Although significant progress has been made from a virological and scientific perspective in preventing and predicting the spread of epidemic diseases, over the past few decades many infectious diseases - from the common flu to HIV, COVID-19, and monkeypox - have continued to threaten global health. Moreover, these diseases pose not only a health issue but also have direct and indirect repercussions on both social and economic dimensions. As a result, recent years have seen a growing demand for more accurate tools and techniques to forecast disease spread and evaluate the effectiveness of pharmaceutical and non-pharmaceutical interventions. By developing mathematical models suited to the complexity of these challenges, the mathematical community has played a crucial role in understanding the biological and epidemiological mechanisms of infectious diseases and, more importantly, in comparing the impact of various containment strategies that policymakers must consider. Advancing in this direction is essential to highlight fundamental aspects of disease transmission mechanisms that remain partially unknown. The research conducted in the last few years and presented in this thesis aims to contribute to this effort by offering significant advances in developing innovative mathematical models and numerical methods to address unresolved issues, enabling faster and more effective responses to emerging diseases. This work presents original contributions aimed at addressing three critical problems frequently faced by the mathematical epidemiological community: (i) providing accurate and reliable forecasts and scenario analyses, (ii) controlling the spread of potential epidemic or pandemic events, and (iii) identifying unknown dynamics in the variables that influence disease transmission mechanisms. Specifically: 1. A novel approach based on ML techniques is proposed for approximating operators between functional spaces, based on Kernel regression theory. This approach has shown valuable performance for reconstructing the map that, given a function describing the temporal evolution of non-pharmaceutical interventions for a specific disease, returns the corresponding evolution of infected individuals, enabling reliable predictions in synthetic scenarios; 2. A new age-stratified compartmental epidemic model, specifically adapted to the spread of SARS-CoV-2, is introduced. Adopting this model as state problem, we solve optimal control problems through a computationally efficient numerical scheme, aiming to minimize the number of infections, deaths, and hospitalizations by optimizing the allocation of vaccine doses across different age groups; 3. We propose a novel neural-network-based architecture designed to automatically learn the differential dynamics governing the evolution of the transmission rate of a disease at stake. This parameter's evolution has been modeled depending on exogenous variables commonly associated with transmission, such as climatic and environmental factors.
Sebbene dal punto di vista virologico e scientifico siano stati fatti ragguardevoli passi in avanti nella prevenzione e nella previsione di diffusione di malattie epidemiche, negli ultimi decenni molte malattie infettive (partendo dalla più comune influenza fino all’HIV, al COVID19 e al vaiolo delle scimmie) hanno minacciato la salute globale. Inoltre, la presenza di tali malattie non costituisce solo una sfida sanitaria, ma ha anche ripercussioni significative, sia dirette che indirette, sul piano sociale ed economico. Di conseguenza, negli ultimi anni `e aumentata significativamente la necessità di sviluppare tecniche e strumenti più precisi, capaci di prevedere l’evoluzione della diffusione di queste malattie e di valutare l’efficacia degli interventi, sia farmaceutici che non farmaceutici, implementabili. Grazie alla formulazione di modelli matematici sempre più adeguati alla complessità di questi problemi, la comunità matematica ha fornito contributi essenziali per la comprensione dei meccanismi biologici ed epidemiologici delle malattie infettive, in particolare per l’analisi comparativa dell’impatto delle diverse strategie di contenimento che i governanti possono adottare. Continuare a progredire in questa direzione è fondamentale per far luce su aspetti cruciali del meccanismo di trasmissione delle malattie, che rimangono ancora in parte sconosciuti. La ricerca condotta negli ultimi anni e presentata in questa tesi si pone proprio questo obiettivo: offrire un contributo significativo allo sviluppo di modelli matematici e metodi numerici per affrontare le questioni ancora aperte, così da rendere le risposte a possibili nuove malattie emergenti sempre più tempestive ed efficaci. In particolare, in questo lavoro sono riportati contributi originali finalizzati a risolvere tre problemi cruciali che la comunità epidemiologica matematica affronta frequentemente: (i) fornire previsioni e analisi di scenario affidabili e robuste, (ii) controllare la diffusione di epidemie o pandemie emergenti, e (iii) identificare dinamiche incognite delle variabili che influenzano i meccanismi di trasmissione delle malattie infettive. Nello specifico, in questa tesi: 1. viene proposto un nuovo approccio basato su tecniche di apprendimento automatico per l’approssimazione di operatori tra spazi di funzioni, fondato sulla teoria dei metodi Kernel per problemi di regressione. Questo approccio consente di ricostruire la mappa che, data una funzione che descrive l’andamento temporale degli interventi non farmaceutici per una determinata malattia, restituisce il corrispondente andamento degli infetti, permettendo di realizzare previsioni affidabili in scenari simulati; 2. viene introdotto un nuovo modello epidemico compartimentale stratificato per età, specificamente adattato alla diffusione di SARS- CoV-2. Utilizzando questo modello come problema di stato, abbiamo risolto problemi di controllo ottimo formulando uno schema numerico computazionalmente efficiente, al fine di minimizzare il numero di infetti, deceduti e ospedalizzati attraverso la allocazione delle dosi vaccinali tra diverse fasce d’età; 3. viene presentata una nuova architettura basata su reti neurali, progettata per apprendere automaticamente la dinamica differenziale che regola l’evoluzione del tasso di trasmissione di una malattia. Difatti, questa quantità varia nel tempo in funzione di variabili esogene, da cui chiaramente dipende la trasmissione della malattia, come fattori climatici e ambientali.
Numerical methods for epidemic forecast and control
Ziarelli, Giovanni
2024/2025
Abstract
Although significant progress has been made from a virological and scientific perspective in preventing and predicting the spread of epidemic diseases, over the past few decades many infectious diseases - from the common flu to HIV, COVID-19, and monkeypox - have continued to threaten global health. Moreover, these diseases pose not only a health issue but also have direct and indirect repercussions on both social and economic dimensions. As a result, recent years have seen a growing demand for more accurate tools and techniques to forecast disease spread and evaluate the effectiveness of pharmaceutical and non-pharmaceutical interventions. By developing mathematical models suited to the complexity of these challenges, the mathematical community has played a crucial role in understanding the biological and epidemiological mechanisms of infectious diseases and, more importantly, in comparing the impact of various containment strategies that policymakers must consider. Advancing in this direction is essential to highlight fundamental aspects of disease transmission mechanisms that remain partially unknown. The research conducted in the last few years and presented in this thesis aims to contribute to this effort by offering significant advances in developing innovative mathematical models and numerical methods to address unresolved issues, enabling faster and more effective responses to emerging diseases. This work presents original contributions aimed at addressing three critical problems frequently faced by the mathematical epidemiological community: (i) providing accurate and reliable forecasts and scenario analyses, (ii) controlling the spread of potential epidemic or pandemic events, and (iii) identifying unknown dynamics in the variables that influence disease transmission mechanisms. Specifically: 1. A novel approach based on ML techniques is proposed for approximating operators between functional spaces, based on Kernel regression theory. This approach has shown valuable performance for reconstructing the map that, given a function describing the temporal evolution of non-pharmaceutical interventions for a specific disease, returns the corresponding evolution of infected individuals, enabling reliable predictions in synthetic scenarios; 2. A new age-stratified compartmental epidemic model, specifically adapted to the spread of SARS-CoV-2, is introduced. Adopting this model as state problem, we solve optimal control problems through a computationally efficient numerical scheme, aiming to minimize the number of infections, deaths, and hospitalizations by optimizing the allocation of vaccine doses across different age groups; 3. We propose a novel neural-network-based architecture designed to automatically learn the differential dynamics governing the evolution of the transmission rate of a disease at stake. This parameter's evolution has been modeled depending on exogenous variables commonly associated with transmission, such as climatic and environmental factors.File | Dimensione | Formato | |
---|---|---|---|
tesi.pdf
accessibile in internet per tutti
Dimensione
19.03 MB
Formato
Adobe PDF
|
19.03 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/232592