Long-term effects of AnorexiaNervosa on cognitive performanceA Neurobiological, Computational, and Structural Synthesis

Executive Summary

Anorexia nervosa (AN) is a severe psychiatric disorder characterized by restrictive energy intake, low body weight, an intense fear of weight gain, and profound body image distortion. Beyond its systemic physiological sequelae, AN is marked by alterations in neuropsychological performance, frontostriatal brain circuitry, and neurostructural integrity. A central question in eating disorder neuroscience is whether observed cognitive impairments represent state-dependent consequences of severe acute starvation or persistent, trait-like neurodevelopmental endophenotypes that predate illness onset and perpetuate chronicity. Synthesizing multi-decade longitudinal cohorts, computational neurobiology studies, structural neuroimaging, and meta-analytic cognitive evaluations reveals a nuanced model of cognitive performance in AN. Core executive inefficiencies—specifically reduced cognitive flexibility (set-shifting) and weak central coherence (a detail-focused cognitive style)—demonstrate strong trait-like properties. These cognitive profiles persist following physical weight restoration and are present in unaffected first-degree relatives, establishing them as candidate neurodevelopmental endophenotypes. Conversely, global structural brain changes, including widespread cortical thinning and gray matter volume reductions, undergo rapid neurostructural recovery during nutritional rehabilitation. However, subtle regional scarring persists within frontal and cingulate networks. Concurrently, computational paradigms demonstrate that prolonged semistarvation shifts behavioral control from goal-directed prefrontal networks to compulsive, frontostriatal habit circuits, explaining why restrictive eating choices endure long after acute weight normalization.

Neuropsychological Architecture: Trait Endophenotypes versus State-Dependent Deficits

Set-Shifting and Executive Flexibility

Cognitive flexibility, assessed via set-shifting paradigms such as the Wisconsin Card Sorting Test (WCST) and the Trail Making Test, reflects the operational ability to adapt thoughts and behaviors to changing environmental rules. Adult individuals with acute AN consistently exhibit impaired set-shifting abilities relative to healthy controls, displaying rigid, perseverative behavioral patterns. Meta-analytic evaluations confirm a moderate overall effect size for set-shifting deficits in adult AN, with Hedge's g ranging from -0.38 to -0.44 and WCST performance exhibiting a Cohen's d of 0.48. Subtype analyses demonstrate notable heterogeneity across clinical presentations. Individuals diagnosed with the restricting subtype of AN (AN-R) exhibit pronounced set-shifting impairments, yielding a Hedge's g of -0.51. Conversely, individuals presenting with the binge-purge subtype (AN-BP) display smaller, statistically non-significant task-based set-shifting differences (g = -0.18). Longitudinal evaluations tracking recovered individuals (AN-REC) confirm that set-shifting deficits endure long after nutritional stabilization and cognitive-affective remission. Interestingly, task-based set-shifting performance in children and adolescents with acute AN often remains intact (d = 0.25). This age-dependent divergence indicates that while set-shifting inefficiencies exist as subtle latent traits early in life, prolonged illness duration and neurodevelopmental maturation interact to solidify executive rigidities into adult chronic profiles.

Central Coherence Dynamics

Central coherence describes the neurocognitive tendency to integrate incoming information into a holistic, contextualized whole. Individuals with AN characteristically display weak central coherence, characterized by an excessive processing bias toward local details at the expense of global integration. Standardized metrics, including the Rey-Osterrieth Complex Figure Test (ROCFT) and the Matching Familiar Figures Test, demonstrate significant local processing biases in acute AN (g = -0.53), comparable to the effect sizes observed in non-underweight eating disorders such as bulimia nervosa (g = -0.70). In clinical contexts, weak central coherence manifests as obsessive preoccupation with microscopic details, such as calorie counting, body checking, and rigid meal presentation rituals. Although performance-based global processing tasks show variable normalization following weight gain, self-reported detail-focused cognitive processing styles remain elevated post-recovery. Self-reported inflexibility and weak central coherence correlate strongly with lower health-related quality of life (QoL), particularly among individuals with the restricting subtype.

