In recent years, the widespread adoption of the Electronic Health Record and the TreC+ platform has made large amounts of data on the use of digital health services available. In this context, this thesis investigates access to the TreC+ platform in the Autonomous Province of Trento with a twofold objective: to describe the main temporal, demographic, and territorial patterns of platform use and to identify potential dependency structures among municipalities that cannot be explained solely by existing administrative divisions. The analysis combines both aggregated data and individual-level access records, providing a comprehensive description of the phenomenon from both an overall and an individual perspective. The systematic component of platform access is modelled using Generalized Additive Models (GAMs), which allow for a flexible representation of the effects of temporal dynamics and demographic and territorial characteristics. The residual component is subsequently analysed using a network-based approach based on the Graphical Lasso to estimate conditional dependencies among municipal time series. The resulting network is then analysed using the Louvain algorithm to identify groups of municipalities exhibiting similar usage patterns. The results reveal substantial territorial heterogeneity in platform use. The estimated residual network exhibits a sparse but non-random structure, indicating the presence of associations among municipalities that persist after accounting for the systematic components of the model. The identified partitions do not fully coincide with the existing Valley Communities, suggesting that geographical proximity and administrative organization alone are insufficient to explain the observed patterns. Furthermore, using these partitions as the grouping structure in predictive models leads to improved predictive performance compared with both a model assuming a common temporal pattern and a model based on the Valley Communities. Overall, this thesis shows that integrating Generalized Additive Models with network analysis not only provides an effective framework for describing patterns in the use of digital health services but also enables the identification of latent territorial structures that are valuable both for interpreting the phenomenon and for improving predictive performance. The proposed approach highlights the potential of combining flexible statistical models with network methods to support the monitoring and planning of digital health services.
Negli ultimi anni la crescente diffusione del Fascicolo Sanitario Elettronico e della piattaforma TreC+ ha reso disponibili grandi quantità di dati sull’utilizzo dei servizi sanitari digitali. In questo contesto, la presente tesi analizza gli accessi alla piattaforma TreC+ nella Provincia Autonoma di Trento con un duplice obiettivo: descrivere le principali dinamiche temporali, demografiche e territoriali del fenomeno e individuare eventuali strutture di dipendenza tra i comuni non spiegabili attraverso le sole suddivisioni amministrative. L’analisi utilizza sia dati aggregati sia dati a livello di singolo accesso, consentendo di descrivere il fenomeno sia nel suo andamento complessivo sia nelle caratteristiche degli accessi individuali. La componente sistematica degli accessi viene modellata mediante modelli additivi generalizzati (GAM), che permettono di rappresentare in modo flessibile gli effetti della dinamica temporale e delle caratteristiche demografiche e territoriali. Le componenti residue vengono quindi analizzate attraverso un approccio di rete basato sul Graphical Lasso, utilizzato per stimare le dipendenze condizionate tra le serie comunali. La rete risultante è successivamente analizzata mediante l’algoritmo di Louvain, al fine di individuare gruppi di comuni caratterizzati da pattern di utilizzo simili. I risultati evidenziano una marcata eterogeneità territoriale nell’utilizzo della piattaforma. La rete stimata sulle componenti residue presenta una struttura sparsa ma non casuale, indicando la presenza di associazioni tra comuni che persistono anche dopo aver tenuto conto delle componenti sistematiche. Le partizioni individuate non coincidono pienamente con le Comunità di Valle, suggerendo che la prossimità geografica e l’organizzazione amministrativa non siano sufficienti a descrivere le dinamiche osservate. Inoltre, il loro utilizzo come struttura di raggruppamento nei modelli previsivi consente di migliorare le prestazioni rispetto sia a un modello con andamento comune sia a un modello basato sulle Comunità di Valle. In conclusione, il lavoro mostra come l’integrazione tra modelli additivi generalizzati e analisi di rete consenta non solo di descrivere efficacemente i pattern di utilizzo dei servizi sanitari digitali, ma anche di individuare strutture territoriali latenti utili sia per interpretare il fenomeno sia per migliorare le prestazioni previsive. L’approccio proposto evidenzia il potenziale dell’analisi congiunta di modelli flessibili e reti statistiche per supportare il monitoraggio e la pianificazione dei servizi sanitari digitali.
Analisi delle strutture di dipendenza latenti e previsione degli accessi al Fascicolo Sanitario Elettronico della Provincia di Trento
BALLARIN, CHIARA
2025/2026
Abstract
In recent years, the widespread adoption of the Electronic Health Record and the TreC+ platform has made large amounts of data on the use of digital health services available. In this context, this thesis investigates access to the TreC+ platform in the Autonomous Province of Trento with a twofold objective: to describe the main temporal, demographic, and territorial patterns of platform use and to identify potential dependency structures among municipalities that cannot be explained solely by existing administrative divisions. The analysis combines both aggregated data and individual-level access records, providing a comprehensive description of the phenomenon from both an overall and an individual perspective. The systematic component of platform access is modelled using Generalized Additive Models (GAMs), which allow for a flexible representation of the effects of temporal dynamics and demographic and territorial characteristics. The residual component is subsequently analysed using a network-based approach based on the Graphical Lasso to estimate conditional dependencies among municipal time series. The resulting network is then analysed using the Louvain algorithm to identify groups of municipalities exhibiting similar usage patterns. The results reveal substantial territorial heterogeneity in platform use. The estimated residual network exhibits a sparse but non-random structure, indicating the presence of associations among municipalities that persist after accounting for the systematic components of the model. The identified partitions do not fully coincide with the existing Valley Communities, suggesting that geographical proximity and administrative organization alone are insufficient to explain the observed patterns. Furthermore, using these partitions as the grouping structure in predictive models leads to improved predictive performance compared with both a model assuming a common temporal pattern and a model based on the Valley Communities. Overall, this thesis shows that integrating Generalized Additive Models with network analysis not only provides an effective framework for describing patterns in the use of digital health services but also enables the identification of latent territorial structures that are valuable both for interpreting the phenomenon and for improving predictive performance. The proposed approach highlights the potential of combining flexible statistical models with network methods to support the monitoring and planning of digital health services.| File | Dimensione | Formato | |
|---|---|---|---|
|
Ballarin_Chiara.pdf
accesso aperto
Dimensione
6.07 MB
Formato
Adobe PDF
|
6.07 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/112236