Main

Rhythmic variations are normally present in physiologic parameters, such as HR, BP, temperature, intracranial pressure, and respiration(1, 2). The variability is complex and is thought to result from the interaction of several components including gain from feedback control loops(3). Analysis of the constituent frequency components can be carried out using spectral analysis. This powerful technique enables a signal in time to be converted into the frequency domain. A cyclic oscillation in the measured signal is reflected in the frequency spectrum as a peak at the frequency of the oscillation. Examination of the spectral content and distribution of spectral power reveals information not available from other methods of analysis.

Using this method to study neonatal HR variability, three major spectral components have been demonstrated. These occur at around 0.05, 0.1, and 0.25 Hz(3–6). The highest frequency peak(>0.2 Hz) is associated with the respiratory cycle(7, 8). The origins of the two lower frequency peaks are less well established, but they are thought to be associated with baroreceptor reflex control (0.1 Hz) and with the thermoregulatory system(0.05 Hz)(2).

The CBFV of neonates has been shown to oscillate at a frequency of between 1 and 5 cycles/min(7, 9, 10). These oscillations appear to be a normal phenomenon in newborn babies(11), and they have also been observed in adults(12). As yet there have been few studies using FFTs to characterize the spectral components of neonatal CBFV. In 1988 Bignall et al.(13) reported use of spectral analysis to examine CBFV signals. They found a variation in CBFV corresponding to respiration frequency, but the short duration of the recordings meant that these authors could not resolve slow variability. Coughtrey et al.(14) used spectral analysis on recordings of 1-min duration and demonstrated several components to the CBFV variability including a respiration-related variability. Zernikow et al.(10) carried out spectral analysis on recordings of CBFV and HR obtained simultaneously from neonates. They found that slow oscillations in CBFV could occur independently of a similar oscillation in HR. A study by Ferrarri et al.(11) confirmed that CBFV oscillations do not simply reflect oscillations in HR.

An alternative approach to the study of physiologic system variability has evolved recently with the finding that the power spectrum of many physiologic variables shows a 1/fα relationship with frequency, where α is a constant close to 1. The physical reasons for this form of relationship are not well understood, but it is thought to be related to the chaotic nature of the system and the complexity of its control mechanisms, suggesting a fractal-like structure. Demonstration of the fractal nature of a system is valuable, because it may promote new ideas about its underlying mechanisms(15). Both HR spectra and BP power spectral densities have previously been reported as having a 1/fα spectral relationship(16, 17).

The aim of this study was to obtain representative estimates of the spectral pattern of neonatal CBFV using a very large number of recordings of sufficient duration to allow adequate resolution of slow variability. The effect of parameters such as PA and GA on the CBFV spectral pattern was also investigated, and we tested the hypothesis that CBFV spectra might follow a 1/fα relationship with frequency. By comparing CBFV spectra with corresponding spectra of HR and BP we attempted to answer the question of whether the CBFV spectra presents information not already available in the other two spectra. If that is the case then CBFV spectra might present useful information for evaluation of cerebral function, to identify pathologic behavior, and to monitor response to treatment.

METHODS

Patients. Data were collected from 106 infants admitted to the Special Care Baby Unit and from 11 term infants on the postnatal ward at the Rosie Maternity Hospital in Cambridge between November 1992 and September 1994. The study was approved by the Cambridge District Ethical Committee, and informed consent was obtained from one or both parents.

Recordings were made from each infant on the day of admission to the unit, and in most cases every 24 h for the first few days of postnatal life. The recording times ranged from 2 to 15 min. In total 522 data records were obtained from 117 infants. Data records of insufficient duration (detailed below) or showing loss of the Doppler signal over a series of cardiac cycles were rejected, reducing the number of infants to 106 and the number of data records available for analysis to 316 (Table 1). After preprocessing of the data, additional files were rejected as described below.

Table 1 Details of patient subgroups
Full size table

The 316 data records were subdivided into groups according to health complications. These are detailed in Table 1. Of the 316 data records, 187 records were obtained from 75 infants with no specific health complications.

