- Email: [email protected]
Effects of non-nutritive sucking on heart rate, respiration and oxygenation: a model-based signal processing approach
Comparative Biochemistry and Physiology Part A 132 (2002) 97–106
Effects of non-nutritive sucking on heart rate, respiration and oxygenation: a model-based signal processing approach夞 ¨ b, H. Devliegerb, P. Casaerb G. Morrena,*, S. Van Huffela, I. Helona, G. Naulaersb, H. Daniels a
Department of Electrical Engineering, ESAT-SCDySISTA, Katholieke Universiteit Leuven, Kasteelpark Arenberg 10, Leuven, Belgium b Department of Paediatrics, University Hospital Gasthuisberg, Leuven, Belgium Received 10 January 2001; received in revised form 3 July 2001; accepted 5 July 2001
Abstract Several studies support the idea that the use of pacifiers can reduce the risk of Sudden Infant Death Syndrome. To investigate the effect of non-nutritive sucking (NNS), we measured heart rate, abdominal respiration, EMG and arterial oxygen saturation of 20 neonates. Also, in 10 of these neonates, changes in cerebral hemoglobin concentrations were acquired by means of near-infrared spectroscopy. Using a parametric technique to model the heart rate as a sum of exponentially damped sinusoids, two main frequency components were found in the heart rate during NNS: a frequency of approximately 0.08 Hz due to the alternation of sucking bursts and pauses, and a frequency of approximately 0.8 Hz that reflects the influence of the respiration. Our analysis shows that it is the alternation of bursts and pauses itself that causes the increased heart rate variability, and that this is not due to increased effort. This suggests that the neuronal mechanism regulating NNS also stimulates the heart rate. From our measurements, no effect of NNS on cerebral or peripheral oxygenation could be found. Furthermore, we show that our model-based signal processing technique is well suited for the analysis of non-stationary biomedical signals. 䊚 2002 Elsevier Science Inc. All rights reserved. Keywords: Heart rate variability; Model-based signal processing; Near-infrared spectroscopy; Non-nutritive sucking; Oxygenation; Polysomnography; Respiration; Sudden Infant Death Syndrome
1. Introduction Sudden Infant Death Syndrome (SIDS), the sudden and unexpected death of an infant where a thorough postmortem investigation fails to demonstrate an adequate cause for death, is still one of the most important causes of death during the first year of life (Brooks, 1982; Cote et al., 1999). Several studies support the idea that the use of pacifiers can reduce the risk of SIDS (Arnestad et al., 1997; Becroft et al., 1998; Fleming et al., 1999; L’Hoir et al., 1998). Possible mechanisms 夞 This paper was presented as part of ISOTT2000 held in Nijmegen, The Netherlands, August 20–25, 2000. The organiser was Berend Oeseburg. *Corresponding author. Tel.: q32-16-321-857; fax: q3216-321-970. E-mail address: [email protected] (G. Morren).
are preventing the tongue from falling back sealing the airways (Cozzi et al., 1979), reducing gastroesophageal reflux periods (Anonymous, 1988) and reducing apnea periods by stimulating respiratory drive (Daniels et al., 1986). Many authors have also found a relationship between the relative risk of SIDS and the degree of maturation of cardiorespiratory control. Therefore, we investigated the effect of non-nutritive sucking (NNS) — sucking without delivery of fluid — on heart rate, respiration and oxygenation. 2. Materials and methods 2.1. Measurements Polysomnographies, where ECG, heart rate, thoracic and abdominal respiratory movements, nasal
1095-6433/02/$ - see front matter 䊚 2002 Elsevier Science Inc. All rights reserved. PII: S 1 0 9 5 - 6 4 3 3 Ž 0 1 . 0 0 5 3 4 - 7
98
G. Morren et al. / Comparative Biochemistry and Physiology Part A 132 (2002) 97–106
Fig. 1. ECG, heart rate, abdominal respiration, SaO2 and EMG during non-nutritive sucking (time scale: 4 s per division).
flow, arterial oxygen saturation (SaO2), EMG of the chin, EEG and EOG of neonates are measured during a 15-h period, are performed frequently in the University Hospital Gasthuisberg. Ten polysomnographies containing several long periods of NNS were selected for this study. In 10 neonates we also measured cerebral oxygenation by nearinfrared spectroscopy (NIRS). Differences in the concentration of oxyhemoglobin (HbO2), deoxyhemoglobin (Hb) and total hemoglobin (HbTots HbO2qHb) were acquired using a NIRO-300 (Hamamatsu, Japan). These measurements lasted for a maximum of 1 h. Therefore, the measurements without NIRS contain more and longer periods of non-nutritive sucking. All subjects had a postmenstrual age (PMA) between 29 and 40 weeks at birth and a PMA between 35 and 49 weeks at the time of the measurements (Helon, 2000). All data were recorded simultaneously with a sampling frequency of 100 Hz by the data acquisition system CODAS (Dataq Instruments, USA) and stored on a PC. All polysomnographic signals are analog and digitized by the CODAS system. The NIRO-300 signals are digital with a sampling rate of 2 Hz. They are converted to analog signals by a sample-and-hold function before their introduction in the CODAS system. In Fig. 1, the most important polysomnographic sig-
nals — ECG, heart rate, abdominal respiration, SaO2 and EMG are shown during a period of NNS. The EMG-signal measures the muscle activity on the chin and is used to identify the periods of NNS. The typical pattern of NNS is shown in Fig. 2. The mean duration of the bursts and pauses is approximately 6 s, but this varies and pauses tend to be longer than bursts. 2.2. Methods Since the CODAS system allows visualizing different channels simultaneously, it is well suited for the analysis of the signals in the time domain. The effects of NNS on heart rate, respiration and oxygenation can also be studied in the frequency domain. The Fourier transform (or fast Fourier transform, FFT), where the signal is represented as a linear combination of complex sinusoids, is most commonly used in the extraction of frequency information of biomedical signals. The power spectrum is the squared amplitude of the FFT. One limitation of the FFT (and power spectrum) is that it gives the frequency content of the signal without providing the time location of the observed frequency components. Therefore, it is not suited for the analysis of non-stationary signals, of which the
G. Morren et al. / Comparative Biochemistry and Physiology Part A 132 (2002) 97–106
99
Fig. 2. EMG on the chin during NNS.
