Predictability of baroreflex sensitivity induced by phenylephrine injection via frequency domain indices computed from heart rate and systolic blood pressure signals during deep breathing
Introduction
It is known that baroreflex sensitivity (BRS) is the amount of change in beat-to-beat interval (RR) against a 1 mm Hg systolic blood pressure deviation [1], [2]. BRS is traditionally measured by bolus injection of vasoactive drugs such as phenylephrine which causes increase in systolic blood pressure (SBP). The increase (decrease) in SBP tempts an increase (decrease) in the corresponding RR. A linear regression is build between mutually varying RR and SBP segments right after the drug injection. The slope of the regression line is recognized as BRS [1]. Since it requires drug injection, this method is classified as a pharmacological method. In the rest of the manuscript this method will be referred to as p-BRS. The clinical importance of BRS has been emphasized by several researches and it is reported that BRS indicates increased risk of sudden cardiac death in myocardial infarction patients and can be used for diagnosis and management of the therapy [3], [4], [5], [6].
Alternative BRS assessment methods, which eliminate the use of drugs, have been introduced over the years. These techniques are classified as non-pharmacological methods. A brief summary of these methods can be found in [2]. Let us here shortly summarize recent findings. The researchers have usually studied the individual correlations between p-BRS and non-pharmacological methods. In [7] the correlation between p-BRS and the transfer function (TF) BRS assessment method was investigated. The authors reported a correlation level of 0.94 in eight subjects. In another study, this relation was investigated with a larger subject database and the correlation between p-BRS and TF has been reported as low as 0.55 [8]. The authors concluded that non-pharmacological methods should not be used in clinical practice as a replacement of p-BRS [8]. In [9] it is argued that p-BRS and non-pharmacological methods should be considered as complementary methods rather than alternative methods since they contain distinct information. Apart from these studies, Davies et al. indicate that BRS can be estimated from RR interval oscillation amplitude at about 0.1 Hz, which is achieved by controlled breathing at this frequency [10]. They report a correlation of 0.81 between the RR oscillation amplitude at 0.1 Hz and BRS index.
The traditional approaches generally investigated the predictive power of individual features. To our knowledge the efficiency of combination of indices in estimating p-BRS was not investigated. In this work, the prediction of p-BRS was achieved from a combination of non-pharmacological indices (predictor features). As a first step, a large pool of indices was extracted from electrocardiogram (ECG) and blood pressure (BP) data which were recorded during deep breathing. The non-pharmacological indices were computed in time and frequency domains using traditional techniques proposed in the past. In addition, a novel frequency domain feature was introduced to estimate the p-BRS values of the second phase. This feature was appended to the set of non-pharmacological indices. In order to eliminate model over fitting, a subset of these time and frequency domain indices were selected to predict the p-BRS. The prediction performance was estimated with a leave one subject out procedure.
The rest of the paper is organized as follows. In the next section, we detail our data acquisition system and experimental paradigm that recorded ECG and BP data. In the following section, we describe our predictors and explain how they are computed from time and frequency domain data. Finally, we provide experimental results and conclude.
Section snippets
Acquisition and pre-processing of electrocardiogram and blood pressure data
Electrocardiogram and BP signals were obtained from 21 subjects (10 male, 11 female) who underwent coronary angiography at Balcali Hospital of Çukurova University, Turkey. The ages of subjects ranged between 36 and 62 with a mean of 49.9 and a standard deviation of 6.5. All subjects had normal coronary arteries and provided informed consent for participating in the study. The study was approved by the local ethics committee. The ECG and BP data were acquired simultaneously with a custom system
Results
Two predictors provided the highest prediction accuracy of 81.6%. The prediction accuracy versus the number of predictors is given in Fig. 3. The correlation between predicted BRS and p-BRS was 0.87 (p = 1.16 × 10−5). A scatter plot of the predicted BRS with p-BRS is shown in Fig. 4. During the leave one out steps the algorithm chose the same indices for all subjects consistently: normalized cross-power about Mayer frequency proposed by the authors and the average magnitude square coherence in HF
Conclusions
Diminished BRS is linked to severe cardiac problems in humans including increased risk of sudden cardiac death in myocardial infarction patients and ventricular arrhythmias [3], [4], [5], [6]. Owing to its value, measurement of BRS by internally forcing a variation in BP via drugs (p-BRS) is a widely accepted method in clinical practice. In this study, we showed that two spectral indices computed from SBP and RR time series during deep breathing can be used to predict the p-BRS with a
Acknowledgement
This study was supported by the Turkish Scientific and Research Council (TUBITAK). We thank Meryem Burduroğlu and Maggie Spaniol for proofreading the manuscript.
References (29)
- et al.
A noninvasive measure of baroreflex sensitivity without blood pressure measurement
Am. Heart J.
(2002) - et al.
Reflex regulation of arterial pressure during sleep in man: a quantitative method for assessing baroreflex sensitivity
Circ. Res.
(1969) - et al.
Advancements in estimating baroreflex function
IEEE Eng. Med. Biol. Mag.
(2001) - et al.
Baroreflex impairment and sudden cardiac death after myocardial infarction
Circulation
(1998) - et al.
Baroreflex sensitivity, clinical correlates, and cardiovascular mortality among patients with a first myocardial infraction. A prospective study
Circulation
(1988) - et al.
Baroreflex sensitivity as a new marker for risk stratification
Z. Kardiol.
(2000) - et al.
Autonomic Mechanisms and Sudden Cardiac Death. New insights from analysis of baroreceptor reflexes in conscious dogs with and without a myocardial infarction
Circulation
(1988) - et al.
Assesment of Baroreflex Sensitivity by means of spectral analysis
Hypertension
(1987) - et al.
Spontaneous indices are inconsistent with arterial baroreflex gain
Hypertension
(2003) - et al.
Time versus frequency domain techniques for assessing baroreflex sensitivity
J. Hypertens.
(2001)
Hemodynamic fluctuations and baroreflex sensitivity in humans: a beat-to-beat model
Am. J. Physiol. Heart Circ. Physiol.
Evaluation of baroreflex sensitivity by the sequence method using blood pressure oscillations and R–R interval changes during deep respiration
Eur. Neurol.
Determinants of heart rate variability during deep breathing: basic findings and clinical application
Clin. Auton. Res.
Controlled breathing protocols probe human autonomic cardiovascular rythms
Am. J. Physiol.
Cited by (4)
Linear and nonlinear analysis of normal and CAD-affected heart rate signals
2014, Computer Methods and Programs in BiomedicineCitation Excerpt :BRS is an indicator of increased risk of SCD in myocardial infarction patients. Arica et al. [27] used HR and systolic pressure signals to assess BRS. The main aim of this paper is to present time, frequency and non-linear features for normal and CAD-affected HR signals.
Linear and non-linear analysis of cardiac health in diabetic subjects
2012, Biomedical Signal Processing and ControlCitation Excerpt :Their proposed method is based on grammatical evolution to discriminate acidemic from normal fetuses. Another application area for heart rate monitoring is to evaluate the risk level for cardiac death in myocardial infarction patients [56]. The lower heart rate variability during deep breathing in patients with diabetic neuropathy was indicated [20,21].
Survey on computer programs and methods for heart diseases prediction and classification
2015, ARPN Journal of Engineering and Applied SciencesAutomated detection of diabetes using higher order spectral features extracted from heart rate signals
2013, Intelligent Data Analysis