Altered Reinforcement Learning from Reward and Punishment in Anorexia Nervosa: Evidence from Computational Modeling

Abstract

Objectives: Anorexia nervosa (AN) is associated with altered sensitivity to reward and punishment. Few studies have investigated whether this results in aberrant learning. The ability to learn from rewarding and aversive experiences is essential for flexibly adapting to changing environments, yet individuals with AN tend to demonstrate cognitive inflexibility, difficulty set-shifting and altered decision-making. Deficient reinforcement learning may contribute to repeated engagement in maladaptive behavior. Methods: This study investigated learning in AN using a probabilistic associative learning task that separated learning of stimuli via reward from learning via punishment. Forty-two individuals with Diagnostic and Statistical Manual of Mental Disorders (DSM)-5 restricting-type AN were compared to 38 healthy controls (HCs). We applied computational models of reinforcement learning to assess group differences in learning, thought to be driven by violations in expectations, or prediction errors (PEs). Linear regression analyses examined whether learning parameters predicted BMI at discharge. Results: AN had lower learning rates than HC following both positive and negative PE (p < .02), and were less likely to exploit what they had learned. Negative PE on punishment trials predicted lower discharge BMI (p < .001), suggesting individuals with more negative expectancies about avoiding punishment had the poorest outcome. Conclusions: This is the first study to show lower rates of learning in AN following both positive and negative outcomes, with worse punishment learning predicting less weight gain. An inability to modify expectations about avoiding punishment might explain persistence of restricted eating despite negative consequences, and suggests that treatments that modify negative expectancy might be effective in reducing food avoidance in AN.

Testing the role of associative learning in evidence‐based treatments for anorexia nervosa

“…Compared to classical conditioning, there are more studies testing reinforcement learning in AN. Most existing studies have found differences in how individuals with AN perform on reinforcement learning behavioral tasks compared to healthy control groups, particularly on tasks testing the ability to update previous learning (e.g., reversal learning) (Bernardoni et al, 2018(Bernardoni et al, , 2021Filoteo et al, 2014;Foerde et al, 2021;Foerde & Steinglass, 2017;Hildebrandt et al, 2018;Pike et al, 2023;Wierenga et al, 2021). However, while most studies have reported significant group differences in reinforcement learning, the specific alterations observed have varied considerably, with evidence of both intact and altered abilities to learn from rewards and/or punishments, shift learned behavior after it no longer is reinforced, and/or form habits in AN (Reilly et al, 2023).…”

“…While data testing these hypotheses are limited, four studies to date have found associations between performance on reinforcement learning tasks (Filoteo et al, 2014;Wierenga et al, 2021) or neural responses within classical conditioning paradigms (DeGuzman et al, 2017;Frank et al, 2018) and weight gain within treatment.…”

Integrating aspects of affect, reward, and cognition to develop more comprehensive models of binge‐eating pathology

“…Note that this represents a departure from previous EMA studies of affect‐binge eating relationships, which characterized the average change in affect across the entire participant sample (Engel et al, 2013; Schaefer, Smith, et al, 2020; Smyth et al, 2007) Finally, behavioral tasks assessing reward learning, inhibitory control, and delay discounting may be adapted to examine both generalized deficits in these domains, as well as disorder‐specific deficits (e.g., through the use of food stimuli). Notably, many of these tasks can be adapted for use with fMRI or electroencephalography, offering an opportunity to assess the neurobiological underpinnings of reward and inhibitory control processes (Balodis et al, 2015; Forester et al, 2022; Kober & Boswell, 2018), and computational modeling approaches can be meaningfully applied to the behavioral or neural data to help clarify the specific processes underlying deficits or aberrations in these domains (Castagna & Crowley, 2021; Reiter et al, 2017; Wiecki & Frank, 2013; Wierenga et al, 2022).…”

Back to the future: Progressing memory research in eating disorders

“…Additionally, when contingencies between stimuli, actions, and outcomes change over time (e.g., if restriction in the presence of food no longer reduces anxiety or becomes associated with negative outcomes), individuals with EDs may fail to adjust their behavior, reflecting poor “reversal learning.” Lab‐based assessment of basic deficits/enhancements of this type of procedural memory in EDs has been quite mixed, likely secondary to differences in study design (e.g., focus on initial learning vs. reversal learning) or population characteristics (diagnosis, age, illness stage). However, there does broadly appear to be evidence for altered outcome‐guided procedural learning among individuals with EDs (Bernardoni et al, 2017; Celone et al, 2011; Foerde & Steinglass, 2017; Hagan & Forbush, 2021; Hildebrandt et al, 2018; Lao‐Kaim et al, 2015; Ritschel et al, 2017; Shott et al, 2012; Wierenga et al, 2022), particularly with regard to reversal learning (Banca et al, 2016; Haynos et al, 2021; Kollei et al, 2018). Relatedly, there is also evidence that individuals with EDs may persist in learned behaviors (e.g., restriction) in response to a specific stimulus (e.g., food), even when the originally reinforcing/punishing outcomes have become irrelevant (e.g., if restriction is now punishing and no longer associated with social reinforcement; i.e., habit) (Davis et al, 2020; Foerde et al, 2021; Gillan et al, 2016; Seidel et al, 2022).…”

“…When coupled with strong experimental design, these models offer the opportunity to evaluate differing influences on a given behavioral outcome, and test competing explanations of observed behavior through model comparison. Indeed, computational modeling techniques have already begun to push ED memory research forward, providing a careful examination of the memory‐related processes and subprocesses that may contribute to EDs (Bernardoni et al, 2017; Berner et al, 2023; Chan et al, 2014; Deguzman et al, 2017; Reiter et al, 2017; Smith et al, 2020; Wierenga et al, 2022). For example, one recent study using computational modeling found that what appears to be outcome‐guided behavior in AN may actually reflect overreliance on learned stimulus–response associations (Foerde et al, 2021).…”