nature changes with time. A method to represent the signal both in the frequency and time domain concurrently is the short-time Fourier transform (STFT) (Lin and Chen, 1996; Oppenheim et al., 1999). In this method, the signal is considered stationary when seen through a window of limited extent. The STFT at time t is the FFT of the signal seen through a window centered at time t. By moving the window, the frequency content can be calculated as a function of time. Using the power spectrum instead of the FFT yields the spectrogram. One drawback of the STFT (and spectrogram) is the trade-off between time and frequency resolution. To increase the frequency resolution, a longer time interval is required, so that the stationary assumption might not be valid any more and the spectral components will be smeared out in the time domain. As an example, the spectrogram of the EMG during a period of non-nutritive sucking is shown in Fig. 3. The alternation between bursts and pauses is clearly visible. During bursts, there is a peak at the sucking frequency of approximately 2.5 Hz. In addition, this peak moves to lower frequencies towards the end of a burst. Another way to analyze the signals is to quantify them directly in the time domain by means of model-based (or parametric) methods. We used a method called HTLSU (Lemmerling, 1999;Van Huffel et al., 1994), where the signal is modeled as a sum of K exponentially damped sinusoids: K
y˘ ns 8 ak expŽ jpk. expwŽydkqj2pfk.nDtx ks1
where y˘ n represents the n-th modeled data point and Dt the constant sampling interval between consecutive data points. The parameters to be determined are the frequencies f k, dampings dk, amplitudes ak and phases pk. The model is a good approximation of the signal if the model order K, which is twice the number of frequency components, is chosen correctly. HTLSU allows quantifying the frequencies accurately with fewer data points than non-parametric methods such as the FFT. Therefore, the method is better suited for non-stationary signals. As an example, the modeling of the respiration is shown in Fig. 4, together with the respiration itself. The analysis with HTLSU allows quantifying the shift in respiratory rate. An interesting generalization is multichannel HTLSU, which allows you to model different signals (called channels) at the same time, e.g. EMG together with ECG or SaO2. In this case, the data from each channel are arranged in separate Hankel matrices and these are stacked together in one block matrix. HTLSU is applied to this matrix in order to extract common frequencies and dampings. Once these common nonlinear parameters are identified, the corresponding amplitudes and phases are calculated separately for each channel. More details about the algorithm and some applications can be found in (Vanhamme, 1999; Vanhamme and Van Huffel, 1998). All signals are modeled with common frequencies f k and dampings dk, but different amplitudes ak and phases pk. Multichannel HTLSU is a powerful tool when looking for mutual influences between different
100
G. Morren et al. / Comparative Biochemistry and Physiology Part A 132 (2002) 97–106
Fig. 3. Time-frequency plot of the EMG during NNS, as shown in Fig. 2. Only the frequency region 0.25–4 Hz, containing the sucking frequency, is shown.
signals. The main contribution of this paper is that we show the usefulness of such a model-based signal processing approach that does not involve any statistical analysis. 3. Results 3.1. Effects of NNS on heart rate The effect of non-nutritive sucking on the heart rate can be detected when analyzing the measurements in the time domain. The heart rate increases during bursts and decreases during pauses. This results in an increase in heart rate variability (HRV) during NNS. An example is shown in Fig. 5. Modeling the heart rate with HTLSU reveals two main frequencies. The frequency of approximately 0.08 Hz is due to the alternation of bursts and pauses. However, another important frequency
component of approximately 0.8 Hz is noted reflecting the influence of the respiration on the heart rate, known as ‘respiratory sinus arrhythmia’ (RSA). Fig. 6 shows the heart rate, modeled with model order 10. The reason for using model order 10 is the shift in respiratory rate during the observed period. The five frequency components are: DC-value, NNS-effect (0.08 Hz) and three RSA-frequencies (between 0.8 Hz and 0.9 Hz). The sucking frequency (2–3 Hz, due to the rapid sucks within the bursts) is not visible, because it is higher than half the heart rate (120–180 bpms 2–3 Hz) and, therefore, the Nyquist-criterion does not hold (Oppenheim et al., 1999). Modeling the normalized heart rate and preprocessed EMG simultaneously using multichannel HTLSU shows the effect even more explicitly. (Normalizing is defined here as subtracting the mean from the signal and consequently dividing it by the standard deviation.) The pre-processing of
G. Morren et al. / Comparative Biochemistry and Physiology Part A 132 (2002) 97–106
101
Fig. 4. Modeling of respiration with HTLSU. The shift in respiratory rate can be quantified in each time window and fluctuates as follows: 0.42 Hz; 0.33 Hz; 0.45 Hz; 0.58 Hz.
Fig. 5. Effect of non-nutritive sucking on heart rate.
