Bayesian Models for Heterogeneous Personalized Health Data

TL;DR

This paper introduces hierarchical Graph-Coupled Hidden Markov Models to track disease spread in small communities, capturing individual differences and covariates, with applications demonstrated on simulated and real data.

Contribution

It develops a novel Bayesian hierarchical HMM framework incorporating link priors and new approximate distributions for better modeling of heterogeneous health data.

Findings

Accurately predicts individual infection probabilities.

Infers personal vulnerability and covariate effects.

Demonstrates effectiveness on real epidemiological data.

Abstract

The purpose of this study is to leverage modern technology (such as mobile or web apps in Beckman et al. (2014)) to enrich epidemiology data and infer the transmission of disease. Homogeneity related research on population level has been intensively studied in previous work. In contrast, we develop hierarchical Graph-Coupled Hidden Markov Models (hGCHMMs) to simultaneously track the spread of infection in a small cell phone community and capture person-specific infection parameters by leveraging a link prior that incorporates additional covariates. We also reexamine the model evolution of the hGCHMM from simple HMMs and LDA, elucidating additional flexibility and interpretability. Due to the non-conjugacy of sparsely coupled HMMs, we design a new approximate distribution, allowing our approach to be more applicable to other application areas. Additionally, we investigate two common link functions, the beta-exponential prior and sigmoid function, both of which allow the development of a principled Bayesian hierarchical framework for disease transmission. The results of our model allow us to predict the probability of infection for each person on each day, and also to infer personal physical vulnerability and the relevant association with covariates. We demonstrate our approach experimentally on both simulation data and real epidemiological records.