The GAs of the infants in this group ranged from 24 to 41 wk. To investigate the effect of GA on the CBFV spectrum, the data records from this group were subdivided into groups according to GA. Similarly the effect of PA was examined by subdividing the data records from the uncomplicated group into groups according to PA. These subgroups are given in Table 1.

The most common complication among the remaining infants was birth asphyxia(59 records, 13 infants). However, the GAs, birth weights, and PAs of the asphyxiated infants are significantly different from those in the uncomplicated group because the asphyxiated infants were all born at term. To make a valid comparison between the asphyxiated and uncomplicated groups, an attempt was made to match (in terms of GA and PA) a sample of records from each group.

Procedures. CBFV was measured using the semicontinuous system described by Evans et al.(18). Recordings were made from the middle cerebral artery using a small 1-cm diameter 4 MHz continuous wave Doppler ultrasound probe, with energy output below the recommended maximum for safety. The CBFV signal was processed by a microcomputer-based system(19), which performs an FFT every 6.25 ms and calculates the peak and intensity weighted mean of the Doppler frequency spectrum. The calculated peak velocity envelope was stored in analog form on a multichannel digital instrumentation tape recorder (PC-108 M Sony, Japan).

Most infants on the Special Care Baby Unit had arterial BP monitored for clinical reasons. This was recorded simultaneously and stored on the digital tape recorder. BP was recorded continuously via either an umbilical or a peripheral arterial catheter using a P23 Spectromed transducer connected to an S&W Quadriscope monitor.

Data analysis; preprocessing of data. On play-back from the digital tape recorder, the CBFV and BP signals were low pass filtered at 30 Hz, and digitized at 200 Hz on a microcomputer. A series of computer programs written in FORTRAN specifically for the purpose was used to analyze the signals. First the signals were low pass filtered by a 20 Hz low pass 8th order Butterworth filter(20), and narrow spikes automatically detected in the CBFV signal were removed by linear interpolation. Plots of the entire duration of each data record were then visually examined. Data containing any obvious artifact were eliminated from the records.

For those data records where BP was available, each occurrence of the end diastolic pressure point within the filtered BP trace was detected, to mark the beginning and end of each cardiac cycle. From this information the R-R interval was estimated.

Intraarterial recordings of BP could not be obtained in normal, healthy term babies (Table 1). In this case the beginning and end of each cardiac cycle was obtained from the foot of the CBFV signal(21). Equivalence of the two methods of extraction of R-R interval (i.e. from BP or CBFV) was checked by using both methods on a sample group of 23 data records matched for GA and PA to the group of term infants.

Values for the mean CBFV and BP were calculated for each cardiac cycle, giving a beat-to-beat series of mean values for the two parameters. HR at each cardiac cycle was calculated from the R-R interval. The series were interpolated, using a third-order polynomial, and resampled at 5 Hz, giving a uniform time axis with an interval of 0.2 s between consecutive data points.

Spectral analysis methods . Analysis by FFT. Spectral analysis of the data were performed using the FFT algorithm. Data of 1024 point lengths were used, giving a total record length of 204.8 s. Before analysis, the signals were low pass filtered at 0.5 Hz using an 8th order Butterworth filter(20), and a least-squares linear regression procedure was used to remove any linear offset and trend from the data. This prevented a large d.c. component from overpowering the higher frequency spectral peaks and reduced any discontinuities at the ends of the data record that may cause spectral leakage. Discontinuities were further reduced by windowing the data with a 20% cosine taper (Tukey) window(20). Any data records containing fewer than 1024 but greater than 800 points were padded to 1024 points by adding zeros to the end of the data record(20). Any records with less than 800 points were discarded. Finally, the power spectrum was calculated for each data record.

Spectral averaging. Mean averaged spectra were obtained by averaging of the individual power spectra. Smoothing of the averaged spectrum was performed using a five-point triangular window. The degrees of freedom for each averaged smoothed spectrum was calculated as 2 × (number of averages) × [0.75 × triangular window width (=5)](22, 23).

