subjective value [3]. To date, no delay discounting studies have addressed this putative choice process directly and its effect on k estimation. Objectives: We aim to challenge the standard method of analysing the delay discounting construct by characterising two parameters that are important for intertemporal decisions: sharpness and discounting rate. We investigated delay discounting in healthy control participants (n1⁄481) and patients with cocaine use disorder (CUD; n1⁄4107), who have been repeatedly shown to have abnormal k. We hypothesise that measuring choice sharpness directly with the hierarchical Bayesian model improves the estimation of k. Methods: We assessed participants’ waiting impulsivity using the Monetary Choice Questionnaire, which requires participants to choose between a small immediate or a larger future reward after a specified delay. For example, "Would you prefer £14 immediately, or £25 in 19 days?” We estimated k by testing several computational methods: the standard hyperbola algorithm [4], a logistic regression [5], single-subject Bayesian, and full hierarchical Bayesian models. Additionally, we included choice sharpness parameter (b) in Bayesian models. The subjective values of each option were calculated using the hyperbolic dis- counting equation. Relative choice preference was calculated as the ratio of delayed and immediate choice value and converted to log odds θ of choosing delayed option with a subject-specific slope b. Results: The Bayesian posterior estimates of log10k and log10b were, respec- tively, -2.24 and 1.74 for control participants and -1.25 and 1.35 for CUD patients (95% highest density interval of group mean difference did not overlap zero). Thus, CUD patients showed steeper discounting and greater indifference in their choices than control participants. The increased k values in CUD patients were evident across all computational methods (standard method, F1,1791⁄4106.3, p