102
G. Morren et al. / Comparative Biochemistry and Physiology Part A 132 (2002) 97–106
Fig. 6. Modeling of heart rate during NNS with HTLSU (model order 10).
the EMG consists of normalizing and taking absolute values. The latter is done to reduce the influence of the sucking frequency, which contains
no valuable information for our analysis. The normalization of both signals is necessary because the signals, expressed in different units, have a
Fig. 7. Modeling of (normalized) heart rate and (pre-processed) EMG during NNS with multichannel HTLSU (model order 2).
G. Morren et al. / Comparative Biochemistry and Physiology Part A 132 (2002) 97–106
103
Table 1 Effect of NNS on heart rate for different subjects. The values between brackets are the minimum and maximum HR increase in the observed period Baby
Duration (s)
Frequency (Hz)
Median HR (bpm)
HR increase by NNS (bpm)
P1 P2 P3 P4 P5 P6 P7 P9
79 31 87 331 178 104 127 84
0.092 0.129 0.077 0.078 0.067 0.070 0.078 0.096
158 149 140 121 140 141 142 128
17 26 11 20 13 9 6 26
different range. Without normalization the signal with the largest range would be better modeled, since the modeling is based on an energy-decomposition. Fig. 7 shows EMG and heart rate, modeled simultaneously with model order 2. Multichannel HTLSU extracts 0.08 Hz (alternation of sucking bursts and pauses) as the most energetic component. The effect of NNS on heart rate can be seen systematically in 9 of the 10 measurements without NIRS. Table 1 gives an overview of the effect for some periods of NNS. The increase in heart rate is shown together with the median heart rate (over the observed period), the duration of the period and the frequency corresponding to the alternation between bursts and pauses (obtained from the multichannel HTLSU modeling of EMG and heart rate). In one measurement without NIRS and the measurements with NIRS, the effect is not always present, probably because these measurements contain fewer and shorter periods of NNS (Helon, 2000) (Table 1). 3.2. Effects of NNS on oxygenation The effect of NNS on cerebral oxygenation is shown in Fig. 8. The heart rate, during a period of NNS, is displayed together with the concentration changes of HbO2, Hb and HbTot. The effect on the heart rate is also reflected in the concentration changes of HbO2 and HbTot. An increase (decrease) in heart rate is, after a time delay of approximately 4 s (Van Huffel et al., 1998), followed by an increase (decrease) in the concentration of HbO2 and HbTot. The effect of NNS on peripheral oxygenation is shown in Fig. 9, which shows the EMG together with the arterial oxygen saturation (SaO2 ). In most
(7–22) (20–32) (9–13) (9–34) (9–14) (7–12) (5–7) (20–33)
cases, there is no change in SaO2 during periods of NNS. When the baby breathes irregularly, however, a decrease in SaO2 is noted. 4. Discussion 4.1. Effects of NNS on heart rate The increase (decrease) in heart rate during bursts (pauses) was observed by other researchers (Casaer, 1979), but never thoroughly investigated. In other studies, it was proved by means of spectral analysis that the variability of the heart rate increases during NNS (Franco et al., 1999). Our analysis shows that the increase in HRV introduced by NNS is not chaotic, but very regular. To our knowledge, the origin of the increased HRV has not been described to date. It cannot be explained by an increased effort due to sucking since there is no cumulative effect. During pauses, the heart rate decreases to the value of the period before NNS. Furthermore, the heart rate increases immediately at the beginning of the sucking bursts; there is no time delay between the onset of sucking and the onset of the increase in heart rate. Therefore, it is hypothesized that the neuronal mechanism regulating NNS also stimulates the heart rate. The increase in HRV introduced by NNS is similar to the increase in HRV introduced by RSA. The latter can be explained by a neuronal mechanism in the brainstem (Berntson et al., 1993). Elementary functions such as respiration and heart contractions, but also NNS in neonates, are controlled from within the brainstem. Neuronal influence in the brainstem could therefore be a plausible explanation for the increased HRV due to NNS. This would imply that NNS is a primary brainstemgenerated activity intimately linked to vital brain-
104
G. Morren et al. / Comparative Biochemistry and Physiology Part A 132 (2002) 97–106
Fig. 8. Effect of NNS on cerebral oxygenation.
stem functions, such as heart rate control and respiration. This link could explain the possible protective effect in SIDS of increased HRV as a result of NNS. 4.2. Effects of NNS on oxygenation From Fig. 8, it can be concluded that there is an increase in cerebral blood volume (CBV) during sucking bursts, since changes in HbTot are proportional to changes in CBV if haematocrit remains constant (Pryds et al., 1990). The increase in heart rate during bursts causes, after a delay,
the increase in CBV. However, since there are concordant changes in HbO2 and HbTot concentrations, this does not indicate a better oxygenation. So, from our measurements, no effect of NNS on cerebral oxygenation could be found. Also, no effect of NNS on peripheral oxygenation could be found. In most cases, no change (apart from noise variations) is noted in arterial oxygen saturation (SaO2) during periods of NNS. NNS during a period of irregular breathing produces a decrease in SaO2, merely as a consequence of poor coordination of breathing and sucking rather than of NNS itself.
G. Morren et al. / Comparative Biochemistry and Physiology Part A 132 (2002) 97–106
105
Fig. 9. Effect of NNS on peripheral oxygenation.
