Cerebral Autoregulation Dynamics in Premature Newborns
Background and Purpose Autoregulation of cerebral blood flow is easily disrupted, and loss of this normal physiological reflex may worsen the neurological outcome for patients undergoing intensive care. We studied the response of cerebral blood flow velocity to changes in mean arterial blood pressure.
Methods Cerebral blood flow velocity was measured with Doppler ultrasonography in one middle cerebral artery for 5-minute periods in 33 babies of gestational age <33 weeks admitted to a neonatal intensive care unit. Two methods of evaluating autoregulation were developed. The first used linear regression analysis of blood flow velocity on blood pressure. Records were classified as showing loss of autoregulation if the regression slope was greater than a critical value. A minimum change in mean arterial blood pressure of 5 mm Hg and a critical slope of 1.5 %/mm Hg were found to be adequate criteria for the classification of records by the regression method. The second method used coherent averaging, a technique similar to that used in recording evoked potentials. Spontaneous transient increases in blood pressure were automatically detected, and the instant corresponding to its maximum rate of rise was used to synchronize averages of the blood pressure and blood velocity transients. The resulting coherent averages were classified into two groups based on the morphology of the cerebral blood flow velocity average.
Results Whereas the regression method allowed the classification of only 51 of 106 records, the coherent average method classified 101 of 106 (95.3%) of the records available. For 51 records that were classified by both methods, there was agreement in 42 cases (82.3%). The coherent average of all records classified as having an active autoregulation showed cerebral blood flow velocity returning to baseline much earlier than blood pressure, suggesting that autoregulation was taking place within 1 to 2 seconds. This pattern was absent in records in which autoregulation was classified as absent.
Conclusions Computerized coherent averaging of the cerebral blood flow velocity response to spontaneous blood pressure transients offers a promising new method for noninvasive bedside assessment of autoregulation in patients undergoing intensive care. The time course for autoregulation, when present, is in agreement with that reported in adults.
Although little is known about the ability of the newborn to regulate cerebral blood flow (CBF) in response to changes in perfusion pressure, there is some evidence that the autoregulatory range is limited compared with that of adults1 2 and that the response is absent altogether in sick infants.3 4 5 6 Preterm infants are particularly vulnerable to both ischemic and hemorrhagic brain injury, which are associated with an increased risk of neurodevelopmental handicap. This fact is often explained by invoking the existence of a pressure-passive cerebral circulation unable to compensate for the changes in arterial blood pressure (BP) that occur frequently during intensive care, despite the paucity of information on the topic.
A reliable, repeatable method to characterize autoregulation would allow further work in this important area. Previous studies have measured CBF with radioactive xenon3 6 or jugular occlusion plethysmography,4 but these methods are disruptive to the infant, cannot be repeated frequently, and are not suitable for study of the speed of response. Analysis of change in CBF velocity (CBFV) in response to change in mean arterial BP (MABP) occurring during clinical deterioration5 or tilting7 seemed promising, but it can only be carried out at irregular intervals, and the stimuli may initiate an autonomic nervous system response and shift the autoregulatory plateau.8 Ahmann et al9 obtained values for CBFV and BP from observations made many hours apart, and the duration of the recordings was not sufficient to ensure that the results were free from inaccuracies in CBFV estimation introduced by the presence of slow cycling.10 11
In adults, Aaslid et al12 developed an elegant method to study the dynamics of autoregulation, using a step change in arterial pressure produced by deflating leg cuffs in healthy volunteers. This approach allowed repeated measurements and provided valuable information about the dynamics of autoregulation. However, deflating a leg cuff does not produce a significant change in the BP of a newborn baby and might not be practical in an intensive care setting. We studied the feasibility of obtaining a measure of autoregulation using spontaneous small changes in BP. Concomitant changes in CBF were estimated by measurements of CBFV with Doppler ultrasonography in one middle cerebral artery. Because the changes in BP and CBFV were of small amplitude, we resorted to coherent averaging to improve the signal-to-noise ratio, a method analogous to that used in the detection of brain evoked potentials.
Coherent averaging has been applied in many different areas to enhance the detection of waveform patterns buried in high-amplitude noise. In medicine, it has found important applications in the detection of visual and auditory evoked responses.13 In this case, a number of segments of an electroencephalogram are averaged in phase with the auditory or visual stimulus. In other applications, averaging can be synchronized by naturally occurring events such as the R wave of the electrocardiogram or by particular features of the signal. If the noise is random and uncorrelated with the signal of interest, it is possible to demonstrate that coherent averaging reduces the variance of the noise by a factor of n, which is exactly the number of transients used in the average.13 Coherent averaging has been previously applied to CBFV recordings from the posterior cerebral artery to detect the cortical blood flow response to visual stimulation.14
Subjects and Methods
We studied a cohort of 33 infants of gestational age <33 weeks admitted consecutively to the Cambridge neonatal intensive care unit. Infants were excluded only if they had a lethal malformation or were admitted more than 12 hours after birth. The study was approved by the Cambridge District Ethical Committee. Informed consent was obtained from one or both parents.
