Figure 1
PRISMA flow diagram for papers included in this study. *Due to the large volume of results, the first 350 items, ordered by relevance, were screened for Google Scholar. **Theoretical papers that proposed a computational approach to AN but did not meet the experimental sample requirements were nevertheless included.
Table 1
Table summarising the characteristics of reviewed studies.
| STUDY | THEME(S) | PARTICIPANT GROUPSWRITTEN AS GROUP NAME (SAMPLE SIZE): MEAN AGE (STANDARD DEVIATION), AND AGE RANGE, IF REPORTED | DIAGNOSTIC CRITERIA FOR THE AN GROUP(S) | COGNITIVE PROCESS | PARADIGM | KEY FINDINGS | CORRELATION OF MODEL PARAMETERS OR TASK PERFORMANCE WITH CLINICALLY RELEVANT FEATURES (SIGNIFICANT AT P < 0.05) |
|---|---|---|---|---|---|---|---|
| Bernardoni et al., 2018 | Reinforcement learning | Acute AN (n = 36): 16.0 (2.6), 12–23 years HC (n = 36): 16.3 (2.6), 12–24 years | Acute AN: participated in a treatment programme | Learning from feedback, decision-making under uncertainty | Probabilistic reversal learning task | Increased learning rates after punishment in AN vs HC. | Not found |
| Bernardoni et al., 2021 | Reinforcement learning | rec-AN (n = 34): 22.3 (2.8), 15–28 years HC (n = 63): 22.0 (2.9), 15–28 years | rec-AN: previously met DSM-IV criteria; no reported symptoms for > 9 months | Learning from feedback, decision-making under uncertainty | Probabilistic reversal learning task | Greater difference in learning rates between punished and rewarded trials in rec-AN vs HC. AN characterised by a learning style associated with low mood. | Not found |
| DeGuzman et al., 2017 | Reinforcement learning | AN (n = 21): 15.2 (2.4), 13–20 years HC (n = 21): 16.4 (1.9), 11–20 years | AN: recruited from a treatment programme; adult participants met DSM-5 criteria Timepoint 1: Acute-AN Timepoint 2: AN-WR: after weight restoration to a BMI ≥ 19.5 | Learning from feedback, punishment sensitivity | Monetary reward task | Negative PEs related to stronger responses in the caudate in Acute AN vs HC. Positive PEs related to stronger responses in the insula in Acute AN vs HC. At timepoint 2, no difference in PE signalling in AN-WR vs HC. | In AN, higher PE signalling in the caudate was associated with worse treatment outcomes. |
| Shott et al., 2012 | Reinforcement learning | AN (n = 21): 25.2 (6.4) Subtype breakdown: AN-R (n = 11), AN-BP (n = 10) HC (n = 19): 27.3 (5.3) | AN: met DSM-IV criteria; recruited from a treatment programme | Feedback learning, decision-making | Category learning task | Impaired implicit category learning in AN vs HC. | Impaired category learning was associated with lower self-reported sensitivity to punishment and higher novelty seeking. |
| Filoteo et al., 2014 | Reinforcement learning; cognitive flexibility | AN-WR (n = 19): 29.7 (6.6), HC (n = 35): 27.7 (5.1) | AN-WR: previously met DSM-IV criteria, no reported symptoms for > 12 months | Learning from feedback, set shifting | Category learning task with rule change | Increased learning speed during initial rule acquisition in AN-WR vs HC. Deficits in set shifting in AN-WR vs HCs. | In AN-WR, a more abnormal learning speed was correlated with shorter duration of weight restoration, and smaller change between the lowest registered and current BMI. |
| Wierenga et al., 2021 | Reinforcement learning | AN (n = 42): 22.8 (9.6), 16–60 years HC (n = 38): 21.6 (4.3), 15–32 years | AN: met DSM-5 criteria; recruited from a treatment programme | Learning from feedback, punishment sensitivity | Probabilistic associative learning task | Lower learning rates and impaired learning from feedback in AN vs HC. | In AN, the magnitude of negative PEs during trials with punishment was associated with worse treatment outcomes. |