4.3. Conclusions Heart rate variability increases during non-nutritive sucking. The alternation of sucking bursts and pauses is accompanied by a simultaneous increase and decrease of the heart rate. This suggests that the neuronal mechanism regulating non-nutritive sucking also stimulates the heart rate. From our measurements, an increase in cerebral blood volume during sucking bursts is observed, but no effect of non-nutritive sucking on cerebral or peripheral oxygenation could be found. Furthermore, it has been shown that our model-based signal processing technique (HTLSU) is well suited for the analysis of non-stationary biomedical signals. Acknowledgments This paper presents research results of the Belgian Programme on Interuniversity Poles of Attraction (IUAP P4-02 and P4-24), initiated by the Belgian State, Prime Minister’s Office-Federal Office for Scientific, Technical and Cultural Affairs, of the European Community TMR Pro-
gramme, Networks, project CHRX-CT97-0160, of the Brite-Euram Programme, Thematic Network BRRT-CT97-5040‘Niconet’, of the Concerted Research Action (GOA) projects of the Flemish Government MEFISTO-666 (Mathematical Engineering for Information and Communication Systems Technology), of the IDO/99/03 project (K.U.Leuven) ‘Predictive computer models for medical classification problems using patient data and expert knowledge’, of the FWO ‘Krediet aan navorsers’ G.0326.98 and the FWO project G.0200.00. References Anonymous, 1988. Gastro-oesophageal reflux and apparent life-threatening events in infancy. Lancet 2, 261–262. Arnestad, M., Andersen, M., Rognum, T.O., 1997. Is the use of dummy or carry-cot of importance for sudden infant death? wsee commentsx. Eur. J. Pediatr. 156, 968–970. Becroft, D.M., Thompson, J.M., Mitchell, E.A., 1998. Epidemiology of intrathoracic petechial hemorrhages in sudden infant death syndrome. Pediatr. Dev. Pathol. 1, 200–209. Berntson, G.G., Cacioppo, J.T., Quigley, K.S., 1993. Respiratory sinus arrhythmia: autonomic origins, physiological mechanisms, and psychophysiological implications. Psychophysiology 30, 183–196.
106
G. Morren et al. / Comparative Biochemistry and Physiology Part A 132 (2002) 97–106
Brooks, J.G., 1982. Apnea of infancy and sudden infant death syndrome. Am. J. Dis. Child. 136, 1012–1023. Casaer, P., 1979. Postural behaviour in newborn infants. Clin. Dev. Med. 72, 70–73. Cote, A., Russo, P., Michaud, J., 1999. Sudden unexpected deaths in infancy: what are the causes? J. Pediatr. 135, 437–443. Cozzi, F., Albani, R., Cardi, E., 1979. A common pathophysiology for sudden cot death and sleep apnoea. The vacuumglossoptosis syndrome. Med. Hypotheses 5, 329–338. Daniels, H., Devlieger, H., Casaer, P., et al., 1986. Nutritive and non-nutritive sucking in preterm infants. J. Dev. Physiol. 8, 117–121. Fleming, P.J., Blair, P.S., Pollard, K, et al., 1999. Pacifier use and sudden infant death syndrome: results from the CESDIy SUDI case control study. CESDI SUDI Research Team wsee commentsx. Arch. Dis. Child. 81, 112–116. Franco, P., Scaillet, S., et al., 1999. Influence of pacifier on sleep characteristics in healthy infants. Pediatr. Pulmonol. 28, 469. Helon, I., 2000. Studie bij pasgeborenen naar de invloed van niet-voedend zuigen op fysiologische parameters en hersencirculatie. Master’s Thesis, Department of electrical engineering, Katholieke Universiteit Leuven, Leuven, Belgium. Lemmerling P., 1999. Structured total least squares: analysis, algorithms and applications. Ph.D.-Thesis, Department of electrical engineering, Katholieke Universiteit Leuven, Leuven, Belgium.
L’Hoir, M.P., Engelberts, A.C., van Well, G.T., et al., 1998. Risk and preventive factors for cot death in The Netherlands, a low-incidence country. Eur. J. Pediatr. 157, 681–688. Lin, Z., Chen, J.D., 1996. Advances in time-frequency analysis of biomedical signals. Crit. Rev. Biomed. Eng. 24, 1–72. Oppenheim, A.V., Schafer, R.W., Buck, J.R., 1999. Discretetime signal processing, 2nd Ed. Prentice Hall, Upper Saddle River. Pryds, O., Greisen, G., Skov, L.L., et al., 1990. Carbon dioxiderelated changes in cerebral blood volume and cerebral blood flow in mechanically ventilated preterm neonates: comparison of near infrared spectrophotometry and 133Xenon clearance. Pediatr. Res. 27, 445–449. Van Huffel, S., Chen, H., Decanniere, C., et al., 1994. Algorithm for time-domain NMR data fitting based on total least squares. J. Magn. Res. A 110, 228–237. Van Huffel S, Craemers J, Lenaerts B, Daniels H, Naulaers G, Casaer P, 1998. Episodes of apnea and bradycardia in the preterm newborn: impact on cerebral oxygenation measured by near infrared spectrophotometry. In: Photon propagation in tissues IV, (Eds) Benaron D, Chance B, Ferrari M, Kohl M, Proceedings of SPIE, Vol. 3566, pp. 112–123. Vanhamme, L., 1999. Advanced time-domain methods for nuclear magnetic resonance spectroscopy data analysis. pH.D.-Thesis, Dept. of electrical engineering, Katholieke Universiteit Leuven, Leuven, Belgium. Vanhamme L, Van Huffel S, 1998. Multichannel quantification of biomedical Magnetic Resonance Spectroscopic signals. In: Advanced Signal Processing Algorithms, Architectures, and Implementations VIII, (Eds) Luk FT, Proceedings of SPIE, Vol. 3461, pp. 237–248.
