ETD

Archivio digitale delle tesi discusse presso l'Università di Pisa

Tesi etd-04092018-144359


Tipo di tesi
Tesi di laurea magistrale
Autore
PANICACCI, SILVIA
URN
etd-04092018-144359
Titolo
Population Health Management exploiting Machine Learning to identify high-risk patients
Dipartimento
INGEGNERIA DELL'INFORMAZIONE
Corso di studi
EMBEDDED COMPUTING SYSTEMS
Relatori
relatore Fanucci, Luca
relatore Francesconi, Paolo
relatore Profili, Francesco
Parole chiave
  • population health management
  • machine learning
  • decision support system
Data inizio appello
07/05/2018
Consultabilità
Completa
Riassunto
Population aging and the increase of chronic conditions incidence and prevalence produce a higher risk of hospitalization or death. This is particularly high for patients with multimorbidity, who usually receive ineffective and inefficient treatments, leading to a great consumption of resources. Identifying as soon as possible high-risk patients becomes an important challenge to improve health care service provision and to reduce costs. Nowadays, population health management, intended as the risk assessment process for de ning patients cohorts and stratifying members by the risk of preventable hospitalizations or death, can be used to identify these "complex" patients. Thanks to the growing computational power, it can exploit machine learning algorithms, intelligent models which learn from experience and have the capability of analysing huge amount of data. The aim of this study is to validate machine learning algorithms (Naive Bayes, CART, C5.0, Conditional Inference Tree, Random Forest, Arti cial Neural Network, LASSO and some combinations of them) to predict the risk of hospitalization or death and to identify high-risk patients, starting from administrative and/or socio-economic data on the residents in the Local Health Unit of Central Tuscany. The high sensitivity, always greater than 70%, points out that a great part of the target patients is correctly identi ed by the models, leading to the improving of the quality of their lives, and that not many chronic patients having really need of speci c treatments are misclassi ed. The models, at last, have a very good discriminatory power, outperforming the methods currently used in Tuscany for the identi cation of high-risk patients (6 vs 40 for the Positive Predictive Ratio metric, which underlines the goodness of the new classi ers). Because of these considerations, the models could be really employed for the selection of the complex patients, to support the general practitioners' decisions and to reduce the risk of hospitalization or death.
File