Premium
Multivariate Hidden Markov Models for disease progression
Author(s) -
Martino Andrea,
Guatteri Giuseppina,
Pagai Anna Maria
Publication year - 2020
Publication title -
statistical analysis and data mining: the asa data science journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.381
H-Index - 33
eISSN - 1932-1872
pISSN - 1932-1864
DOI - 10.1002/sam.11479
Subject(s) - multivariate statistics , unobservable , hidden markov model , disease , computer science , normality , outcome (game theory) , multivariate analysis , econometrics , machine learning , artificial intelligence , markov chain , markov model , statistics , medicine , mathematics , mathematical economics
Disease progression models are a powerful tool for understanding the development of a disease, given some clinical measurements obtained from longitudinal events related to a sample of patients. These models are able to give some insights about the disease progression through the analysis of patients histories and can be also used to predict the future course of the disease for an individual. In particular, Hidden Markov Models are suitable for disease progression since they model the latent unobservable states of the disease. In this work, we propose a HMM where the outcome is multivariate and its components are not independent; to accomplish our aim, since we do not make any usual normality assumptions, we model the outcome using copulas. We first test the performance of our model in a simulation setting and show the validity of the method. Then, we study the course of Heart Failure, applying our model to an administrative dataset from Lombardia Region in Italy, showing how episodes of hospitalization can give information about the disease status of a patient.