Effects of non-nutritive sucking on heart rate, respiration and oxygenation: a model-based signal processing approach夞 ¨ b, H. Devliegerb, P. Casaerb G. Morrena,*, S. Van Huffela, I. Helona, G. Naulaersb, H. Daniels a
Department of Electrical Engineering, ESAT-SCDySISTA, Katholieke Universiteit Leuven, Kasteelpark Arenberg 10, Leuven, Belgium b Department of Paediatrics, University Hospital Gasthuisberg, Leuven, Belgium Received 10 January 2001; received in revised form 3 July 2001; accepted 5 July 2001
Abstract Several studies support the idea that the use of pacifiers can reduce the risk of Sudden Infant Death Syndrome. To investigate the effect of non-nutritive sucking (NNS), we measured heart rate, abdominal respiration, EMG and arterial oxygen saturation of 20 neonates. Also, in 10 of these neonates, changes in cerebral hemoglobin concentrations were acquired by means of near-infrared spectroscopy. Using a parametric technique to model the heart rate as a sum of exponentially damped sinusoids, two main frequency components were found in the heart rate during NNS: a frequency of approximately 0.08 Hz due to the alternation of sucking bursts and pauses, and a frequency of approximately 0.8 Hz that reflects the influence of the respiration. Our analysis shows that it is the alternation of bursts and pauses itself that causes the increased heart rate variability, and that this is not due to increased effort. This suggests that the neuronal mechanism regulating NNS also stimulates the heart rate. From our measurements, no effect of NNS on cerebral or peripheral oxygenation could be found. Furthermore, we show that our model-based signal processing technique is well suited for the analysis of non-stationary biomedical signals. 䊚 2002 Elsevier Science Inc. All rights reserved. Keywords: Heart rate variability; Model-based signal processing; Near-infrared spectroscopy; Non-nutritive sucking; Oxygenation; Polysomnography; Respiration; Sudden Infant Death Syndrome
1. Introduction Sudden Infant Death Syndrome (SIDS), the sudden and unexpected death of an infant where a thorough postmortem investigation fails to demonstrate an adequate cause for death, is still one of the most important causes of death during the first year of life (Brooks, 1982; Cote et al., 1999). Several studies support the idea that the use of pacifiers can reduce the risk of SIDS (Arnestad et al., 1997; Becroft et al., 1998; Fleming et al., 1999; L’Hoir et al., 1998). Possible mechanisms 夞 This paper was presented as part of ISOTT2000 held in Nijmegen, The Netherlands, August 20–25, 2000. The organiser was Berend Oeseburg. *Corresponding author. Tel.: q32-16-321-857; fax: q3216-321-970. E-mail address: [email protected] (G. Morren).
are preventing the tongue from falling back sealing the airways (Cozzi et al., 1979), reducing gastroesophageal reflux periods (Anonymous, 1988) and reducing apnea periods by stimulating respiratory drive (Daniels et al., 1986). Many authors have also found a relationship between the relative risk of SIDS and the degree of maturation of cardiorespiratory control. Therefore, we investigated the effect of non-nutritive sucking (NNS) — sucking without delivery of fluid — on heart rate, respiration and oxygenation. 2. Materials and methods 2.1. Measurements Polysomnographies, where ECG, heart rate, thoracic and abdominal respiratory movements, nasal
1095-6433/02/$ - see front matter 䊚 2002 Elsevier Science Inc. All rights reserved. PII: S 1 0 9 5 - 6 4 3 3 Ž 0 1 . 0 0 5 3 4 - 7
98
G. Morren et al. / Comparative Biochemistry and Physiology Part A 132 (2002) 97–106
Fig. 1. ECG, heart rate, abdominal respiration, SaO2 and EMG during non-nutritive sucking (time scale: 4 s per division).
flow, arterial oxygen saturation (SaO2), EMG of the chin, EEG and EOG of neonates are measured during a 15-h period, are performed frequently in the University Hospital Gasthuisberg. Ten polysomnographies containing several long periods of NNS were selected for this study. In 10 neonates we also measured cerebral oxygenation by nearinfrared spectroscopy (NIRS). Differences in the concentration of oxyhemoglobin (HbO2), deoxyhemoglobin (Hb) and total hemoglobin (HbTots HbO2qHb) were acquired using a NIRO-300 (Hamamatsu, Japan). These measurements lasted for a maximum of 1 h. Therefore, the measurements without NIRS contain more and longer periods of non-nutritive sucking. All subjects had a postmenstrual age (PMA) between 29 and 40 weeks at birth and a PMA between 35 and 49 weeks at the time of the measurements (Helon, 2000). All data were recorded simultaneously with a sampling frequency of 100 Hz by the data acquisition system CODAS (Dataq Instruments, USA) and stored on a PC. All polysomnographic signals are analog and digitized by the CODAS system. The NIRO-300 signals are digital with a sampling rate of 2 Hz. They are converted to analog signals by a sample-and-hold function before their introduction in the CODAS system. In Fig. 1, the most important polysomnographic sig-
nals — ECG, heart rate, abdominal respiration, SaO2 and EMG are shown during a period of NNS. The EMG-signal measures the muscle activity on the chin and is used to identify the periods of NNS. The typical pattern of NNS is shown in Fig. 2. The mean duration of the bursts and pauses is approximately 6 s, but this varies and pauses tend to be longer than bursts. 2.2. Methods Since the CODAS system allows visualizing different channels simultaneously, it is well suited for the analysis of the signals in the time domain. The effects of NNS on heart rate, respiration and oxygenation can also be studied in the frequency domain. The Fourier transform (or fast Fourier transform, FFT), where the signal is represented as a linear combination of complex sinusoids, is most commonly used in the extraction of frequency information of biomedical signals. The power spectrum is the squared amplitude of the FFT. One limitation of the FFT (and power spectrum) is that it gives the frequency content of the signal without providing the time location of the observed frequency components. Therefore, it is not suited for the analysis of non-stationary signals, of which the
G. Morren et al. / Comparative Biochemistry and Physiology Part A 132 (2002) 97–106
99
Fig. 2. EMG on the chin during NNS.