The sampling frequency of the data was 5 Hz, which allows the spectrum to be calculated up to a Nyquist frequency of 2.5 Hz, with a frequency resolution of 5/1024 Hz (approximately 0.005 Hz). However, the original sampling of the data was on a beat-to-beat basis, and therefore the maximum frequency to which a meaningful spectrum can be calculated is half of the beat-to-beat sampling frequency; in other words, half of the mean HR. The mean HR of neonates is approximately 2 Hz, therefore the spectra should be valid up to approximately 1 Hz. In this study, however, the data have been low pass filtered at 0.5Hz.

Comparison of spectra. The following methods were used to establish equivalence or otherwise of mean spectra obtained from subgroups of data records. First, the total power of the averaged spectra over the entire frequency band 0-0.5 Hz was calculated by integrating the power spectra over that frequency band. An F test was used to compare the total power values of different spectra. Spectra were then divided into three frequency bands, 0-0.02 Hz, 0.02-0.08 Hz (LF), and 0.08-0.5 Hz in accordance with previous studies(8, 10). The LF band covers the frequency range (i.e. 1-5 cycles/min) in which oscillations in CBFV have been detected by other groups. The ratio of spectral power in the LF band to the total spectral power over the entire band (0-0.5 Hz) was calculated. The total power reflects changes in overall signal variability, and the LF ratio indicates whether redistribution of signal power is taking place between different spectral regions. However, the LF ratio cannot reflect changes in spectral pattern within the LF band, and for this purpose we used aχ2 test described by Bendat and Piersol(23). This was performed on the normalized LF spectra, and as such was sensitive to differences in spectral pattern rather than spectral power in the LF region. In all statistical tests a value of p < 0.05 was considered significant.

1/fα spectrum. The calculated power spectra were tested in this study for an inverse relationship with frequency, by applying a 1/fα model to the data. If the model is valid, a logarithmic plot of power spectral density against frequency should give a straight line having a negative slope of -α. Linearity was tested using linear regression analysis, and evaluation of the coefficient of determination(r2). The autocorrelation of the residuals was tested using the Durbin-Watson test(24). This determines whether the residuals from the linear regression are serially independent and distribution of the data about the regression line is random, which is an assumption made in linear regression analysis. A t test was used to compare slopes(i.e. values of α).

RESULTS

Averaged spectra from infants showing no clinical complications. The CBFV power spectrum for each data record from the group of infants showing no specific health complications was obtained. Individual spectra were averaged over the whole group, and the resulting averaged spectrum is shown in Figure 1. The same procedure was carried out for the BP and HR records, and these are also shown in Figure 1. The number of infants and records for which good quality, simultaneous recordings of CBFV, BP, and HR were available was 61 and 157, respectively. Each of the three spectra has been normalized by dividing by the total power in the spectrum up to 0.5 Hz. Visual inspection of the spectra suggests a similar spectral pattern for all three variables, with a large LF power content and very little spectral power above approximately 0.1 Hz. Spectra are displayed only to 0.3 Hz because very little spectral power exists above this frequency. A comparison was made between the three spectra, and results are given in Table 2. The significant difference in LF spectral pattern between CBFV and both BP and HR spectra, although not apparent in Figure 1, results from the very large number of data records (namely 157) used in the analysis.

Figure 1
figure 1

Mean power spectra of CBFV, BP, and HR from infants with no complications. Spectra are normalized by area. The LF range between 0.02 and 0.08 Hz is indicated.

Full size image
Table 2 Comparison of mean CBFV, BP, and HR spectra for premature infants without other health complications
Full size table

The spectra in Figure 1 were tested for a 1/fα relationship with frequency by plotting the logarithm of the spectral power against the logarithm of the frequency. The resulting plots are shown in Figure 2. In this case no smoothing of the spectra has been carried out so as not to affect the linear regressions. The plots have been linearly offset from each other for clarity; this does not affect their respective values of α.

