Interoceptive Prediction Error and the Bayesian Brain Fog Framework
The predictive processing framework proposes that the brain continuously generates predictions about bodily state and updates them against sensory evidence. This section applies this framework to ME/CFS: chronic interoceptive prediction errors — the brain’s model of body state persistently diverging from actual physiological signals — could produce the fatigue, cognitive effort-intolerance, and sensory hypersensitivity characteristic of ME/CFS without requiring ongoing peripheral pathology. This is not a psychosomatic hypothesis; rather, it explains how peripheral dysfunction (e.g., autonomic, immune) becomes encoded in predictive brain circuits and perpetuates symptoms even during partial physiological recovery.
1 The Free-Energy Principle and Predictive Processing
The free-energy principle, proposed by Friston, holds that biological systems minimize surprise (formally: variational free energy, a bound on the surprise of sensory observations) by maintaining an internal generative model of their environment and updating it against incoming sensory evidence (Friston 2010). In predictive processing terms: perception is the brain’s best explanation of sensory data, not a direct read-out; prediction errors (mismatches between model and reality) propagate upward to update the model; action drives the body to fulfill the model’s predictions. Applied to interoception the brain’s modeling of the body’s internal state this framework predicts that persistent, genuine perturbations in afferent bodily signals will produce persistent, amplified prediction errors experienced phenomenologically as fatigue, effort-intolerance and pain (Friston 2010).
2 Interoceptive Hyper-Vigilance in ME/CFS
In ME/CFS, chronic peripheral dysfunction (autonomic dysregulation, immune activation, impaired oxygen delivery) generates persistently aberrant afferent interoceptive signals. The brain’s generative model, attempting to minimize prediction error, increases the precision weighting assigned to interoceptive signals effectively amplifying internal body awareness and reducing the threshold for detecting deviation from homeostatic norms. This precision upweighting manifests clinically as: heightened heartbeat discrimination accuracy, lower pain pressure thresholds, cognitive effort-intolerance, and the subjective experience of fatigue as a physiological state rather than a psychological one. Importantly, this is not a psychosomatic mechanism; it is the brain’s rational response to genuine physiological disorder. (Certainty: 0.50. Note: the inflammation→interoceptive hub pathway (Inflammation Changes How the Brain Senses the Body, cert 0.70) provides biological plausibility for this computational framework — both derive from the same root evidence (Harrison 2009 n=16, Zhang/Wager 2025, Savitz/Harrison 2018) and are not independent reinforcement. A Phase 7 +0.05 bump was applied and subsequently reversed on review as circular — both hypotheses share the same evidence base. Mechanism plausible; direct tests of prior beliefs and precision weighting in ME/CFS are limited.)
Empirical support comes from case-control studies in post-infective fatigue syndrome: patients demonstrate significantly higher accuracy on heartbeat discrimination tasks, lower pressure pain thresholds, and a distinct cardiac response profile characterized by insensitivity to task difficulty and absence of habituation (Kadota et al. 2010). Heightened interoceptive sensitivity correlated strongly with concurrent symptoms. Complementary structural neuroimaging shows increased grey matter in the insula (the primary interoceptive cortex) in ME/CFS versus controls, with white matter reductions in brainstem pathways. Tenderness (a proxy for central sensitization) and interoceptive symptom burden are strongly correlated in ME/CFS and Gulf War Illness (Chen et al. 2025), suggesting that nociplastic and interoceptive mechanisms may co-amplify each other.
3 Distinction from Psychosomatic Frameworks
The interoceptive prediction error model is explicitly distinguished from psychosomatic or functional somatic syndrome models in two critical respects. First, the primary afferent signals in ME/CFS are genuinely abnormal (reduced VO2max, impaired autonomic function, elevated lactate, abnormal cytokine profiles), not normal signals misinterpreted by an anxious mind. Second, the model predicts that symptom persistence should track the persistence of peripheral dysfunction, not psychological intervention alone — consistent with the observation that CBT/GET produces no sustained physiological improvement on objective exercise testing (Keller et al. 2024).