nature changes with time. A method to represent the signal both in the frequency and time domain concurrently is the short-time Fourier transform (STFT) (Lin and Chen, 1996; Oppenheim et al., 1999). In this method, the signal is considered stationary when seen through a window of limited extent. The STFT at time t is the FFT of the signal seen through a window centered at time t. By moving the window, the frequency content can be calculated as a function of time. Using the power spectrum instead of the FFT yields the spectrogram. One drawback of the STFT (and spectrogram) is the trade-off between time and frequency resolution. To increase the frequency resolution, a longer time interval is required, so that the stationary assumption might not be valid any more and the spectral components will be smeared out in the time domain. As an example, the spectrogram of the EMG during a period of non-nutritive sucking is shown in Fig. 3. The alternation between bursts and pauses is clearly visible. During bursts, there is a peak at the sucking frequency of approximately 2.5 Hz. In addition, this peak moves to lower frequencies towards the end of a burst. Another way to analyze the signals is to quantify them directly in the time domain by means of model-based (or parametric) methods. We used a method called HTLSU (Lemmerling, 1999;Van Huffel et al., 1994), where the signal is modeled as a sum of K exponentially damped sinusoids: K
y˘ ns 8 ak expŽ jpk. expwŽydkqj2pfk.nDtx ks1
where y˘ n represents the n-th modeled data point and Dt the constant sampling interval between consecutive data points. The parameters to be determined are the frequencies f k, dampings dk, amplitudes ak and phases pk. The model is a good approximation of the signal if the model order K, which is twice the number of frequency components, is chosen correctly. HTLSU allows quantifying the frequencies accurately with fewer data points than non-parametric methods such as the FFT. Therefore, the method is better suited for non-stationary signals. As an example, the modeling of the respiration is shown in Fig. 4, together with the respiration itself. The analysis with HTLSU allows quantifying the shift in respiratory rate. An interesting generalization is multichannel HTLSU, which allows you to model different signals (called channels) at the same time, e.g. EMG together with ECG or SaO2. In this case, the data from each channel are arranged in separate Hankel matrices and these are stacked together in one block matrix. HTLSU is applied to this matrix in order to extract common frequencies and dampings. Once these common nonlinear parameters are identified, the corresponding amplitudes and phases are calculated separately for each channel. More details about the algorithm and some applications can be found in (Vanhamme, 1999; Vanhamme and Van Huffel, 1998). All signals are modeled with common frequencies f k and dampings dk, but different amplitudes ak and phases pk. Multichannel HTLSU is a powerful tool when looking for mutual influences between different
100
G. Morren et al. / Comparative Biochemistry and Physiology Part A 132 (2002) 97–106
Fig. 3. Time-frequency plot of the EMG during NNS, as shown in Fig. 2. Only the frequency region 0.25–4 Hz, containing the sucking frequency, is shown.
signals. The main contribution of this paper is that we show the usefulness of such a model-based signal processing approach that does not involve any statistical analysis. 3. Results 3.1. Effects of NNS on heart rate The effect of non-nutritive sucking on the heart rate can be detected when analyzing the measurements in the time domain. The heart rate increases during bursts and decreases during pauses. This results in an increase in heart rate variability (HRV) during NNS. An example is shown in Fig. 5. Modeling the heart rate with HTLSU reveals two main frequencies. The frequency of approximately 0.08 Hz is due to the alternation of bursts and pauses. However, another important frequency
component of approximately 0.8 Hz is noted reflecting the influence of the respiration on the heart rate, known as ‘respiratory sinus arrhythmia’ (RSA). Fig. 6 shows the heart rate, modeled with model order 10. The reason for using model order 10 is the shift in respiratory rate during the observed period. The five frequency components are: DC-value, NNS-effect (0.08 Hz) and three RSA-frequencies (between 0.8 Hz and 0.9 Hz). The sucking frequency (2–3 Hz, due to the rapid sucks within the bursts) is not visible, because it is higher than half the heart rate (120–180 bpms 2–3 Hz) and, therefore, the Nyquist-criterion does not hold (Oppenheim et al., 1999). Modeling the normalized heart rate and preprocessed EMG simultaneously using multichannel HTLSU shows the effect even more explicitly. (Normalizing is defined here as subtracting the mean from the signal and consequently dividing it by the standard deviation.) The pre-processing of
G. Morren et al. / Comparative Biochemistry and Physiology Part A 132 (2002) 97–106
101
Fig. 4. Modeling of respiration with HTLSU. The shift in respiratory rate can be quantified in each time window and fluctuates as follows: 0.42 Hz; 0.33 Hz; 0.45 Hz; 0.58 Hz.
Fig. 5. Effect of non-nutritive sucking on heart rate.