The High-Performing Cognitive Paradox

Despite executive inefficiencies in set-shifting and central coherence, broad intellectual capacity (IQ) in AN is fully preserved or superior to general population averages, with mean IQ scores positioned 5.9 to 10.8 points above normative controls. Computerized cognitive testing batteries, such as CogTrack, demonstrate that acutely ill AN patients display faster reaction times during attention span tasks and significantly higher accuracy in grammatical reasoning compared to healthy controls. Domain-specific testing shows that sustained attention, working memory, episodic memory, and retrieval speed remain unaffected by acute starvation. This paradoxical profile highlights a distinct cognitive architecture: acute starvation does not induce generalized encephalopathic cognitive decline, but rather shifts neurocognitive processing toward hyperfocused, detail-oriented, and rules-based execution.

Endophenotypic Marker Validation

Family and twin studies provide compelling evidence that set-shifting inefficiencies and weak central coherence function as heritable endophenotypes rather than secondary consequences of malnutrition. Unaffected first-degree relatives, including parents and biological sisters, display significant impairments in cognitive flexibility and central coherence relative to healthy control families, exhibiting moderate-to-large effect sizes. Monozygotic twin pairs discordant for AN demonstrate higher concordance in set-shifting performance than dizygotic twin pairs, underscoring the genetic underpinning of these cognitive traits. These cognitive markers predate the clinical onset of eating psychopathology, generating a neurocognitive profile characterized by behavioral rigidity and perfectionism that elevates vulnerability to AN under stressful environmental conditions.

| Neuropsychological Domain | Assessment Paradigms | Acute AN Profile | Weight-Restored / Recovered Profile | Trait / Endophenotype Status |

|---|---|---|---|---|

| Cognitive Flexibility (Set-Shifting) | Wisconsin Card Sort (WCST), Trail Making Test B | Moderate-to-severe impairment in adults (g = -0.38 to -0.51); intact in early adolescents. | Persistent impairment in AN-REC (g = -0.38 to -0.44). | Validated Endophenotype: Present in unaffected first-degree relatives and MZ twins. |

| Central Coherence | Rey-Osterrieth Complex Figure, Embedded Figures Test | Weak global processing; strong local detail bias (g = -0.53). | Self-report details persist; performance metrics show partial recovery. | Candidate Endophenotype: Shared familial local processing preference. |

| Attention & Reaction Speed | CogTrack Battery, Focused Attention Tasks | Superior reaction time accuracy; hyper-focused processing. | Maintained high performance; no significant decay post-weight restoration. | State-Enhanced / Phenotypic: Driven by heightened perfectionism and self-control. |

| Grammatical Reasoning | Computerized Logical Reasoning Batteries | Heightened accuracy compared to healthy control cohorts. | Stable across longitudinal assessments. | Preserved / Phenotypic: Reflects preserved high broad intellectual functioning (IQ). |

| Working & Episodic Memory | Digit Span, Verbal Learning, Pattern Recognition | Broadly intact; minor variations linked strictly to extreme BMI nadir. | Fully normalized with nutritional restoration. | State-Dependent: Minor acute fluctuations resolve upon refeeding. |

Computational Neurobiology and Fronto-Striatal Circuitry

Reinforcement Learning and Prediction Error Processing

Computational psychiatry provides a mechanistic framework for understanding how cognitive processing becomes distorted in AN. Reinforcement learning (RL) models parameterize how individuals update internal reward and punishment expectations based on prediction errors (\(\delta\)), defined as the mathematical mismatch between expected and actual outcomes: where \(R_t\) represents the received reward at time \(t\), and \(V_t\) is the expected value. The expected value is updated via the learning rate parameter (\(\alpha\)): Individuals with AN exhibit altered reinforcement learning parameters characterized by elevated harm avoidance, high punishment sensitivity, and reduced learning rates from positive feedback. Behavioral tasks paired with computational modeling demonstrate that acute and weight-restored AN individuals learn significantly less from positive reinforcement feedback than healthy controls. This blunted positive reward learning rate correlates directly with clinical symptom severity and persists following complete weight restoration, indicating a long-term impairment in neurocomputational reward processing.