Figure 2
figure 2

Spectra of Figure 1 plotted on a log-log scale. Regression lines (given in Table 3) are also plotted. Spectra are linearly offset along spectral power axis. Key: CBFV(▪), BP (□), HR (•).

Full size image

Linear regression was initially carried out on the logarithmic data over the entire frequency range up to 0.5 Hz, and the results of the regression analyses over this frequency range are given in Table 3. The Durbin-Watson statistic was calculated, and results of this test are also given in Table 3. It is clear from this test that even though the linear regressions are highly significant, the residuals are not serially independent.

Table 3 Results of fitting a 1/fα model to the CBFV, HR, and BP spectra, from Figure 2
Full size table

Examination of Figure 2 would suggest that the serial dependence of the residuals is due to a change in slope, this becoming more negative at frequencies above approximately 0.06 Hz. Two separate regression analyses were performed on the data points above and below 0.06 Hz, and the resulting regression lines are plotted in Figure 2 and detailed in Table 3. For all three variables the difference in slope between the two frequency ranges is highly significant atp < 0.001.

Over the higher frequency range, 0.06-0.5 Hz, the regression slopes for the three variables are markedly different (p < 0.001) with the BP spectrum falling most rapidly with frequency. Consecutive residuals are not independent, suggesting the 1/fα model is not valid over the range 0.06-0.5 Hz. Over the lower frequency range, 0.005-0.06 Hz, the three slopes are not significantly different, and for both the CBFV and the BP, the residuals are not serially correlated, suggesting a valid 1/fα model.

Effect of GA. Figure 3 shows the averaged spectra of CBFV (in absolute units) from subgroups of the infants with no complications, classified according to GA. Results of a statistical analysis of Figure 3 are given in Table 4 with similar analyses for the effects of GA on BP and HR spectra. The difference in the numbers of infants and data records observed between Table 1 and Tables 4–6 are due to a small number of BP and HR signals showing a high incidence of transients or sharp artifacts. These records were not included in the spectral analysis. The table contains (1) the mean CBFV, BP, and HR for each of the subgroups studied, (2) the total power to 0.5 Hz,(3) the spectral power in the LF range, (4) the ratio of LF spectral power to total spectral power, and (5) the results of the χ2 test comparing the spectral pattern in the LF range. The results of each GA subgroup were compared statistically with those from the 36-41-wk GA subgroup (i.e. the term infants) for the CBFV and HR spectra using the tests described in “Methods.” For the BP spectra the reference was the 32- to 35-wk GA subgroup. No significant effect of GA was observed in any of the CBFV parameters listed in Table 4. However, the results indicate that both the BP and HR signal power increase significantly with GA. Although the CBFV total power also increases for the subgroups with 32-35-wk and 36-41-wk GA, this was not statistically significant. Finally, the mean HR was found to be significantly reduced for the subgroup of term infants in relation to the three subgroups of premature babies.

Figure 3
figure 3

Mean power spectra of CBFV grouped according to GA.Table 4 gives statistical results. The LF range between 0.02 and 0.08 Hz is shown.

Full size image
Table 4 Effect of GA on the CBFV, BP, and HR spectra
Full size table

Effect of PA. Figure 4 shows the averaged spectra of CBFV from subgroups of the infants classified according to PA;Table 5 gives the results of the statistical analyses, including the BP and HR spectra. No significant differences were observed in the mean values of CBFV, BP, and HR in relation to the 0-12-h subgroup which was used as the reference for the statistical tests. On the other hand, total power significantly increased with PA for all three variables, except HR for the 12-24-h subgroup. Interestingly enough, only CBFV showed significant changes in LF spectral pattern with PA, after 24 h of postnatal life. This change is also reflected by the gradual reduction in LF power ratio with increasing PA.

Figure 4
figure 4

Mean power spectra of CBFV grouped according to PA.Table 5 gives statistical results. The LF range between 0.02 and 0.08 Hz is shown.

