Quantitative investigation of the impact of noise on heart rate variability and complexity analysis in trauma patients

Abstract

Objectives: Despite extensive studies in the past few decades on the significance and meaning of the many different measures of heart rate variability (HRV) and heart rate complexity (HRC), few studies exist to assess the influence of noise on these measures. The aims of this study were to review selected measures of HRV and HRC and investigate how noise affects these calculations, particularly in trauma patients who received at least 1 life-saving intervention (LSI) vs. those who received none. Methods: A total of 108 prehospital trauma patients with blunt or penetrating injuries admitted to a level I trauma center in Houston, TX, were selected based upon availability of electrocardiogram (ECG) waveform data and manual verification of all R-R interval (RRI) sequences. Eighty-two patients received at least 1 LSI, whereas the remaining 26 patients received none. Electrocardiograms were approximately 15 to 20 minutes long and acquired at a sampling frequency of 375 Hz. Noise was defined to be any source that alters the true RRI sequence of an ECG and, consequently, modifies calculation of an ECG-derived metric. Noise was both simulated and naturally obtained. Artificial modifications of RRI sequences corresponding to patient records were carried out by dropping every Mth multiple of the true RRI sequence for M= 10, 9, ., 1. Furthermore, ECGs of patient records were loaded into an in-house-developed R-wave detection algorithm to produce detected RRI sequences. Selected metrics were calculated afterwards, and means between LSI and non-LSI patient groups were then compared using Wilcoxon tests. The criterion standard for obtaining true means was manual verification of R waves and subsequent metric calculations. Results: There were no significant differences between LSI and nonLSI patients in age, sex, and transport time. 82 patients underwent 142 LSIs. Mean age was 37±14, 76% were male, 86% had blunt injury, and injury severity score was 17 ± 11. Furthermore, numerical relationships between mean values of patient groups were preserved, regardless of amount of noise (LSI vs non-LSI; sample entropy: 0.8 ± 0.3, 1.1±0.4, P= .002; quadratic sample entropy: 3.0±0.9, 3.5± 0.7, P = .004; multiscale entropy: 0.5 ± 0.4, 1.0 ± 0.6, P = .001; Poincare variability ratio: 0.5 ± 0.2, 0.4 ± 0.2, P = .370; fractal scaling exponent: 1.0 ± 0.2, 0.9 ± 0.2, P = .101; autocorrelation coefficient: 0.1 ± 0.1, 0.1 ± 0.0, P = .004; degree of nonstationarity: 0.7 ± 0.2, 0.6 ± 0.2, P= .039; SD: 45.8 ± 60.6, 40.4 ± 25.7, P= .309; successive differences: 44.5 ± 70.3, 35.7 ± 31.3, P= .454). However, there were only statistical differences (P b .01) between groups for entropy measures and autocorrelation. Importantly, P values for sample and multiscale entropy were the lowest of all measures. Conclusions: This study demonstrated that under some noisy environments, selected measures of HRV and HRC can still discriminate between patient groups. In particular, sample and multiscale entropy have predictive power and promising clinical use for the trauma patient cohort.