102
G. Morren et al. / Comparative Biochemistry and Physiology Part A 132 (2002) 97–106
Fig. 6. Modeling of heart rate during NNS with HTLSU (model order 10).
the EMG consists of normalizing and taking absolute values. The latter is done to reduce the influence of the sucking frequency, which contains
no valuable information for our analysis. The normalization of both signals is necessary because the signals, expressed in different units, have a
Fig. 7. Modeling of (normalized) heart rate and (pre-processed) EMG during NNS with multichannel HTLSU (model order 2).
G. Morren et al. / Comparative Biochemistry and Physiology Part A 132 (2002) 97–106
103
Table 1 Effect of NNS on heart rate for different subjects. The values between brackets are the minimum and maximum HR increase in the observed period Baby
Duration (s)
Frequency (Hz)
Median HR (bpm)
HR increase by NNS (bpm)
P1 P2 P3 P4 P5 P6 P7 P9
79 31 87 331 178 104 127 84
0.092 0.129 0.077 0.078 0.067 0.070 0.078 0.096
158 149 140 121 140 141 142 128
17 26 11 20 13 9 6 26
different range. Without normalization the signal with the largest range would be better modeled, since the modeling is based on an energy-decomposition. Fig. 7 shows EMG and heart rate, modeled simultaneously with model order 2. Multichannel HTLSU extracts 0.08 Hz (alternation of sucking bursts and pauses) as the most energetic component. The effect of NNS on heart rate can be seen systematically in 9 of the 10 measurements without NIRS. Table 1 gives an overview of the effect for some periods of NNS. The increase in heart rate is shown together with the median heart rate (over the observed period), the duration of the period and the frequency corresponding to the alternation between bursts and pauses (obtained from the multichannel HTLSU modeling of EMG and heart rate). In one measurement without NIRS and the measurements with NIRS, the effect is not always present, probably because these measurements contain fewer and shorter periods of NNS (Helon, 2000) (Table 1). 3.2. Effects of NNS on oxygenation The effect of NNS on cerebral oxygenation is shown in Fig. 8. The heart rate, during a period of NNS, is displayed together with the concentration changes of HbO2, Hb and HbTot. The effect on the heart rate is also reflected in the concentration changes of HbO2 and HbTot. An increase (decrease) in heart rate is, after a time delay of approximately 4 s (Van Huffel et al., 1998), followed by an increase (decrease) in the concentration of HbO2 and HbTot. The effect of NNS on peripheral oxygenation is shown in Fig. 9, which shows the EMG together with the arterial oxygen saturation (SaO2 ). In most
(7–22) (20–32) (9–13) (9–34) (9–14) (7–12) (5–7) (20–33)
cases, there is no change in SaO2 during periods of NNS. When the baby breathes irregularly, however, a decrease in SaO2 is noted. 4. Discussion 4.1. Effects of NNS on heart rate The increase (decrease) in heart rate during bursts (pauses) was observed by other researchers (Casaer, 1979), but never thoroughly investigated. In other studies, it was proved by means of spectral analysis that the variability of the heart rate increases during NNS (Franco et al., 1999). Our analysis shows that the increase in HRV introduced by NNS is not chaotic, but very regular. To our knowledge, the origin of the increased HRV has not been described to date. It cannot be explained by an increased effort due to sucking since there is no cumulative effect. During pauses, the heart rate decreases to the value of the period before NNS. Furthermore, the heart rate increases immediately at the beginning of the sucking bursts; there is no time delay between the onset of sucking and the onset of the increase in heart rate. Therefore, it is hypothesized that the neuronal mechanism regulating NNS also stimulates the heart rate. The increase in HRV introduced by NNS is similar to the increase in HRV introduced by RSA. The latter can be explained by a neuronal mechanism in the brainstem (Berntson et al., 1993). Elementary functions such as respiration and heart contractions, but also NNS in neonates, are controlled from within the brainstem. Neuronal influence in the brainstem could therefore be a plausible explanation for the increased HRV due to NNS. This would imply that NNS is a primary brainstemgenerated activity intimately linked to vital brain-
104
G. Morren et al. / Comparative Biochemistry and Physiology Part A 132 (2002) 97–106
Fig. 8. Effect of NNS on cerebral oxygenation.
stem functions, such as heart rate control and respiration. This link could explain the possible protective effect in SIDS of increased HRV as a result of NNS. 4.2. Effects of NNS on oxygenation From Fig. 8, it can be concluded that there is an increase in cerebral blood volume (CBV) during sucking bursts, since changes in HbTot are proportional to changes in CBV if haematocrit remains constant (Pryds et al., 1990). The increase in heart rate during bursts causes, after a delay,
the increase in CBV. However, since there are concordant changes in HbO2 and HbTot concentrations, this does not indicate a better oxygenation. So, from our measurements, no effect of NNS on cerebral oxygenation could be found. Also, no effect of NNS on peripheral oxygenation could be found. In most cases, no change (apart from noise variations) is noted in arterial oxygen saturation (SaO2) during periods of NNS. NNS during a period of irregular breathing produces a decrease in SaO2, merely as a consequence of poor coordination of breathing and sucking rather than of NNS itself.
G. Morren et al. / Comparative Biochemistry and Physiology Part A 132 (2002) 97–106
105
Fig. 9. Effect of NNS on peripheral oxygenation.