The Goal-Directed to Habitual Control Shift

A critical neurobiological mechanism explaining the chronicity of AN is the pathological transition from goal-directed behavioral control to rigid, habitual control within frontostriatal networks. Goal-directed behavioral control relies on prefrontal structures—primarily the ventromedial prefrontal cortex (vmPFC) and ventral striatum—to evaluate actions based on prospective outcome values. In contrast, habitual behavioral control is mediated by dorsal frontostriatal circuits, including the dorsolateral striatum and dorsal caudate, executing automated stimulus-response routines independent of current outcome value. In the initial stages of AN, caloric restriction often functions as a goal-directed behavior driven by conscious weight-loss goals or affective regulation. However, severe prolonged semistarvation accelerates the neuroplastic migration of food-choice processing from ventral striatal goal evaluation to dorsal striatal habit execution. Once habitual control consolidates, restrictive eating actions transform into automated neurobiological routines. Exposure to food cues automatically triggers restrictive choices regardless of metabolic hunger, physical starvation, or expressed recovery goals. Longitudinal modeling confirms that the computational strength of disorder-specific food-intake habits predicts the long-term persistence of restrictive eating choices following inpatient weight normalization.

Neurostructural Dynamics and Brain Restoration

Macrostructural Brain Alterations in Acute Starvation

Acute AN produces widespread, macrostructural brain alterations visible on structural Magnetic Resonance Imaging (sMRI). Severe malnutrition induces pseudoatrophy, characterized by global gray matter volume (GMV) reductions, structural cortical thinning across almost all brain regions, white matter microstructural degradation (reduced fractional anisotropy on diffusion tensor imaging), and ventricular enlargement. Multimodal MRI paradigms demonstrate that these structural reductions are driven primarily by global changes rather than isolated regional atrophy. After statistical adjustment for total global gray matter volume and mean whole-brain cortical thickness, most regional group differences between acute AN patients and controls disappear. The neurobiological drivers of acute pseudoatrophy include cellular and glial remodeling, astrocyte shrinkages, and alterations in brain tissue macromolecular content rather than simple systemic dehydration.

Refeeding Kinematics and Neurostructural Recovery

Longitudinal MRI tracking of adult and adolescent patients undergoing inpatient refeeding demonstrates rapid structural neuroplasticity upon nutritional restoration. Cortical thickness recovers at an average rate of approximately 0.06 mm per month during structured weight rehabilitation. The temporal trajectory of structural brain restoration follows a nonlinear growth curve, with the steepness of recovery concentrated in the early phases of treatment. During initial refeeding, as BMI increases from severe underweight status (BMI ≤ 15.5 kg/m²) toward intermediate thresholds (BMI ≈ 17.5 kg/m²), the rate of cortical thickening and subcortical volume expansion reaches its peak. As weight restoration approaches physiological normalization (BMI ≥ 17.5 kg/m²), the rate of structural brain regeneration gradually plateaus. Patient age serves as a major biological predictor of structural brain restitution. Younger adolescent AN patients display near-complete normalization of global gray matter and cortical thickness following short-term weight restoration, reflecting heightened developmental neuroplasticity. Conversely, older adult patients exhibit diminished restoration rates. Crucially, the capacity for neurostructural recovery is independent of total illness duration, demonstrating that even individuals with long-standing AN retain significant brain restoration potential upon full nutritional recovery.