4 Why Graded Exposure Fails
Graded exposure therapy operates on the premise that catastrophizing and fear-avoidance maintain symptoms by preventing corrective experiences. This is appropriate when avoidance is maintained by maladaptive beliefs about safe activities. However, two lines of evidence directly challenge this premise in chronic fatigue. First, Poort et al. (2021, n=134, RCT mediation analysis) found that changes in catastrophizing did not mediate fatigue improvement in either GET or CBT — fatigue reduction was mediated by physical conditioning and perceived activity, not by cognitive variables (Poort et al. 2021). If catastrophizing does not drive treatment outcomes even in CBT for fatigue, targeting it is not supported by mechanistic evidence. Second, two-day CPET demonstrates that the belief “exercise causes harm” is physiologically accurate: Day 2 objective capacity declines are reproducible and measurable (Keller et al. 2024) (Campen, Rowe, and Visser 2020), though a 2026 null replication did not find group-average decline (Mancini et al. 2026) (elevated RPE and CI are consistent). Exposing patients to an activity that genuinely produces physiological harm (NLRP3 activation, ASIC3 upregulation, immune surge) to demonstrate it is “safe” is both theoretically inconsistent with the interoceptive error model and inconsistent with the physiological evidence.
5 Therapeutic Implications: Interoceptive Retraining
Consistent with the model, therapeutic approaches that improve interoceptive resolution — the ability to discriminate internal signals accurately rather than amplify them non-specifically — rather than suppress interoceptive awareness, are mechanistically predicted to be beneficial:
- Heart rate variability (HRV) biofeedback: Provides high-resolution, real-time autonomic feedback, potentially improving the precision and accuracy of cardiovascular interoceptive prediction, reducing allostatic load.
- Pacing within energy envelope: By remaining below the ventilatory threshold, pacing prevents the generation of genuine metabolic danger signals, reducing the afferent input driving prediction error.
- Interoceptive awareness training: Mindfulness-based approaches that explicitly target interoceptive attention have been explored in chronic pain and fatigue; effects are modest and require careful implementation to avoid effort-based exacerbation in ME/CFS. No interoceptive training/rehabilitation trials have been conducted in ME/CFS specifically — a complete evidence gap.
6 Inflammation-to-Interoception: The Immune-Interoceptive Axis
The sections above describe interoceptive predictive processing in abstract computational terms. This section provides the biological anchor: the experimental evidence that peripheral inflammation directly modulates the brain’s interoceptive hubs.
When the immune system is activated, the brain regions responsible for tracking the body’s internal state—particularly the insula and anterior cingulate cortex—change their activity patterns. This means the immune dysfunction documented in ME/CFS has a direct anatomical target: the same brain areas that tell you whether your heart is racing, your gut is moving, or your muscles are tense.
Harrison et al. (2009, n=16) gave healthy volunteers a typhoid vaccine, which triggers a brief, controlled immune response. Brain scans showed that this inflammatory signal reduced activity in the mid-cingulate and increased connectivity between the insula and anterior cingulate cortex during an emotional task (Harrison et al. 2009). In other words, inflammation alters the brain’s body-sensing network within hours—in healthy people. Savitz and Harrison (2018) reviewed the full evidence and found that immune signals reach the brain through two routes: (1) the vagus nerve, which carries chemical signals from the body directly to the brainstem, and (2) leaky spots in the blood-brain barrier where immune molecules can diffuse into brain tissue (Savitz and Harrison 2018).
Zhang, Wager et al. (2025, n=200+, Nature Neuroscience) used ultra-high-resolution 7 Tesla fMRI to map this entire body-sensing network in detail: anterior insula, dorsal anterior cingulate, medial prefrontal cortex, hypothalamus, amygdala, and brainstem autonomic nuclei (Zhang, Wager, et al. 2025). These are precisely the regions that appear abnormal in ME/CFS brain scans. Baraniuk et al. (2025) showed that tenderness (a measure of pain sensitivity) and body-sensing symptom burden are strongly correlated in ME/CFS and Gulf War Illness, suggesting that pain amplification and body-sensing amplification happen in the same brain regions (Chen et al. 2025).
