Hampel’s LMS in the Analysis of Online Monitoring Data

Abstract

We present procedures for online signal extraction from intensive care data. These filtering methods use high-breakdown linear regression methods in moving time windows. In particular, we concentrate on the performance of Least Median of Squares (short: LMS) regression in this context. Comparing the LMS to other robust regression techniques, we discuss its merits for preserving clinically relevant patterns such as trends, abrupt shifts and extremes and for the removal of irrelevant spikes or outliers.