CBFV was measured using the semicontinuous system described by Evans et al.15 A small (0.5-cm diameter) continuous-wave Doppler ultrasound probe of power intensity <50mW · cm−2 was securely fixed to the skin overlying one middle cerebral artery using Dalzofoam (Seton). BP was measured continuously through an umbilical or peripheral arterial catheter using a P23 Spectromed transducer connected to a Simonsen and Weel Quadriscope monitor. Simultaneous recordings of CBFV (peak velocity envelope) and BP were obtained for at least 5 minutes and stored on a digital instrumentation tape recorder (PC-108M, Sony). Recordings were made at <6, 12, and 24 hours and on subsequent days if the infant had a functioning arterial catheter.
Recorded signals were low-pass-filtered at 30 Hz and converted to digital format at a rate of 200 samples per second on a microcomputer. Narrow spikes in the CBFV signal were detected and removed by linear interpolation. Both CBFV and BP signals were low-pass filtered with an eighth-order Butterworth zero-phase digital filter with a cutoff frequency of 20 Hz. The filtered BP signal was used to estimate the RR interval and to mark the beginning and the end of each cardiac cycle.
CBFV and BP data from individual transient BP peaks were averaged together to produce a coherent average response.13 To compute coherent averages, the mean values of CBFV and BP were calculated for each cardiac cycle of the recording. The resulting beat-to-beat sequences of mean CBFV and BP values were interpolated with a third-order polynomial and resampled with an interval of 0.2 seconds to produce signals with a uniform time axis. The position of peaks in the resampled BP signal was automatically detected, and the foot of each peak was also detected by a foot-seeking algorithm.16 Peaks were only accepted if they were at least 6 seconds apart and if their relative amplitude (peak to foot) was ≥2% of the baseline value. The largest peaks (up to a maximum number of 25) were detected for each record. The position of the maximum derivative of each BP peak was used as the point of synchronism for coherent averaging. Although coherent averaging reduces the influence of random noise on the estimated CBFV response to a transient peak in BP, it is well known that CBFV recordings show low-frequency oscillations and other large-amplitude artifacts that are not random.10 11 17 To remove these interferences, signals were accepted into the average only if CBFV and BP had a correlation coefficient of r>.3 for the 3 seconds preceding the point of synchronism. To test whether the results of averaging could be due to artifact, averages were also obtained using random alignment between the pressure peaks and the CBFV signal, using the same number of waveforms.
Two distinct methods were used to classify each individual record with respect to the presence or absence of autoregulation. In the first method, the 5-minute recording was split into contiguous 8-second intervals, and the mean values of CBFV and BP were calculated for each interval. For recordings with mean BP changes greater than the minimum pressure change (ΔPmin), a linear regression of CBFV on BP was performed using BP as the independent variable. Recordings were classified as showing absence of autoregulation (group B) if the regression had a slope significantly greater than the critical minimum slope (SLmin). Otherwise, the recording was classified as showing the presence of autoregulation (group A). This criterion implies that regressions that are not statistically significant (P<.05) will be classified as showing an active autoregulation independent of the slope value. Reference values for ΔPmin and SLmin are 5 mm Hg9 and 0.5 %/mm Hg,5 respectively. The effect of different choices of ΔPmin and SLmin on the classification based on linear regression was studied by varying ΔPmin between 3 and 7 mm Hg and SLmin between 0 and 2%/mm Hg.
The second classification of individual records was based on the morphology of the CBFV coherent average. The total sample of records was randomly split into two groups, and the coherent average of each group was computed. Using the correlation coefficient between the CBFV average of each record and the group coherent average (for the 10-second period that follows the foot of the BP transient), individual records were reallocated to the group corresponding to the highest correlation. A new coherent average was computed for the two groups, and the process was repeated until there were no more transitions between the two groups.
A contingency table was used to compare the results of the two independent methods of classification, and the degree of agreement was tested with Cohen’s κ.18 Differences between clinical variables were assessed with the t test. A value of P≤.05 was adopted as the criterion for statistical significance.
Thirty-three infants of gestational age <33 weeks were admitted during the study period, and 120 5-minute recordings were made. No parents refused to give permission for their infant to be studied, and the small amount of handling involved in making the recordings was well tolerated by all subjects. Fourteen records were discarded because of damped BP signals or excessive movement artifact, leaving 106 suitable records for further analysis.
