An Improved Method for Using Sample Entropy to Reveal Medical Information in Data from Continuously Monitored Physiological Signals

Medical devices, especially wearables, are being under fast development for continuous monitoring of physiological signals. These devices generate a huge amount of continuous time series data. To derive meaningful and useful information out of these data, the adoption of nonlinear statistics is usually essential. Sample entropy is becoming a widely used nonlinear statistics to extract the information contained in continuous time series data for disease diagnosis and prognosis. However, missing values commonly exist in the physiological time series data. How to minimize the influence of missing points on the calculation of entropy remains an important problem in practice. In this paper, we propose a new method to handle missing values in this area. Unlike the usual ways by modifying the input data, such as direct deletion, our method keeps the data unchanged and modifies the calculation process, which employs a less intrusive way of dealing with missing values. Our research demonstrates that our method is effective and applicable to RR interval data in entropy analysis. Therefore, our proposed method may serve as an effective tool for dealing with missing values in the analysis of sample entropy for physiological signals.