| JOURNAL HOME | CME HOME | THIS MONTH | PAST ISSUES | ETOC | COLLECTIONS |
| AUTHORS | REVIEWERS | EDITORIAL BOARD | FEEDBACK | RSS | HELP |
| ||||||||||||||
| ||||||||||||||
|
|
|
|
||||||||||||
|
||||||||||||||
*Department of Anaesthesiology and Intensive Care Medicine University Hospital Schleswig-Holstein hanss@anaesthesie.uni-kiel.de
Department of Experimental and Applied Physics Christian-Albrechts-University Kiel, Germany
To the Editor:
With interest we read the article by Laitio et al. (1) reporting a predictive value of heart rate variability (HRV) on the incidence of postoperative ischemic events. The authors claimed a significantly higher fractal scaling exponent difference (
1) between night and day in 12 patients with postoperative ischemic events compared with 16 patients without ischemia. We agree with the authors that perioperative HRV assessment may be a useful tool to detect patients at risk of ischemic events. However, there are some points on which we wish to comment.
First, the authors claim that their study was performed according to the standards of HRV measurement defined previously (2). However, the recommended frequency range for ECG sampling should be at least 250 to 500 Hz or more rapidly (2,3). A sampling rate of 128 Hz, as used by the authors, is not sufficient for HRV analysis for the following reasons. Typical variations of RR-intervals are <20 ms (4). With a sampling rate of 128 Hz, the accuracy of RR-interval estimation is limited to 8 ms. If this random contribution to RR-interval determination is considered, any subsequent spectral analysis is questionable. Especially, the estimation of scaling exponents, which is a delicate task anyhow, will be influenced by this imprecision. This is reflected by the large scattering of spectral power in all frequency domains as published by the authors (1). Therefore, the rather small night to day differences in the ischemic patients may mainly originate from undersampling and are not proof of differences in the autonomic regulation of the patients. Finally, HRV values are not thought to be normally distributed (2). Nevertheless, parametric tests were used for statistical analysis between groups. Given the considerable overlap, the differences are not that impressive, and it remains questionable whether nonparametric tests would have yielded significant results.
References
Department of Anesthesiology, Turku University Hospital, timo.laitio@tyks.fi Department of Theoretical Physics, Turku University, Turku, Finland Division of Cardiology, Department of Medicine, Oulu University Hospital, Oulu, Finland Department of Anesthesiology, Turku University Hospital Division of Cardiology, Department of Medicine, Oulu University Hospital, Oulu, Finland Department of Biostatistics, University of Turku Departments of Physiology and Clinical Physiology, Turku University Hospital Turku PET Centre and the Department of Pharmacology and Clinical Pharmacology, University of Turku, Turku, Finland
In Response:
We appreciate the comments on our study (1) of the predictive value of fractal heart rate variability (HRV) for postoperative prolonged ischemia by Hanss et al. They comment on the use of Holter recorder with 128 Hz sampling rate. This is a very important issue since most of the HRV studies before the year 2000 have been done with Holter recordings with the sampling rate of 128 Hz (2). The Holter recordings of our study were performed during the years 1995–1997. However, we cannot completely agree with Hanss et al.
First, they claim that especially the analysis of fractal scaling exponents will be influenced by the sampling rate. Second, they claim that the statistical analysis should have been performed with nonparametric methods because HRV values are not thought to be normally distributed.
Recently, the accuracy of RR interval detection with the sampling rate of 128 Hz (8-ms timing accuracy) and 1000 Hz (1-ms timing accuracy) was studied by Tapanainen et al. (2). Conventional and dynamic HRV measures were used. They showed that SDNN, a time domain measure, short-term (1) and long-term (2) fractal analysis, can be done adequately with both sampling rates. However, the high-frequency spectral components of conventional measures and the approximate entropy (ApEn) of dynamic measures were significantly affected by the sampling rate. The difference between the 1-ms and the 8-ms timing accuracies was 11 and 14% with ApEn and LF/HF ratio, respectively. The difference with the fractal measure 1 was only 3%. Thus, it is not likely that there would exist any significant difference in scaling exponent 1 whether analyzed with 128 Hz or with 250 Hz, or even 1000 Hz. Notably, the Task Force (3) has no exact recommendations for sampling rate in the analysis of HRV with nonlinear (i.e., dynamic) measures, since at that time such methods were rather new. Furthermore, the ability of scaling exponent 1 measured with 1000 Hz and with 128 Hz sampling rates to identify patients with earlier myocardial infarction (MI) from the healthy subjects was similar (2). However, ApEn was able to identify MI patients as measured with 128 Hz but not with 1000 Hz sampling rate. The main result of our study (1) was that the preoperative night-day difference of 1 predicted postoperative prolonged (>10 min) myocardial ischemia in traumatic hip fracture patients. Other HRV measures had no predictive value and ApEn was not used.
We agree with Hanss et al. that part of the conventional and dynamic analyses of HRV should be performed with a higher sampling rate than 128 Hz. However, it seems to be that the fractal scaling measures are not seriously biased by the lower sampling rate. Thus, we strongly believe that the predictive value of scaling exponent 1 will remain if measured with the higher timing accuracy of the Holter recorder. However, conventional Holter recordings should be used with caution in the in the case when frequency domain measures and some new dynamic measures are used, especially ApEn (2).
Hanss et al. also criticized the use of parametric tests in the statistical analyses. It is true that HRV values are not usually normally distributed. However, we did not use parametric methods to test the predictive value of the night-day difference of the HRV parameters. Instead, the logistic regression analyses were used that do not require the normal distribution of the data (4). The paired t-test was only used to analyze the difference between night and day values within group that had no predictive value and had, thus, no importance in our study. However, we also analyzed the difference between the day and night values within groups with a nonparametric test (i.e., Wilcoxon matched pairs test). The results did not change.
It can be concluded that the predictive value of preoperative HRV for postoperative MI and death should be studied more with various surgical patients and with larger patient populations. The potential benefit of HRV in risk stratification of high-risk patients is strongly supported by numerous studies done with various cardiac patients. Instead, only one study has been published to date, which shows that preoperative HRV can predict postoperative death (5). The Holter recordings of that study were also done with the sampling rate of 128 Hz. However, it is clear that high-resolution Holter recorders are more accurate than the low-resolution recorders and should be used in future studies in this field.
References
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
