Computational Psychiatry of Fatigue: Towards predicting individual symptoms from cognition and brain connectivity
OPEN ACCESS
Loading...
Author / Producer
Date
2025
Publication Type
Doctoral Thesis
ETH Bibliography
yes
Citations
Altmetric
OPEN ACCESS
Data
Rights / License
Abstract
Fatigue is a pervasive symptom in medicine, associated with a wide variety of disorders and profoundly impacting patients' quality of life. Yet, there are no biomarkers for fatigue, and its pathophysiology remains poorly understood. Thus, diagnosis is predicated on patients' subjective accounts and based on excluding so-called secondary causes of fatigue. Beyond this, treatment mostly relies on a trial-and-error procedure, with limited efficacy.
The current evidence suggests a heterogeneous pathophysiological basis for fatigue. Several potential mechanisms have been suggested, that can be expressed in combination in the same patient. These include: (1) the presence of structural brain lesions in key areas such as monoamine nuclei or the lateral hypothalamus, as well as (2) the presence of peripheral or central inflammation. In addition, (3) a novel computational theory, the allostatic self-efficacy (ASE) theory, focused on the role of interoception and metacognition, suggests that fatigue arises from the metacognitive diagnosis of lost control over bodily states in a context of chronic dyshomeostasis.
The first two chapters of this thesis are dedicated to investigating these different hypotheses using two studies: a retrospective study conducted on data from the UK Biobank (n>2000) and a cross-sectional study conducted in people with MS (n=75). In both studies, we used supervised learning with nested cross-validation and elastic net regularization to predict either the presence of fatigue symptoms (classification) or fatigue severity (regression). Model performance as well as feature importance were evaluated, in order to assess which features most significantly contributed to prediction. Our results underscore the role of clinical, particularly sleep-related information, in the prediction of fatigue. Cognitive factors are found to significantly correlate with fatigue levels, in accordance with the ASE theory, although they do not lead, by themselves, to statistically significant performance on unseen data. Finally, imaging features were found to not significantly improve prediction when combined with clinical information and, by themselves, they did not yield a consistent and statistically significant association with fatigue levels, nor significant performance on unseen data.
Following this, we dive into more methodological work. The third chapter of this thesis contains a proposal for an extension to regression Dynamic Causal Modeling (rDCM) that allows for the estimation of time-varying connectivity on whole-brain networks. To this effect, a sliding window with clustering via Variational Bayesian Gaussian Mixture Modeling (VBGMM) is proposed and tested with simulated data from networks of increasing complexity. We show the feasibility of estimating time-varying whole-brain effective connectivity from simulated functional magnetic resonance imaging (fMRI) data using the proposed framework. Future work will focus on testing this method on empirical data.
Finally, after focusing on fMRI connectivity analyses, we turn to electrophysiological data, which offers less spatial resolution, but a higher temporal resolution. Models of effective connectivity also exist for this type of data and we review one of the most advanced formulations of DCM: conductance-based DCM for cross-spectral densities.
Taken together, our work highlights not only which predictors are most robustly associated with fatigue, but we also observe that several of these features are easy and inexpensive to obtain, making them more likely to be used in the clinical context. Our methodological work addresses potential avenues for further testing of the role of connectivity-derived information.
Ultimately, this thesis aims to help further our understanding of fatigue, and hopes to contribute to progress in both research and clinical care.
Permanent link
Publication status
published
External links
Editor
Contributors
Examiner : Stephan, Klaas Enno
Examiner : Pessiglione, Mathias
Book title
Journal / series
Volume
Pages / Article No.
Publisher
ETH Zurich
Event
Edition / version
Methods
Software
Geographic location
Date collected
Date created
Subject
Computational Psychiatry; Fatigue; fMRI; DCM; UK Biobank; EEG
Organisational unit
03955 - Stephan, Klaas E. / Stephan, Klaas E.
02140 - Dep. Inf.technologie und Elektrotechnik / Dep. of Inform.Technol. Electrical Eng.
Notes
Funding
Related publications and datasets
References: