Impulsive decision-making predicts the course of substance-related and addictive disorders

Background This study investigated whether patterns of impulsive decision-making (i) differ between individuals with DSM-5 substance use disorders (SUD) or non-substance-related addictive disorders (ND) and healthy controls, and (ii) predict the increase of SUD and ND severity after one year. Methods In a prospective-longitudinal community study, 338 individuals (19–27 years, 59% female) were included in one of three groups: SUD (n = 100), ND (n = 118), or healthy controls (n = 120). Group differences in four impulsive decision-making facets were analyzed with the Bayesian priors: delay discounting (mean = 0.37, variance = 0.02), probability discounting for gains and for losses (each − 0.16, 0.02), and loss aversion (− 0.44, 0.02). SUD and ND severity were assessed at baseline and after 1 year (n = 312, 92%). Predictive associations between decision-making and SUD/ND severity changes were analyzed with the Bayesian prior: mean = 0.25, variance = 0.016. Results Compared with controls, the SUD group displayed steeper delay discounting and lower probability discounting for losses; the ND group displayed lower probability discounting for losses (posterior probabilities > 98%). SUD symptom increase after 1 year was predicted by steeper delay discounting and lower loss aversion; ND symptom increase by lower probability discounting for losses and lower loss aversion (posterior probabilities > 98%). There was low evidence for predictive relations between decision-making and the quantity-frequency of addictive behaviours. Discussion Impulsive decision-making characterizes SUD and ND and predicts the course of SUD and ND symptoms but not the engagement in addictive behaviours. Strength of evidence differed between different facets of impulsive decision-making and was mostly weaker than a priori expected. Electronic supplementary material The online version of this article (10.1007/s00213-020-05567-z) contains supplementary material, which is available to authorized users.


Recruitment and participants Respondents and non-respondents
From 2013 to 2016, 18.000 inhabitants aged between 19 and 27 randomly taken from the registration office files of Dresden were invited by post to participate in our study and 1.856 responded to the invitation letter (10.3 %). We descriptively compared respondents and non-respondents regarding their birth year and gender. Respondents were fore likely female and more likely from the birth years 1992 to 1995 compared to 1988 to 1991 or 1996 (Table S1).

Table S1
Gender and birth years of respondents and non-respondents to our study invitation by post.

Sample size calculation
The sample size for the whole project was estimated using Stata 13 for multiple linear regression with power (1-β) = .80, significance level α= 0.05, and five covariates (age, gender, IQ, income, school graduation). We assumed moderate group differences at baseline with R 2 =.05 according previous studies comparing individuals with gambling disorder, alcohol dependence, and Tourette syndrome on measures of executive functions and decision-making (Goudriaan et al. 2005;Goudriaan et al. 2006). The necessary sample size would have been N=235 in total. Furthermore, we assumed a drop-out rate of 30% during the first funding phase of the project (3 years). The final estimated sample size was 330 with 110 in each group.

Measurements
Addictive disorder groups at baseline Note: M = Means; SD = standard deviations 1) Abitur is the German school-leaving qualification required for university entrance.
2) Two participants refused to provide information.
Symptom severity according to groups

Data analyses
Sample for priors to test the first hypothesis (group differences) Table S4 Demographic characteristics of the sample from a previous study (Bernhardt et al. 2017), which was used to estimate the prior distributions.

Table S5
Decision-making parameters separately for the substance use disorder (SUD) group, the non-substancerelated addictive disorder (ND) group, and the control group.   Results of the reverse Bayesian analyses 1), 2) (with posterior, prior, and likelihood distributions) of the group differences in the decision-making parameters at baseline between the substance use disorder (SUD) group or the non-substance-related addictive disorder (ND) group and the control group. Note: Baseline demographic characteristics (age, gender, IQ, income, and school graduation) were included as control variables in all analyses. 1) For the two facets of impulsive decision-making, where we have drawn conclusions for group differences (delay discounting, probability discounting for losses), we have performed a "reverse-Bayes" analysis. Starting from the current prior used for the paper, we reduced the expected value in 0.1 steps until we identified the most pessimistic prior expectation that still allows the conclusion with a probability of 95% (leaving the variance unchanged).
2) For the two facets of impulsive decision-making where we did not draw conclusions for group differences, but the literature suggests weak evidence for group differences (probability discounting for gains, loss aversion), we used a prior with an expected value of zero (leaving the variance unchanged).

Table S8
Results of the Bayesian linear regression analyses (with posterior, prior, and likelihood distributions) of the group differences in the decision-making parameters at baseline between the substance use disorder (SUD) group or the non-substance-related addictive disorder (ND) group and the control group as reference group. Results are without extreme values according to Stata box plot (values outside the lower quartile -1.5 inter quartile range (IQR) and the upper quartile + 1.5 IQR).  Predictors were the dummy-coded groups (substance use disorder (SUD) group or non-substance-related addictive disorders (ND) group, control group as reference). Outcomes were the logarithmic k or λ (for mixed gambles) values as indicators of impulsive decision-making. Positive group differences (x axis) indicate that the SUD or ND group had higher values compared to the controls group. We hypothesized that steeper delay discounting (group difference > 0), lower probability discounting for gains, lower probability discounting for losses, and decreased loss aversion (group differences < 0 each) characterize individuals with SUD and ND compared to healthy controls. As priors for regression coefficients we used normal distributions with expectations and variances as estimated in a previous study from our lab (Bernhardt et al. 2017). Outcomes were the differences between substance use disorder (SUD) or non-substance-related addictive disorder (ND) severity (number of fulfilled criteria) at follow-up minus baseline. Predictors and outcomes were both z-standardized, yielding standardized regression coefficients that have the same range as correlations (x axis). We hypothesized that steeper delay discounting (correlation > 0), lower probability discounting for gains, lower probability discounting for losses, and lower loss aversion (correlations < 0 each) predict increased addictive disorder severity, i.e. an increased number of fulfilled criteria. Concerning the priors, we assumed a probability of 95% that the true association would range between 0 and 0.5 (resp. -0.5 and 0) which corresponds with an expectation of 0.25 (-0.25) and a standard deviation of 0.127 in a normal distribution.