Nonstationary Multivariate Gaussian Processes for Electronic Health Records

TL;DR

This paper introduces a nonstationary multivariate Gaussian process model for electronic health records, capturing time-varying relationships among clinical variables to improve prediction and uncover latent correlations.

Contribution

The authors develop a novel nonstationary Gaussian process framework with time-dependent parameters and efficient inference methods, applied to EHR data for enhanced predictive accuracy.

Findings

Better predictive performance than stationary models

Uncovered latent correlation processes across vitals

Potentially predictive of patient risk

Abstract

We propose multivariate nonstationary Gaussian processes for jointly modeling multiple clinical variables, where the key parameters, length-scales, standard deviations and the correlations between the observed output, are all time dependent. We perform posterior inference via Hamiltonian Monte Carlo (HMC). We also provide methods for obtaining computationally efficient gradient-based maximum a posteriori (MAP) estimates. We validate our model on synthetic data as well as on electronic health records (EHR) data from Kaiser Permanente (KP). We show that the proposed model provides better predictive performance over a stationary model as well as uncovers interesting latent correlation processes across vitals which are potentially predictive of patient risk.