Full size image
Table 5 Effect of postnatal age on the CBFV, BP, and HR spectra
Full size table

Birth asphyxia. Selecting a subgroup of records from infants without complications to match the GA and PA of the asphyxiated group led to an appropriate matching as reflected by the mean ± SD values of GA and PA given in Table 6. The corresponding average CBFV power spectra for the two groups are shown in Figure 5. There is a clear increase in spectral power for newborns with asphyxia in the very LF range, with a corresponding reduction in power in the range 0.03-0.2 Hz. The net result, reflected by the statistical analysis presented in Table 6, is a slight and nonsignificant increase in total power for the group with asphyxia. On the other hand there was a very significant increase in mean CBFV in the asphyxiated group, corresponding to a 161% change in the DC value (not represented in Fig. 5). The differing spectral patterns between the normal and asphyxiated groups over the LF range (Fig. 5) also led to a very significant statistical difference when assessed with the χ2 test(Table 6).

Table 6 Comparison of spectra from asphyxiated infants with age-matched uncomplicated infants
Full size table
Figure 5
figure 5

Mean power spectra of CBFV grouped according to complications. Table 6 gives statistical results. The LF range between 0.02 and 0.08 Hz is shown.

Full size image

On the HR spectra, asphyxia produced complementary changes as neither the mean value nor the LF power distribution were significantly different between the two groups. In this case the total HR power is considerably reduced in the group with asphyxia when compared with the uncomplicated group(Table 6).

DISCUSSION

The present study confirmed our original hypothesis that the variability of neonatal CBFV, as described by spectral analysis, contributes different information from that carried by the HR and BP power spectra. This conclusion follows from estimations of the global average power spectra of a large number of records, as well as from an analysis of the effects of GA, PA, and asphyxia on spectral parameters.

Although considerable variability exists between individual spectra, by averaging the spectra we have been able to determine a characteristic mean CBFV spectral pattern. Our results have shown that the CBFV spectrum averaged over a large number of data records has a generally similar spectral pattern to the averaged BP and HR spectra, with a large LF power content and very little spectral power above approximately 0.1 Hz (Fig. 1). A more detailed statistical investigation, however, showed the CBFV spectral pattern over the LF band to be significantly different from both the BP and HR spectral patterns. This is an important observation, because it suggests that CBFV LF oscillations do not simply reflect oscillations in BP or HR, unless their effect is manifested through a nonlinear mechanism.

An increasing number of biologic and physiologic phenomena have been shown to present the properties of chaotic systems(25). Such systems usually display a 1/fα spectrum(15), and we tested the hypothesis of whether CBFV might present this kind of behavior. CBFV recordings show many properties of random signals, such as large short-term fluctuations as well as spectral broadness. On the other hand, our conceptual models usually assume that CBFV is controlled within tight limits by autoregulation(26). Chaotic systems are neither random nor deterministic, but their behavior shows characteristics of both(25), and this could explain the dual behavior of CBFV. In addition, chaotic systems are usually described by nonlinear dynamic equations, relying on a relatively small number of parameters(25), and thus would lead to distinctly different mathematical models when compared with the classical multivariate linear approach(26).

1/fα spectral analysis of neonatal CBFV, HR, and BP has not been reported previously. Over the whole frequency range available (0.005-0.5 Hz) the corresponding values of α were 1.51 for CBFV, 1.73 for BP, and 1.21 for HR. These values are significantly different and confirm that CBFV has a different spectral pattern from those of BP and HR. Values of α for CBFV for adults could not be found in the literature. For HR our value ofα is close to that reported for adults(17, 27, 28). In the case of BP, however, we obtained α = 1.73, whereas previous studies in adults have obtained values around 1.2(16, 27) and 1.8(29). However, a direct comparison with previously reported values of α is not justified, because these previous studies were performed on adults. Moreover, they were performed over 24-h periods, and as a result, the frequency range studied differs significantly from ours, in most cases being <0.001 Hz.

