A signal abnormality index for arterial blood pressure waveforms

Fidelity of the arterial blood pressure (ABP) waveform is often critical to computations that involve processing of the waveform. The ABP waveform may be used for cardiac output (CO) estimation, suppression of ECG-based false alarms, and tracking long term trends such as mean pressure. An abnormal ABP waveform may cause these applications to give undesirable results. In this paper, we present a signal abnormality index (SAI) algorithm that detects abnormal beats in ABP waveforms. The algorithm flags ABP beats by intelligently setting constraints on physiologic, noise/artifact, and beat-to-beat variation values. Compared to an expert annotator, the algorithm's sensitivity=1, specificity=0.91, positive predictive value=0.73, and negative predictive value=l. We also perform a sensitivity analysis to show SAI's robustness. Finally, we use SAI to identify clean ABP segments from our study population and show that CO estimation error is 30% less in the clean subset than in the entire population.