(Certainty: 0.70. (0.65→0.70: reinforced by Zhang/Wager 2025 7T fMRI allostatic-interoceptive system mapping, n=200+, Nature Neuroscience). The pathway from inflammation to altered body-sensing is well-established in healthy people given a temporary immune challenge. It has NOT been directly measured in ME/CFS patients. The brain network map is well-replicated in large samples and provides the anatomical blueprint.)
The inflammation-to-interoception pathway is detailed in Chapter Immune System Dysfunction (Section Sickness Behavior as Overarching Integrative Framework), where the three cytokine-to-brain routes (vagal, humoral, transport) are described (Dantzer et al. 2000) (McCusker and Kelley 2013). The Harrison (2009) and Savitz/Harrison (2018) evidence completes the circuit: the same interoceptive regions mapped by Zhang, Wager et al. (2025) are the targets of inflammation-induced modulation. This makes a specific, testable prediction: if fatigue in ME/CFS is driven by CNS-confined inflammation (as Omdal et al. 2026 (Omdal et al. 2026) demonstrated — normal peripheral cytokines with disabling fatigue), the interoceptive dysfunction should persist even when blood cytokines are normal, because the relevant inflammatory signaling is occurring behind the blood-brain barrier, at hypothalamic mast cells, circumventricular organs, and vagal paraganglia.
7 The Interoceptive Accuracy-Sensibility Dissociation
A critical question for the interoceptive model in ME/CFS is whether patients have elevated interoceptive accuracy (consistent with hypervigilance) or reduced interoceptive accuracy (consistent with signal corruption). The answer may be: both, depending on which dimension is measured.
ME/CFS patients are predicted to show a specific pattern: they are intensely aware of their body (every heartbeat, every gut sensation feels amplified and hard to ignore), yet when tested objectively—for example, counting their own heartbeats without touching their pulse—they perform worse than healthy people. They feel more body sensation but perceive it less accurately.
This dissociation is defined by the Garfinkel et al. (2015) framework, which splits body-sensing into three layers (Garfinkel et al. 2015): (1) interoceptive accuracy (how well you actually detect internal signals), (2) interoceptive sensibility (how much body sensation you subjectively feel), and (3) interoceptive awareness (how accurately you know whether you’re good or bad at detecting signals—metacognitive insight). The predicted ME/CFS pattern—high sensibility, low accuracy, low metacognitive awareness—would distinguish it from: healthy people (low-normal sensibility, high accuracy), anxiety disorders (where both accuracy and sensibility tend to be elevated, with preserved metacognitive insight), and autism spectrum conditions (where hyper-precise top-down predictions can produce elevated accuracy with variable sensibility and awareness).
Why would this happen? When inflammatory signals traveling up the vagus nerve corrupt the body’s data stream, the brain compensates by turning up the volume on all body signals—producing hypervigilance—but the underlying data is too noisy to resolve, so accuracy and metacognitive insight remain poor. The review by Quadt, Critchley, and Garfinkel (2018) established that interoceptive dysfunction appears across many conditions, with the insula as the central hub (Quadt, Critchley, and Garfinkel 2018).
Partial evidence exists from fibromyalgia, a condition that frequently overlaps with ME/CFS: Valenzuela-Moguillansky et al. (2017) found that fibromyalgia patients show elevated body attention alongside decreased accuracy on body-localization tasks—the same dissociation pattern, but measured through touch rather than internal body signals (Valenzuela-Moguillansky, Reyes-Reyes, and Gaete 2017). (Origin: brainstorm)
(Certainty: 0.45. The Garfinkel tripartite framework is well-validated in healthy and clinical populations, and the fibromyalgia data provides indirect support, but no study has directly measured this dissociation in ME/CFS using the standard heartbeat detection protocol combined with the TAS-20 alexithymia scale and the MAIA sensibility scale. Falsification condition: If ME/CFS patients show normal or elevated accuracy alongside elevated sensibility on the heartbeat task—meaning they really are hyper-accurate, not just hyper-aware—then the corrupted-signal model is wrong. If both accuracy and sensibility are normal, the entire interoceptive framework lacks psychophysical evidence in ME/CFS. Note: This is one of the most immediately testable predictions in the paper—a single-session study could resolve it.)