Residual Neurostructural Alterations

Despite substantial global normalization following weight restoration, specialized surface-based morphometry reveals lingering regional neurostructural anomalies in weight-restored cohorts. Focal residual cortical thinning persists in the superior frontal cortex of adult AN patients following full weight restoration. Similarly, a thinned cortical profile remains within the right pars orbitalis—a region heavily implicated in response inhibition and impulse regulation—in weight-maintained recovered outpatients. Persistent microstructural alterations also occur within the orbitofrontal cortex (OFC) and insular cortex. Because the insula and OFC govern interoceptive awareness, taste processing, and primary reward valuation, lingering structural alterations within these hubs may represent a neurobiological substrate for enduring body image distortion and altered hedonic responses to food.

| Structural Parameter | Acute Starvation State (acAN) | Weight-Restored State (wrAN) | Long-Term Recovered State (recAN) | Primary Biological Mechanism |

|---|---|---|---|---|

| Global Cortical Thickness | Widespread, profound cortical thinning across all lobes. | Rapid normalization (0.06 mm/month recovery rate). | Completely comparable to healthy controls. | Cellular fluid dynamics, macromolecular restoration, glial remodeling. |

| Global Gray Matter Volume (GMV) | Substantial global GMV deficit; elevated CSF space. | Marked volume expansion; rapid initial recovery curve. | Reversible in non-chronic/younger cases; minor persistent deficits in long-term adult cohorts. | Reversal of starvation-induced pseudoatrophy. |

| Superior Frontal Cortex Thickness | Marked focal thinning. | Partial restoration; residual thinning remains vs controls. | Subtle focal thinning persists in subset of chronic adult cases. | Potential localized neurostructural scarring or premorbid regional marker. |

| Right Pars Orbitalis | Severe cortical thinning. | Persistent cortical thinning despite BMI normalization. | Lingering thinning in weight-restored adult outpatients. | Frontostriatal inhibitory control vulnerability. |

| Orbitofrontal & Insula Microstructure | Altered cortical folding and volume loss. | Relative volume elevations or structural reorganization. | Persisting structural alterations in reward valuation networks. | Altered interoceptive and hedonic reward circuit coding. |

Neuroendocrine and Neurotrophic Mediators of Cognitive Function

Brain-Derived Neurotrophic Factor Dynamics

Brain-derived neurotrophic factor (BDNF) is a primary neurotrophin regulating neuronal survival, synaptic plasticity, dendritic growth, and hypothalamic energy balance. Genetic investigations link the Met allele of the functional Val66Met polymorphism (rs6265) in the BDNF gene to increased susceptibility for the restricting subtype of AN. The Met allele alters activity-dependent BDNF secretion, impairing synaptic plasticity and reward learning within frontostriatal networks. Preclinical models of AN demonstrate that severe food restriction downregulates Bdnf mRNA expression within the prefrontal cortex (PFC) and dorsal striatum, directly impairing cognitive flexibility on behavioral tasks such as the Y-maze. Progressive refeeding restores Bdnf levels in the dorsal striatum, confirming a direct metabolic link between caloric intake and neurotrophic support in striatal circuits. In clinical human literature, peripheral serum BDNF patterns display complex dynamics. While smaller longitudinal studies suggest peripheral BDNF levels are suppressed during acute starvation and rise to normal—or even supranormal—levels following multi-year weight restoration, a large-scale mega-analysis (N = 389) controlling for age, sex, and sample storage duration found no statistically significant cross-sectional or longitudinal differences in serum BDNF between acute AN, long-term recovered AN, and healthy controls. Peripheral serum BDNF levels do not serve as a reliable state or trait biomarker for cognitive recovery in individual patients, despite confirmed central neurotrophic alterations within frontostriatal circuits.

