A stochastic model characterizing the occurrence of premature ventricular contractions (PVCs) is described and tested for consistency with clinical data. Single-lead EKG signals from 2 1 patients hospitalized in the CCU were recorded and the intervals between ectopic beats were stored in the computer for a subsequent time series analysis. The PVC occurrence could be adequately described by locally stationary renewal processes for 86% of the patients, and for 52% consistency with the more specific Poisson process was established. Adopting the Poisson process as a pertinent model, a statistically efficient procedure for the on-lnne detection of PVC rate changes was developed.
The proposed detection scheme performed without significant degradation for the much wider class of data consistent with nonspecific renewal processes. The detection procedure provides for the partition of the record of PVC occurrence into a sequence of segments with constant rates, thereby greatly reducing the amount of data for monitoring.