Biomarkers as individual indicators and interconnected systems: statistical approaches across clinical contexts

Biomarkers play a central role in modern medicine and biomedical research. They are widely used to characterize biological processes, monitor disease progression, assess disease risk, and evaluate safety and response to therapeutic interventions. Advances in laboratory technologies and high-throughput methods have substantially increased the number and diversity of available biomarkers, creating unprecedented opportunities to investigate complex biological systems. However, this growing complexity also poses substantial challenges. Biomarkers often vary over time, reflecting dynamic physiological and pathological processes, and they may interact with one another, mirroring the organization of biological systems as interconnected structures rather than isolated components. As these challenges become increasingly pronounced, advanced statistical methods are required to adequately capture temporal variability, interdependencies, and the underlying structure of biomarker systems. Traditionally, biomarkers have been studied as single entities—serving as predictors, outcomes, or covariates—with the primary aim of estimating their associations with clinical endpoints. This approach has proven successful over the past decades in identifying relevant indicators of biological and clinical processes. At the same time, however, such an approach may overlook important information that emerges when biomarkers are considered as components of an interconnected system rather than as isolated variables. The aim of this thesis is to investigate biomarker data from two different perspectives. The first perspective focuses on biomarkers considered as individual entities, examining both cross-sectional measurements and their longitudinal evolution over time. In this context, longitudinal analyses of Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) antibody trajectories are conducted to characterize the immunological response following infection, and risk analyses are performed to evaluate their association with post-coronavirus disease 2019 (COVID-19) syndrome. For these objectives, classical statistical methods are applied. The second part of the thesis moves beyond classical approaches by conceptualizing biomarkers as components of an interconnected system rather than as isolated variables. This perspective is developed within the field of aging research, with the aim of identifying underlying biological processes associated with aging. To this end, network-based statistical methods are employed to model interdependencies among biomarkers and to uncover higher-order structures that may reflect coordinated biological mechanisms. The thesis is structured into three main parts. Part I provides background on biomarkers, including their definitions, classifications, and principal clinical applications. Part II focuses on biomarker analyses in the context of post-COVID-19. Specifically, it investigates long-term predictors of post-COVID-19, with particular attention to immunological biomarkers, and examines the kinetics and duration of the immune response following SARS-CoV-2 infection. Part III introduces a network modelling framework applied to aging research. Chapter 3 presents the development of the MixMashNet framework, an R package designed to estimate, analyse, and interpret single layer and multilayer networks. This framework allows the identification of biomarker communities, central and bridging elements, and higher-level summaries that capture system-level organization beyond isolated effects. Chapter 4 applies this framework to circulating biomarker data from the Swedish National Study on Aging and Care in Kungsholmen (SNAC-K) cohort, with the aim of characterizing aging-related biological processes through a network-based perspective. The thesis concludes with a discussion of the main findings and their implications for biomarker research.

I biomarcatori svolgono un ruolo centrale nella medicina moderna e nella ricerca biomedica. Sono ampiamente utilizzati per caratterizzare i processi biologici, monitorare la progressione delle malattie, valutare il rischio e analizzare la sicurezza e la risposta agli interventi terapeutici. I progressi nelle tecnologie di laboratorio e nei metodi ad alta processività hanno aumentato in modo significativo il numero e la varietà dei biomarcatori disponibili, offrendo nuove opportunità per lo studio di sistemi biologici complessi. Tuttavia, questa crescente complessità comporta anche sfide rilevanti. I biomarcatori variano nel tempo e possono interagire tra loro, riflettendo sistemi biologici interconnessi piuttosto che componenti isolati. Diventano quindi necessari metodi statistici avanzati per cogliere variabilità temporale, interdipendenze e struttura dei sistemi di biomarcatori. Tradizionalmente, i biomarcatori sono stati studiati come entità singole—utilizzati come predittori, esiti o covariate—con l’obiettivo di stimare la loro associazione con endpoint clinici. Sebbene questo approccio si sia dimostrato efficace, può trascurare informazioni rilevanti che emergono considerando i biomarcatori come un sistema interconnesso. L’obiettivo di questa tesi è analizzare i biomarcatori da due prospettive. La prima li considera come entità individuali, analizzando misurazioni trasversali e traiettorie longitudinali nel tempo. In questo contesto, vengono studiate le traiettorie degli anticorpi contro il SARS-CoV-2 e la loro associazione con la sindrome post-COVID-19, utilizzando metodi statistici classici. La seconda prospettiva supera l’approccio tradizionale, concependo i biomarcatori come componenti di un sistema interconnesso. Questa analisi è sviluppata nell’ambito della ricerca sull’invecchiamento, utilizzando metodi statistici basati su reti per modellare le interdipendenze e identificare strutture di ordine superiore associate a processi biologici coordinati. La tesi è strutturata in tre parti. La Parte I introduce definizioni, classificazioni e applicazioni cliniche dei biomarcatori. La Parte II si concentra sul contesto post-COVID-19, analizzando predittori a lungo termine e la dinamica della risposta immunitaria. La Parte III presenta un framework di modellazione a rete applicato all’invecchiamento. Il Capitolo 3 introduce il framework MixMashNet, un pacchetto R per l’analisi di reti a singolo e multilivello, mentre il Capitolo 4 ne applica l’approccio ai dati dello studio SNAC-K. La tesi si conclude con una discussione dei principali risultati e delle loro implicazioni per la ricerca sui biomarcatori.