Leptin and Semistarvation-Induced Hyperactivity

Leptin, an adipocyte-derived hormone, signals energy reserves to the central nervous system to regulate long-term energy balance. In acute AN, severe loss of adipose tissue suppresses circulating leptin to hypoleptinemic levels. Hypoleptinemia directly mediates semistarvation-induced hyperactivity (SIH), driving excessive, compulsive physical movement. Regression analyses indicate that suppressed leptin levels—rather than elevated cortisol—predict the severity of physical hyperactivity in acute AN. Upon nutritional refeeding, leptin levels surge alongside body fat restoration. In individuals with long-standing, highly severe AN, rapid leptin surges during early inpatient refeeding (BMI < 16 kg/m²) correlate paradoxically with heightened subjective ratings of depression, anxiety, and stress. This relationship suggests that rapid neuroendocrine signaling shifts during early weight restoration exert acute psychological strain before physiological adaptation occurs.

Autobiographical Memory and Autobiographical Planning

Cognitive evaluations in long-standing AN demonstrate persistent deficiencies in Autobiographical Memory (AM) specificity. Individuals with chronic AN exhibit overgeneral memory (OGM), recalling personal past events as broad, categorical summaries rather than detailed, context-specific episodes. Longitudinal evaluations tracking chronic AN cohorts over a 3-year follow-up period demonstrate high temporal stability of overgeneral memory and impaired Episodic Future Thinking (EFT). Deficiencies in retrieving specific personal memories do not deteriorate further over time, nor do they resolve with minor BMI fluctuations. The stability of overgeneral autobiographical memory impairs personal narrative construction, identity consolidation, and adaptive social problem-solving, contributing to the persistent psychosocial disability seen in severe and enduring AN.

Long-Term Clinical Outcomes and Therapeutic Innovations

Multi-Decade Longitudinal Recovery Trajectories

Longitudinal outcome studies tracking AN cohorts over 20 to 25 years provide crucial perspective on recovery trajectories. At a 9-year follow-up, approximately 31.4% of individuals with AN achieve full clinical recovery. However, when cohorts are re-evaluated at 20 to 25 years post-intake, recovery rates increase significantly to between 62.8% and 76%. In contrast to bulimia nervosa—where recovery occurs rapidly within the first decade—recovery from AN is protracted, with roughly half of individuals who remain ill at 9 years progressing to full recovery by 22 years. Major risk factors predicting poor long-term clinical and neurocognitive outcomes include lower BMI at initial presentation, comorbid major depressive disorder at intake, extended duration of illness (>10 years), elevated baseline maturity fears, and persistent binge-purge behaviors. These multi-decade outcome data counter arguments for shifting to purely palliative management in severe and enduring AN, demonstrating that meaningful clinical and neurocognitive improvement remains achievable even after decades of illness.

Cognitive Remediation Therapy Efficacy

Given the prominence of set-shifting inefficiencies and weak central coherence in AN, Cognitive Remediation Therapy (CRT) was developed as an adjunctive neurocognitive intervention. CRT utilizes interactive exercises to increase metacognitive awareness, encouraging patients to recognize their cognitive styles, shift mental sets, and cultivate global processing strategies. Extensive meta-analyses and randomized controlled trials (RCTs) evaluating CRT yield a clear distinction between metacognitive process awareness and objective clinical remediation:

*Metacognitive and Process Benefits: CRT consistently increases patients' self-awareness of cognitive rigidity, improves treatment engagement, and enhances therapeutic rapport.*

*Objective Task and Clinical Metrics: Systematic reviews and meta-analyses show limited-to-no short-term superiority of CRT over control psychological therapies or non-specific cognitive training regarding performance-based neurocognitive task scores, rate of BMI gain, core eating disorder psychopathology, or generic health-related quality of life.*

The limited translation of task-based CRT gains into broader clinical symptom remission highlights the challenge of targeting entrenched frontostriatal habits through abstract cognitive exercises alone.

Synthesis of Multi-Order Insights and Clinical Directions

Synthesizing neuropsychological, computational, structural, and neuroendocrine findings reveals that long-term cognitive performance in anorexia nervosa is dictated by a dynamic interaction between preexisting neurodevelopmental traits and starvation-induced neurobiological alterations.

Trait-State Dualism and Endophenotypic Vulnerability

C