| Chan et al., 2014 | Value-based decision-making | AN (n = 94): 25.6 (8.5) BN (n = 63): 26.9 (10.8) HC (n = 67): 25.5 (6.7) Data were collected across three sites. | AN at two sites (n = 81/94): met DSM-IV criteria AN one site (n = 13/94): EDDS questionnaire (Stice et al., 2000) | Probabilistic decision-making | IGT | Impaired task performance in AN vs HC, characterised by lower memory parameter estimates, indicating greater reliance on most recent outcomes for decision-making. Decreased loss sensitivity in AN vs HC in two out of three AN samples. | In AN, the learning/memory parameter from a prospect-valence learning model was positively correlated with BMI. |
| Verharen et al., 2019 | Value-based decision-making | Study 1: AN (n = 60): 27.3 (9.9) HC (n = 55): 24.5 (8.3) Study 2: AN (n = 216): 22.3 (7.3) | AN: met DSM-IV criteria; recruited from a treatment programme | Probabilistic decision-making | IGT | Reduced loss aversion parameter in AN vs HC (gains and losses have similar impact on behaviour in AN, contrary to HC where losses have bigger impact on future choices). | Not found |
| Jenkinson et al., 2023 | Value-based decision-making | Clinical study: Acute AN (n = 31): 24.9 (8.7). AN-WR (n = 23): 26.1 (7.5) HC low disordered eating (n = 38): 22.9 (3.3) HC high disordered eating (n = 35): 22.5 (4.7) Non-clinical studies: Study 0 + Study 1: (n = 170): 32.3 (10.6) Study 2 (n = 315): 23.4 (6.6) | Acute AN: met DSM-5 criteria; participated in a treatment programme AN-WR: diagnosed by a clinician as no longer meeting DSM-5 criteria; no reported symptoms for > 12 months HC high and low disordered eating: BMI, EDE-Q (Fairburn & Beglin, 1994) restraint scores | Decision-making under uncertainty, risk aversion | BART with a body-size condition | Risk taking in Acute AN and AN-WR was modulated by values related to increasing/decreasing body size. Corresponding computational changes in risk aversion, but not loss aversion, were seen in AN-WR. | Not found |
| Decker et al., 2015 | Value-based decision-making | Timepoint 1: Acute AN (n = 59): 25.0 (7.5) HC (n = 39): 24.7 (7.6) Timepoint 2: AN-WR (n = 43) HC (n = 31) | Acute AN: met DSM-5 criteria; participated in a treatment programme Timepoint 2, AN-WR: measurements after weight restoration to a BMI ≥ 19.5 | Intertemporal decision-making, inhibitory control | Delay discounting task | Lower delay discounting rates in Acute AN vs. HC. The difference at timepoint 1 was driven by AN-R. At timepoint 2, no difference in delay discounting rates in AN-WR vs HC. | Not found |
| King et al., 2016 | Value-based decision-making | AN (n = 34): 15.7 (2.5), 12-22 years HC (n = 34): 16.1 (2.4), 12-22 years | AN: met DSM-IV criteria; participated in a treatment programme | Intertemporal decision-making, inhibitory control | ICT | No difference in delay discounting rates in AN vs HC, faster decision speed in AN vs. HC. Decreased brain activation in lateral prefrontal and posterior parietal regions associated with decision-making in AN vs HC. | Not found |
| King et al., 2020 | Value-based decision-making | AN-WR (n = 36): 22.2 (3.3), 17–27 years HC (n = 36): 21.2 (3.4), 17-27 years | AN-WR previously met DSM-IV criteria; no reported symptoms for > 12 months | Intertemporal decision-making, inhibitory control | ICT | No differences in behavioural or brain measures in AN-WR vs. HC. | Not found |
| Ritschel et al., 2015 | Value-based decision-making | AN (n = 34): 15.3 (2.7) rec-AN (n = 33): 21.7 (3.1) HC (n = 54): 18.8 (4.4) | AN: met DSM-IV criteria; recruited from a treatment programme rec-AN: previously met DSM-IV criteria, no reported symptoms for > 6 months | Intertemporal decision-making, inhibitory control | ICT | No difference in delay discounting parameter (k) in AN vs HC. | Not found |
| Steinglass et al., 2012 | Value-based decision-making | Acute AN (n = 36): 24.8 (6.4) HC (n = 28): 25.9 (6.7) | Acute AN: met DSM-IV criteria; participated in a treatment programme | Intertemporal decision-making, inhibitory control | Titration task | Decreased temporal discounting in Acute AN vs HC. | Not found |
| Steinglass et al., 2017 | Value-based decision-making | HC (n = 75): 29.0 (7.6) AN (n = 27): 27.7 (7.5) OCD (n = 50): 29.2 (5.8) SAD (n = 44): 30.0 (4.0) | AN: met DSM-IV criteria; participated in a treatment programme (different stages) | Intertemporal decision-making, inhibitory control | Titration task; ICT | Decreased temporal discounting in AN vs HC. No significant difference in HC vs OCD and SAD. | Anxiety was associated with decreased temporal discounting in all groups. |
| Foerde et al., 2021 | Model-based and model-free control | AN (n = 41): 27.1 (7.0) HC (n = 53): 25.6 (5.0) | AN: met the DSM-5 criteria; recruited from a treatment programme | Goal-directed behaviour, learning, decision-making | Two-step decision task with monetary and food-specific condition | Decreased model-based contribution to learning in AN vs HC in monetary and food conditions. | Not found |
| Onysk & Seriès, 2022 | Model-based and model-free control | ED (n = 35): 30.6 (4.5), 18-38 years HC (n = 32): 26.4 (4.6), 18-38 years | ED: reported being on restrictive diet in an attempt to lose weight; scored ≥ 14 on disordered eating and body image preoccupation on the EAT-26 (Garner et al., 1982) and AAI (Veale et al., 2014) | Goal-directed behaviour, learning, decision-making, body image preoccupation | Two-step decision task with monetary and body image disturbance condition | Decreased model-based and model-free contributions to learning in ED vs HC, with a greater effect in the body image disturbance condition. | The difference between model-based learning in the neutral and body image disturbance condition correlated with self-reported scores on disordered eating and body image preoccupation questionnaires. |
| Pike et al., 2023 | Cognitive flexibility; reinforcement learning | rec-AN (n = 25): 23.5 (3.8) Subclinical ED (n = 25): 23.8 (2.8) HC (n = 32): 25.0 (6.4) | rec-AN: self-reported former AN diagnosis from a healthcare professional; no reported symptoms for > 12 months; did not meet DSM-5 criteria for ED subclinical ED: scored ≥ 20 on the EAT- 26 (Garner et al., 1982). | Adaptive learning, set shifting | The volatility task | Elevated learning rate adjustments in response to volatility in rec-AN vs HC. | Not found |
| Neuser et al., 2020 | Theory-based accounts; value-based decision-making | Punishment sensitivity, reward sensitivity | Bandit task simulations | Low and invariable reward sensitivity associated with lower calorie intake in model simulations. | |||
| Rigoli & Martinelli, 2021 | Theory-based accounts; value-based decision-making | Choice evaluation | High reference point as a proposed explanation for behavioural manifestations of AN. |
[i] Note. Abbreviations: AN = Anorexia Nervosa; AN-R = Anorexia Nervosa Restricting Subtype; AN-BP = Anorexia Nervosa Binge-Purge Subtype; HC = Healthy Control; AN-WR = Weight Restored Anorexia Nervosa; BN = Bulimia Nervosa; ED = Eating Disorder; OCD = Obsessive-Compulsive Disorder; SAD = Social Anxiety Disorder; rec-AN = Recovered Anorexia Nervosa; IGT = Iowa Gambling Task; BART = Balloon Analogue Risk Task; ICT = Intertemporal Choice Task; PE = Prediction Error; BMI = Body Mass Index; DSM-IV/5 = Diagnostic and Statistical Manual of Mental Disorders, 4th/5th Edition; EDDS = Eating Disorder Diagnostic Scale; EAT-26 = Eating Attitudes Test; AAI = Appearance Anxiety Inventory; EDE-Q = Eating Disorders Examination Questionnaire.