Danciut, Garfinkel et al. (2024, n=33 MS patients + 33 controls) provided the first empirical demonstration that interoceptive deficits correlate with fatigue severity in a neuroinflammatory disease (Danciut, Garfinkel, et al. 2024). White matter dysconnectivity in interoceptive network tracts correlated with both interoceptive accuracy deficits and fatigue severity. MS provides a cross-disease template: neuroinflammation → white matter damage in interoceptive pathways → interoceptive accuracy deficit → fatigue. The Manjaly, Harrison, Critchley et al. (2019) review proposed the same pathway for MS fatigue specifically, linking interoceptive prediction error (via anterior cingulate and insula) to pathological fatigue (Manjaly et al. 2019).
Greenhouse-Tucknott et al. (2022) proposed a model that explains fatigue not as a direct readout of how tired your muscles are, but as a prediction the brain makes about how much effort an action will cost (Greenhouse-Tucknott et al. 2022). The brain combines real body signals (muscle state, metabolic byproducts, autonomic activity) with its own expectations based on past experience. In ME/CFS, the brain has accumulated a long history of evidence that even small actions produce disproportionate suffering. The result: it learns to overestimate how effortful any action will be, producing the experience of fatigue before the body has actually reached its physical limit.
The same framework has been applied to multiple sclerosis fatigue by Manjaly, Harrison, Critchley et al. (2019), who proposed that MS fatigue comes from disrupted body-signal processing in the anterior cingulate and insula—the same network implicated in ME/CFS (Manjaly et al. 2019). Two neuroinflammatory conditions, two independent research groups, one computational model.
The model makes specific predictions: (1) ME/CFS patients should overestimate how hard an upcoming task will feel, above and beyond what their measured exercise capacity would predict; (2) this overestimation should track inflammatory markers and fatigue severity, but not depression—distinguishing it from simple loss of motivation; (3) real-time heart rate variability biofeedback should help recalibrate these expectations more in patients with high body awareness than those with low; (4) treatments that reduce inflammation should reduce perceived effort for the same workload.
Seth (2011) added a deeper layer: the same brain system that produces fatigue predictions also produces the basic sense of being present in your body (Seth, Suzuki, and Critchley 2011). When the mismatch between prediction and reality is chronic and unresolvable, the result is not just fatigue but a feeling of disconnection—the body feels alien, uncontrollable, a source of threat rather than a home. This matches the anhedonia and depersonalisation that many ME/CFS patients describe. (Origin: brainstorm)
(Certainty: 0.40. (0.35→0.40: reinforced by cross-disease convergence with Manjaly/Critchley 2019 MS fatigue model, cert 0.75 (Manjaly et al. 2019)). The model is theoretically well-developed but has not been tested in ME/CFS. The MS parallel shows the same computational framework applied to a related neuroinflammatory condition. The model’s main strength is that it makes specific, falsifiable predictions. Falsification condition: If perceived effort in ME/CFS is explained entirely by measured exercise capacity—with no additional contribution from inflammatory markers or HRV—then the “learned expectation” layer is unnecessary and the simpler explanation (peripheral limitation only) is correct.)
8 Emotional-Interoceptive Dysfunction as Downstream Biology
The emotional processing abnormalities documented in ME/CFS—alexithymia, flattened affect, emotional-autonomic dissociation—are primarily biologically driven: downstream consequences of the same immune-to-brain signaling that produces the sickness behavior program, refracted through the computational lens of interoceptive predictive processing. Psychological factors (emotional suppression, negative affect amplification) can exacerbate these symptoms but do not cause them — the same bidirectional relationship that applies to fatigue (pain worsens mood; low mood amplifies pain perception without causing the underlying pathology). The interoceptive hierarchy framework is also explored in Section Speculative Cross-Disease Connections at lower certainty (0.30–0.40) as speculative extensions to cross-disease neuropsychiatric phenotypes; this section provides the mechanistic evidence base for those speculations.
