The use of multivariate autoregressive modelling for analyzing dynamical physiological responses of individual critically ill patients

Authors:
Kristien Van Loon;Jean-Marie Aerts;Geert Meyfroidt;Greta Van den Berghe;Daniel Berckmans
Affiliations:
Division Measure, Model & Manage Bioresponses, Katholieke Universiteit Leuven, Leuven, Belgium;Division Measure, Model & Manage Bioresponses, Katholieke Universiteit Leuven, Leuven, Belgium;Department of Intensive Care Medicine, University Hospital Gasthuisberg, Leuven, Belgium;Department of Intensive Care Medicine, University Hospital Gasthuisberg, Leuven, Belgium;Division Measure, Model & Manage Bioresponses, Katholieke Universiteit Leuven, Leuven, Belgium
Venue:
ISBMDA'06 Proceedings of the 7th international conference on Biological and Medical Data Analysis
Year:
2006

Citing 2
Cited 0

Time Series Analysis: Forecasting and Control

Time Series Analysis: Forecasting and Control
Using Bayesian networks in the construction of a bi-level multi-classifier. A case study using intensive care unit patients data

Artificial Intelligence in Medicine

Quantified Score

Hi-index	0.00

Visualization

Abstract

We attempted to find a way to distinguish survivors and non-survivors on the basis of the differences in the dynamics in both patient classes using multivariate autoregressive (MAR) time series analysis techniques. Time series data of 11 physiological variables were used to calculate MAR models. Data were taken from a subset of patients, with an intensive care unit length of stay of at least 20 days, from a database of a previously published randomized controlled trial [1]. The methodology was developed on 20 and validated on 16 patients. Based on the MAR coefficients, impulse response curves were simulated to describe the contributions of a single variable to fluctuations in another. The impulse responses of non-survivors had a tendency to be either more instable or to return to the initial level after a longer time than the responses of survivors did. This allowed us to distinguish survivors from non-survivors in the development cohort with a sensitivity of 0.70 and a selectivity of 1.00. This result was confirmed in the validation set where a sensitivity of 0.63 and a selectivity of 1.00 were reached.