Considering the opportunities given by EHR data towards the development of AI algorithms within healthcare, and the potential impact that such approaches may have if successfully deployed to the real world after overcoming trust-related limitations, in this thesis, we propose three new specific frameworks to develop interpretable and easy-to-use AI-based tools to predict disease progression-related outcomes using data collected in EHR systems. We complete the description of each framework with a case study tied to a disease and its corresponding data flow obtained from EHR infrastructures. The first framework (Chapter 2) is designed to develop interpretable and easy-to-use risk scores. To provide an explicative case study, the proposed framework is applied to obtain a tool to predict hospitalizations caused by heart failure using the administrative claims of 176,018 patients with diabetes extracted from the EHR system of the Veneto Region, Italy. The second framework (Chapter 3), concerns the development of explainable deep learning-based natural language processing algorithms that extract relevant structured information from unstructured free-form text. The case study for this second framework concerns the development of a tool to identify a past hospitalization related to cardiovascular complications using the text of 197,411 routine visits undergone by 16,876 patients with diabetes followed at the diabetes outpatient clinic of the University Hospital of Padua. Finally, the third framework (Chapter 4) is designed as a solid methodological core to allow models to be fairly trained and evaluated while ensuring model interpretability and the possibility of deploying them to real-world clinical software. This framework is model-agnostic and can be applied without specific restrictions related to the underlying methodologic approach or domain of application. The case study for this last framework concerns the development and deployment of a tool useful to predict death due to amyotrophic lateral sclerosis using data from 2,209 patients collected during daily clinical practice by neurologists at two centers of excellence in Turin, Italy, and Lisbon, Portugal. Ultimately, this thesis demonstrates the potential of AI algorithms built by exploiting EHR data through proper development workflows that ensure solid model development and validation as well as result interpretability.

Interpretable Artificial Intelligence to Predict Disease Progression-Related Outcomes via Electronic Healthcare Records Data / Guazzo, Alessandro. - (2024 Mar 20).

Interpretable Artificial Intelligence to Predict Disease Progression-Related Outcomes via Electronic Healthcare Records Data

GUAZZO, ALESSANDRO
2024

Abstract

Considering the opportunities given by EHR data towards the development of AI algorithms within healthcare, and the potential impact that such approaches may have if successfully deployed to the real world after overcoming trust-related limitations, in this thesis, we propose three new specific frameworks to develop interpretable and easy-to-use AI-based tools to predict disease progression-related outcomes using data collected in EHR systems. We complete the description of each framework with a case study tied to a disease and its corresponding data flow obtained from EHR infrastructures. The first framework (Chapter 2) is designed to develop interpretable and easy-to-use risk scores. To provide an explicative case study, the proposed framework is applied to obtain a tool to predict hospitalizations caused by heart failure using the administrative claims of 176,018 patients with diabetes extracted from the EHR system of the Veneto Region, Italy. The second framework (Chapter 3), concerns the development of explainable deep learning-based natural language processing algorithms that extract relevant structured information from unstructured free-form text. The case study for this second framework concerns the development of a tool to identify a past hospitalization related to cardiovascular complications using the text of 197,411 routine visits undergone by 16,876 patients with diabetes followed at the diabetes outpatient clinic of the University Hospital of Padua. Finally, the third framework (Chapter 4) is designed as a solid methodological core to allow models to be fairly trained and evaluated while ensuring model interpretability and the possibility of deploying them to real-world clinical software. This framework is model-agnostic and can be applied without specific restrictions related to the underlying methodologic approach or domain of application. The case study for this last framework concerns the development and deployment of a tool useful to predict death due to amyotrophic lateral sclerosis using data from 2,209 patients collected during daily clinical practice by neurologists at two centers of excellence in Turin, Italy, and Lisbon, Portugal. Ultimately, this thesis demonstrates the potential of AI algorithms built by exploiting EHR data through proper development workflows that ensure solid model development and validation as well as result interpretability.
Interpretable Artificial Intelligence to Predict Disease Progression-Related Outcomes via Electronic Healthcare Records Data
20-mar-2024
Interpretable Artificial Intelligence to Predict Disease Progression-Related Outcomes via Electronic Healthcare Records Data / Guazzo, Alessandro. - (2024 Mar 20).
File in questo prodotto:
File Dimensione Formato  
Tesi_Alessandro_Guazzo.pdf

embargo fino al 20/03/2027

Descrizione: Tesi_Alessandro_Guazzo
Tipologia: Tesi di dottorato
Licenza: Altro
Dimensione 10.07 MB
Formato Adobe PDF
10.07 MB Adobe PDF Visualizza/Apri   Richiedi una copia
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11577/3511514
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
  • OpenAlex ND
social impact