A major difference between our results and those reported previously is the finding that the linear fitting on the log-log scale has significantly correlated residuals as determined by the Durbin-Watson test. Breaking the linear regressions into two frequency ranges (Fig. 2) has improved the linear fitting, although only CBFV and BP have shown independent residuals in the range 0.005-0.06 Hz. Overlooking this limitation, the corresponding values of α are not significantly different for the three variables in the range 0.005-0.06 Hz but are so in the range 0.06-0.5 Hz. Close inspection of the results published by other investigators show a similar departure from a single linear relation on the log-log spectral plot of HR(28, 30) and BP(16). Shono et al.(31) reported a change in slope(increase in α) at a frequency between 0.01 and 0.1 Hz in neonatal (and fetal) HR. Kobayashi and Musha(28) showed a similar phenomenon in adult HR. Further work, involving longer records and different populations, is required to determine whether CBFV should be regarded as the output of a chaotic system to guide future research into the determinants of CBFV variability.

The effects of GA, PA, and perinatal asphyxia on the CBFV spectra allow some further insight into its relationship to the HR and BP spectra. Hitherto, neonatal HR variability has been the most studied of these three signals, and our results are in broad agreement with previous investigations that have also shown increases in HR variability with GA and PA(8, 32–36). As discussed by Chatow et al.(8), these changes probably reflect autonomic nervous system maturation of the premature newborn. Moreover, the striking reduction in total spectral power, which we observed for newborns with perinatal asphyxia, is in agreement with the findings of Divon et al.(37), although those investigators have considered only the frequency band around the breathing rate. The mechanism for the reduced variability in asphyxia is thought to involve the removal of influence from higher centers on cardiorespiratory control. An intact, well oxygenated CNS is essential for normal variability, and hypoxia due to severe respiratory distress syndrome also reduced long-term HR variability in preterm infants(33, 35, 38, 39).

A lot less is known about the variability and frequency spectra of arterial BP in premature and term newborns. Mean arterial BP has been shown to increase with GA and PA(40–42). This trend is reflected in our data, although differences were not significant and it was not possible to record the BP in the term group. Perlman et al.(43) associated the increased variability of CBFV, which might be implicated in the occurrence of IVH, to an increased variability in BP. However, Miall-Allen et al.(44) have calculated the CV for intravascular beat-to-beat measurements of systolic pressure and found that the risk of IVH was inversely related to the amount of time a newborn presents CV > 5%. Because the total power figures presented in Tables 4 and 5 are a close approximation to the signal variance, it is possible to calculate the CV for our data, keeping in mind that these correspond to mean BP rather than the systolic values adopted by Miall-Allen et al.(44). The minimum value we found was 2.1% for the first 12 h after birth, this value rising and staying above 3.45% after 12 h PA. As for GA, values of 3.0% or less are obtained for the two first subgroups in Table 4 (24-31 wk), reaching 3.96% for 32-35-wk GA. If the interpretation of Miall-Allen et al.(44) is correct, then the subgroups at highest risk for IVH in our study would be the more premature neonates(24-31-wk GA) in the first 12 h after birth, in agreement with clinical observation studies.

