Risk stratification of ICU patients using arterial blood pressure waveforms

Abstract

Identifying patients at high risk for adverse events is very important to the practice of clinical medicine. Non-invasive ECG-based methods of risk stratification such as T wave Alterans, Morphological Variability, and Heart Rate Variability extract prognostic information from the electrocardiograph. However, there is still a wealth of data collected from ICU patients and left unused every year that can augment risk-stratification methods. This thesis extends non-invasive risk stratification to Arterial Blood Pressure (ABP) Waveforms. We derive and analyze classifiers based on the morphological distance time series (derived from beat-to-beat morphology changes in the ABP waveform) including ASDNNmd, SDANNmd, rMSSDmd, the MVABP score etc. We also derive and analyze classifiers based on the Downstroke Time Series (derived from the decay from peak systole to diastole) including ASDNNDownstroke, SDANNDownstroke, rMSSDDownstroke, etc. While this body of work suggests the classifiers we developed are not effective in risk stratification of ICU patients, we discuss other methods which may extract prognostic information from the ABP waveform more effectively.