Operationalising synergy in multimorbidity and symptom networks

Open Access
Authors
Supervisors
Co-supervisors
Award date 27 November 2026
ISBN
  • 9789465348148
Number of pages 314
Organisations
  • Faculty of Science (FNWI) - Informatics Institute (IVI)
Abstract
Complex biomedical systems contain interactions that cannot always be captured by conventional pairwise analyses. Understanding these interactions requires methods capable of identifying information that emerges only when variables are considered jointly, known as synergy. This thesis develops and applies information-theoretic approaches to quantify, interpret, and scale the analysis of higher-order dependencies in multimorbidity and symptom networks.
Using Partial Information Decomposition (PID) and related measures, the thesis demonstrates how pairwise networks can overlook higher-order dependencies and introduces computationally efficient search algorithms for identifying synergistic variable combinations in high-dimensional data. These methods are applied across diverse biomedical datasets, including cardiovascular risk factors, metabolomics, ageing cohorts, and psychological assessments. Applications to multimorbidity show that symptoms provide unique information about functional limitations beyond signs and diseases, particularly among individuals with multiple chronic conditions. To improve interpretability, the Breadth–Uniformity–Synergy–Total (BUST) framework characterises synergistic relationships across multiple outcomes and population subgroups, revealing context-dependent combinations of health indicators.
The thesis further introduces embedding-based PID (ePID), extending information decomposition to larger networks by compressing multivariate context into a target-relevant representation. Applications to depressive symptom networks and psychometric data demonstrate how redundant and synergistic information can be distinguished at system scale.
Together, these contributions establish a computational framework for uncovering and interpreting higher-order dependencies in complex datasets, advancing our understanding of which combinations of variables provide meaningful information, for whom, and under what circumstances.
Document type PhD thesis
Language English
Downloads
Permalink to this page
cover
Back