Tables 4 and 5 also show significant changes in total power for BP with both GA and PA and significant changes in LF spectral pattern with GA. These changes mimic those observed in HR variability, as discussed above, but its causality cannot be ascertained because of the closed-loop structure represented by the baroreflex and the hemodynamic feedback of HR on BP(45). On the other hand, it is important to characterize the variability of BP as a first step toward a better understanding of CBFV variability. In the absence of cerebral autoregulation, changes in mean arterial BP will be reflected as changes, and hence variability, of CBFV. We have shown previously(46) that with an intact autoregulation CBFV responds to transient changes in BP within 1-2 s. As a first approximation this autoregulatory response can be modeled as a high pass frequency filter(47), and our results then suggest a cutoff frequency somewhere above 0.3 Hz. As a consequence of the finding that most of the CBFV spectra is restricted to frequencies <0.3 Hz (Fig. 1), the total power expressed in Tables 4–6 corresponds to the frequency range where we could expect autoregulation to attenuate the variability of arterial BP that is transmitted to the cerebral circulation. Therefore, the total power ratio between CBFV and BP represents an approximate indicator of autoregulatory performance. This ratio is approximately 1.5 for the first two subgroups of GA in Table 4 (24-27 and 28-31 wk), but falls to 1.07 for the 32-35-wk subgroup. As for the effect of PA, the CBFV/BP power ratio is 1.41 for the period 0-12 h and falls to the range 1.04-1.11 for the other three subgroups considered. The relative values of these power ratios are suggestive of a less efficient autoregulation in the premature newborn (24-31-wk GA) as well as in the first 12 h after birth. An alternative interpretation would be an increased CBFV variability in these subgroups of GA and PA from sources not related to the BP variability(7).

Limited information is still available about the origins of the slow (LF) oscillations frequently observed in CBFV recordings(7, 9). Our results reject the hypothesis that these oscillations are merely a reflection of the LF variability of HR and BP. This conclusion follows from the spectral patterns (e.g.Figs. 1 and 3) and the corresponding χ2 tests (Tables 2, 4, and 5) showing different behavior of the LF spectral pattern for the three spectra. In the analysis of HR variability, it is customary to examine the relative power content of different spectral bands as we have attempted to do with the LF ratio. However, we found the interpretation of this parameter fraught with problems arising from concomitant changes in the power content of the very LF harmonics(<0.02 Hz), e.g. Figs. 3–5; or in the high frequency band (>0.08 Hz). In addition, changes in spectral power within the LF band are not reflected by the LF ratio. For these reasons we resorted to the χ2 test, which we found to be the most appropriate measure to characterize differences in shape between two different spectral samples(23). The χ2 test has detected significant changes in LF spectral pattern between the average spectra of HR or BP in relation to that of CBFV (Fig. 1,Table 2). For the case of CBFV, Figure 4 shows that the LF spectral band is relatively more prominent for the lower PA subgroups, suggesting the presence of slow oscillations whose incidence is likely to be reduced at later periods as previously observed by Coughtrey et al.(7). A similar observation applies to the most premature group in Figure 3 (24-27-wk GA), but in this case theχ2 test did not show a significant difference. This result is in agreement with the findings of Michel et al.(48) showing a lack of correlation between the occurrence of slow oscillations and GA.

By matching a subgroup of term babies to the asphyxiated group we have attempted to control for the effects of GA and PA on the CBFV and HR spectra(Table 6). We confirmed previous observations(49, 50) that mean CBFV (and perhaps CBF) is considerably elevated in newborns suffering from perinatal asphyxia. This result has been interpreted by previous investigators as a sign of vasoparalysis or exhaustion of cerebral autoregulation. One original finding of the present study, though, is that the mean CBFV power spectra shows differences between asphyxiated neonates and normal babies (Fig. 5), with a highly significant difference in the LF frequency band (Table 6). The LF ratio in Table 6 and the LF band pattern in Fig. 5 suggest that LF oscillations would be more readily apparent in the normal group than in the asphyxiated one. This observation gives credence to the view that LF oscillations are a reflection of a normal autoregulatory system in the newborn(48). Whether the absence of oscillations or a shift toward the very LF region of the spectra (Fig. 5) is an indication of autoregulatory failure remains to be studied.

As a conclusion, we have shown that the frequency spectra of beat-to-beat changes in CBFV provides additional information about pathophysiologic states of the neonatal cerebral circulation to that which can be extracted from HR or BP spectra or standard estimations of mean CBFV. Further work is required to assess the clinical value of spectral analysis of CBFV in neonatal care, ideally using longer recordings to improve the resolution of LF variability. More studies are also needed of BP variability in the newborn to shed light on its role in the etiology of IVH and to establish its contribution to CBFV variability.