4.3. Conclusions Heart rate variability increases during non-nutritive sucking. The alternation of sucking bursts and pauses is accompanied by a simultaneous increase and decrease of the heart rate. This suggests that the neuronal mechanism regulating non-nutritive sucking also stimulates the heart rate. From our measurements, an increase in cerebral blood volume during sucking bursts is observed, but no effect of non-nutritive sucking on cerebral or peripheral oxygenation could be found. Furthermore, it has been shown that our model-based signal processing technique (HTLSU) is well suited for the analysis of non-stationary biomedical signals. Acknowledgments This paper presents research results of the Belgian Programme on Interuniversity Poles of Attraction (IUAP P4-02 and P4-24), initiated by the Belgian State, Prime Minister’s Office-Federal Office for Scientific, Technical and Cultural Affairs, of the European Community TMR Pro-
gramme, Networks, project CHRX-CT97-0160, of the Brite-Euram Programme, Thematic Network BRRT-CT97-5040‘Niconet’, of the Concerted Research Action (GOA) projects of the Flemish Government MEFISTO-666 (Mathematical Engineering for Information and Communication Systems Technology), of the IDO/99/03 project (K.U.Leuven) ‘Predictive computer models for medical classification problems using patient data and expert knowledge’, of the FWO ‘Krediet aan navorsers’ G.0326.98 and the FWO project G.0200.00. References Anonymous, 1988. Gastro-oesophageal reflux and apparent life-threatening events in infancy. Lancet 2, 261–262. Arnestad, M., Andersen, M., Rognum, T.O., 1997. Is the use of dummy or carry-cot of importance for sudden infant death? wsee commentsx. Eur. J. Pediatr. 156, 968–970. Becroft, D.M., Thompson, J.M., Mitchell, E.A., 1998. Epidemiology of intrathoracic petechial hemorrhages in sudden infant death syndrome. Pediatr. Dev. Pathol. 1, 200–209. Berntson, G.G., Cacioppo, J.T., Quigley, K.S., 1993. Respiratory sinus arrhythmia: autonomic origins, physiological mechanisms, and psychophysiological implications. Psychophysiology 30, 183–196.
106
G. Morren et al. / Comparative Biochemistry and Physiology Part A 132 (2002) 97–106
Brooks, J.G., 1982. Apnea of infancy and sudden infant death syndrome. Am. J. Dis. Child. 136, 1012–1023. Casaer, P., 1979. Postural behaviour in newborn infants. Clin. Dev. Med. 72, 70–73. Cote, A., Russo, P., Michaud, J., 1999. Sudden unexpected deaths in infancy: what are the causes? J. Pediatr. 135, 437–443. Cozzi, F., Albani, R., Cardi, E., 1979. A common pathophysiology for sudden cot death and sleep apnoea. The vacuumglossoptosis syndrome. Med. Hypotheses 5, 329–338. Daniels, H., Devlieger, H., Casaer, P., et al., 1986. Nutritive and non-nutritive sucking in preterm infants. J. Dev. Physiol. 8, 117–121. Fleming, P.J., Blair, P.S., Pollard, K, et al., 1999. Pacifier use and sudden infant death syndrome: results from the CESDIy SUDI case control study. CESDI SUDI Research Team wsee commentsx. Arch. Dis. Child. 81, 112–116. Franco, P., Scaillet, S., et al., 1999. Influence of pacifier on sleep characteristics in healthy infants. Pediatr. Pulmonol. 28, 469. Helon, I., 2000. Studie bij pasgeborenen naar de invloed van niet-voedend zuigen op fysiologische parameters en hersencirculatie. Master’s Thesis, Department of electrical engineering, Katholieke Universiteit Leuven, Leuven, Belgium. Lemmerling P., 1999. Structured total least squares: analysis, algorithms and applications. Ph.D.-Thesis, Department of electrical engineering, Katholieke Universiteit Leuven, Leuven, Belgium.
L’Hoir, M.P., Engelberts, A.C., van Well, G.T., et al., 1998. Risk and preventive factors for cot death in The Netherlands, a low-incidence country. Eur. J. Pediatr. 157, 681–688. Lin, Z., Chen, J.D., 1996. Advances in time-frequency analysis of biomedical signals. Crit. Rev. Biomed. Eng. 24, 1–72. Oppenheim, A.V., Schafer, R.W., Buck, J.R., 1999. Discretetime signal processing, 2nd Ed. Prentice Hall, Upper Saddle River. Pryds, O., Greisen, G., Skov, L.L., et al., 1990. Carbon dioxiderelated changes in cerebral blood volume and cerebral blood flow in mechanically ventilated preterm neonates: comparison of near infrared spectrophotometry and 133Xenon clearance. Pediatr. Res. 27, 445–449. Van Huffel, S., Chen, H., Decanniere, C., et al., 1994. Algorithm for time-domain NMR data fitting based on total least squares. J. Magn. Res. A 110, 228–237. Van Huffel S, Craemers J, Lenaerts B, Daniels H, Naulaers G, Casaer P, 1998. Episodes of apnea and bradycardia in the preterm newborn: impact on cerebral oxygenation measured by near infrared spectrophotometry. In: Photon propagation in tissues IV, (Eds) Benaron D, Chance B, Ferrari M, Kohl M, Proceedings of SPIE, Vol. 3566, pp. 112–123. Vanhamme, L., 1999. Advanced time-domain methods for nuclear magnetic resonance spectroscopy data analysis. pH.D.-Thesis, Dept. of electrical engineering, Katholieke Universiteit Leuven, Leuven, Belgium. Vanhamme L, Van Huffel S, 1998. Multichannel quantification of biomedical Magnetic Resonance Spectroscopic signals. In: Advanced Signal Processing Algorithms, Architectures, and Implementations VIII, (Eds) Luk FT, Proceedings of SPIE, Vol. 3461, pp. 237–248.