When you have the flu, you don’t just feel tired—you lose interest in food, withdraw from social contact, feel mentally slow, and everything hurts more than it should. This coordinated package of symptoms is called sickness behavior, and it’s produced by the same immune-to-brain signaling that causes fatigue. Normally it resolves when the infection clears. In ME/CFS, this program appears to get stuck in the “on” position.
The emotional symptoms that clinicians often label as separate problems—difficulty identifying feelings (alexithymia, documented in CFS by Maroti et al. 2017 (Maroti, Molander, and Bileviciute-Ljungar 2017)), a gap between what the body shows and what the person feels (emotional-autonomic dissociation, found experimentally by Rimes et al. 2016 (Rimes et al. 2016)), and the way negative mood amplifies physical symptoms more in CFS than in healthy people (Van Den Houte et al. 2017 (Van Den Houte et al. 2017))—can all be understood as different expressions of the same underlying problem (Dantzer et al. 2008) (Dantzer 2008).
Barrett’s Theory of Constructed Emotion provides a framework for understanding why (Barrett 2025). Emotions are not hardwired circuits that just “turn on”—they are predictions the brain constructs by combining body signals with past experience. If the body signals arriving at the brain are corrupted by inflammation (at the vagus nerve and the leaky spots in the blood-brain barrier), the brain’s emotion predictions become systematically unreliable. What gets labeled “alexithymia” may simply be the brain trying to build an emotion from degraded raw materials. (Origin: brainstorm)
(Certainty: 0.45. The immune→sickness behavior pathway is well-established. Reframing emotional symptoms as part of this same pathway—rather than as separate psychological problems—extends the model but has not been directly tested as a unified framework in ME/CFS. The Barrett theory provides plausibility from general neuroscience. Falsification condition: If alexithymia scores in ME/CFS do not track inflammatory markers, or if treatments that reduce inflammation and fatigue leave alexithymia unchanged, then the emotional symptoms are not primarily driven by the immune-to-brain pathway. If emotional abnormalities in ME/CFS are indistinguishable in severity from those in matched depression patients without ME/CFS, the disease-specific pathway is not supported.)
Alexithymia—difficulty identifying and naming your own feelings—is elevated in ME/CFS, but not in the way you’d expect from a personality trait. Maroti et al. (2017, n=43 CFS, 34 exhaustion syndrome, 43 controls) found that CFS patients scored high specifically on the “Difficulty Identifying Feelings” part of the standard questionnaire, while scoring normally on “Externally-Oriented Thinking” (a cognitive style of not paying attention to emotions) (Maroti, Molander, and Bileviciute-Ljungar 2017). This pattern—trouble naming feelings but normal attention to emotions—suggests the problem is not psychological avoidance but a breakdown in the raw data the brain needs to construct emotions.
Think of it this way: to know you’re anxious, your brain needs reliable signals about your heart rate, gut sensations, muscle tension, and breathing. If those signals are noisy or degraded—corrupted by inflammation at the vagus nerve and brainstem—the brain cannot assemble them into a clear emotion label. Barrett’s Theory of Constructed Emotion predicts exactly this: emotions are built from body signals, and when the signals are degraded, the construction fails (Barrett 2025).
Bileviciute-Ljungar and Friberg (2020, n=30 ME/CFS) found that lower emotional awareness correlated with more frequent awakenings on overnight sleep recordings (Bileviciute-Ljungar and Friberg 2020). This creates a vicious cycle: immune activation disrupts sleep, poor sleep further degrades body-sensing, and degraded body-sensing makes emotions even harder to identify.
Importantly, some patients may have always had difficulty with emotions (trait alexithymia) while others developed it only after getting sick (acquired alexithymia). Both can coexist in the same population—the interoceptive model predicts the second mechanism without ruling out the first. (Origin: brainstorm)
(Certainty: 0.45. Alexithymia elevation in CFS has been found in two studies (Maroti 2017, Bileviciute-Ljungar 2020), but both are small-to-moderate samples from overlapping research groups. The link to body-sensing has not been directly tested: no study has measured both heartbeat detection accuracy and alexithymia in the same ME/CFS patients. Falsification condition: If alexithymia scores do NOT correlate with heartbeat detection accuracy in ME/CFS, then alexithymia is not a body-sensing deficit—it must be explained by something else (stress hormones, pre-existing psychology, or other mechanisms). A stronger test: alexithymia should correlate with heartbeat detection accuracy in patients who report that their emotional difficulties started after ME/CFS onset, but not in those who say they always had them.)
In healthy people, emotional experience and the body’s physical response are normally in sync: when you feel distressed, your heart rate rises, your palms sweat, your muscles tense. In ME/CFS, this coupling can break.
Rimes et al. (2016, n=40 CFS, 40 controls) asked participants to suppress their emotions while watching upsetting material. Healthy controls showed the expected pattern: trying to suppress emotions produced a clear spike in skin conductance (a measure of sympathetic nervous system activity). CFS patients showed the opposite: their skin conductance stayed flat—their bodies didn’t react—but they reported feeling more distressed, not less (Rimes et al. 2016). The subjective experience of emotion had become detached from the body’s autonomic response.
A small fMRI study by Wortinger et al. (2017, n=10 adolescent CFS, 10 controls) found that CFS patients activate different brain circuitry during emotional tasks: the anterior cingulate and prefrontal cortex worked differently, even though behavioral performance was normal (Wortinger et al. 2017). This suggests the brain is recruiting extra resources to compensate for degraded processing at lower levels—the mental equivalent of straining to hear someone in a noisy room.
This dissociation has a direct clinical consequence: patients report emotional distress, but clinicians see normal vital signs (blood pressure, heart rate, temperature) and may dismiss the distress as exaggerated or fabricated. The absence of visible body signs is not evidence that the distress isn’t real—it is evidence that the normal body substrate of emotion has been disrupted. Van Den Houte et al. (2017) showed the same phenomenon from the other direction: experimentally inducing a bad mood amplified physical symptoms more in CFS and fibromyalgia patients than in controls (Van Den Houte et al. 2017). The amplification is real—mood genuinely affects symptoms—but it reflects a sensitised circuit, not a made-up problem.
Emotional processing abnormalities in CFS have now been found across different methods: experimental tasks (Rimes 2016), questionnaires (Brooks et al. 2017, n=67 CFS (Brooks, Chalder, and Rimes 2017)), and brain imaging (Wortinger 2017). A competing explanation focuses on stress hormones: Kang et al. (2026) reviewed evidence that HPA axis dysfunction in ME/CFS—cortisol dysregulation leading to hippocampal changes—could independently produce cognitive and emotional symptoms (Kang et al. 2026). The two pathways (body-sensing disruption and hormone disruption) are not rivals—both likely operate and probably interact.
(Certainty: 0.40. Each study is small-to-moderate, single-center, and unreplicated. The convergence across different methods provides triangulation but not independent replication. Confirmation in larger adult ME/CFS samples using combined autonomic recording, brain imaging, and emotional tasks is needed. Falsification condition: If the correlation between body response and self-reported emotion during emotional tasks—the coherence between what the body shows and what the person feels—is the same in CFS patients as in controls, then the dissociation is not specific to the disease and may reflect general distress rather than disrupted body-sensing.)
Clinical implication. These findings support the NICE NG206 (2021) position that emotional symptoms in ME/CFS should not be treated as primary evidence of psychological causation (National Institute for Health and Care Excellence 2021). Alexithymia screening should prompt evaluation for underlying immune-autonomic dysfunction alongside — not instead of — psychological assessment when clinically indicated. Emotional processing difficulties are biologically real: their primary treatment target is the upstream biology, but this does not mean patients should be denied access to evidence-based psychological support for the distress these symptoms cause. The analogy: no one would treat myositis fatigue with activity scheduling alone without investigating muscle inflammation — but offering the patient psychological support for coping with chronic illness is appropriate and empirical. The same principle applies to emotional symptoms in ME/CFS. Trait alexithymia (premorbid) and acquired alexithymia (inflammation-driven) may coexist within the patient population — the interoceptive model predicts the latter mechanism without excluding the former.