A record with large BP transients is plotted in Fig 1A⇓ and 1B, with arrows indicating the points of synchronism detected for coherent averaging. A record with smaller BP transients but with a spontaneous large fall in mean BP followed by a similar change in mean CBFV is presented in Fig 1C⇓ and 1D⇓. This record was classified as group B (absence of autoregulation) by the linear regression method.
With a reference value of ΔPmin=5 mm Hg, only 51 of the 106 records could be used for the classification based on linear regression because 55 did not have a change in mean BP of >5 mm Hg. With a reference value of SLmin=0.5 %/mm Hg, 18 records from 16 infants were considered to show autoregulation because the slope of the regression line was not significant (group A, Fig 2A⇓). Fig 2B⇓ shows the regression lines for 33 records from 22 infants that had a significant slope greater than SLmin (0.5 %/mm Hg) and were classified as showing an absence of autoregulation. Eleven babies had recordings in both groups. Increases in SLmin led to a reduction in the fraction of records classified as showing an absence of autoregulation. Similarly, changes in ΔPmin determined the total number of records that could be used for the classification based on linear regression. The effects of changes in ΔPmin and SLmin on the linear regression classification are shown in Table 1⇓.
Classification based on the morphology of the CBFV average made use of 101 of the 106 individual records available. One record did not have BP transients with amplitude >2%, and four records did not have transients satisfying the r>.3 condition (see “Methods”). The final coherent averages for CBFV and BP for the two groups separated by the algorithm described above are represented in Fig 3⇓. The time course of the change in BP was similar for the two groups, but there was a marked difference in the temporal pattern of CBFV response. While the CBFV waveform represented in Fig 3A⇓ returned to baseline values before the BP pulse, the group B average continued to increase after the BP average returned to its baseline. We associate the waveform of Fig 3A⇓ with the presence of autoregulation and the one in Fig 3B⇓ with loss of autoregulation. The first group included 60 records from 28 patients; the second included 41 records from 24 patients. Nineteen patients had records in both groups.
The number of records available for comparison of the two independent methods of classification was restricted by the condition involving ΔPmin (Table 1⇑). For the reference condition (ΔPmin=5 mm Hg, SLmin=0.5 %/mm Hg), there was agreement between the two methods for 37 records of a total of 51, with 18 in group A and 19 in group B. This leads to a coefficient of agreement (Cohen’s κ) of 0.489, which is highly significant (P<.001). The corresponding values of κ for other values of ΔPmin and SLmin are given in Table 1⇑. All these values are highly significant (P<.005). In Table 1⇑ it can be observed that for any value of ΔPmin the highest values of κ are obtained for SLmin=1.5 %/mm Hg. In particular, for ΔPmin=5 mm Hg there are 42 correct classifications for the 51 records (82.3 %) corresponding to a κ of 0.649.
By adopting ΔPmin=5 mm Hg and SLmin=1.5 %/mm Hg, the coherent averages of CBFV and BP can be calculated for the group A and B records, as classified by the linear regression method, and are plotted in Fig 4⇓. These waveforms are strikingly similar to the ones depicted in Fig 3⇑, which resulted from the coherent average classification. Furthermore, these patterns of CBFV disappear when random alignment is adopted between the BP and CBFV transients.
With values of ΔPmin=5 and SLmin=1.5 %/mm Hg maintained for the classification based on linear regression, analysis of the clinical variables recorded prospectively during the study showed that mean BP was significantly lower in group B (absence of autoregulation) when compared with the mean of group A (P<.005, Table 2⇓). Gestational age and Po2 were also significantly different in the two groups (P=.04). Table 2⇓ also shows that ΔP, the maximum pressure change in the regression data, was significantly greater in group A compared with group B. As expected from the decision criteria adopted to separate the two groups, the mean normalized regression slope of the group B records was much higher than the mean of those from group A (3.26 versus 0.87 %/mm Hg).
The occurrence of significant changes in CBF (or CBFV) produced by changes in MABP has been used previously as an indication of an impaired cerebral autoregulation.3 4 5 6 9 Unfortunately, different authors used different criteria to assess the change in CBF and CBFV; as a result, no standardized procedures exist to evaluate the status of autoregulation in either adults or infants. Some studies used only two pressure measurements to assess the corresponding change in CBF.1 4 9 Linear regression analysis, using multiple measurements of CBF (or CBFV) and MABP, has been preferred by others,3 5 6 19 and we followed this approach because of its statistical superiority. Nevertheless, the critical regression slope that should be adopted to reject the null hypothesis of an intact autoregulation in neonates remains to be defined. Jorch and Jorch5 selected a value of 0.50 %/mm Hg as the threshold to classify newborns with impaired autoregulation. The observational study of Pryds et al6 gave a mean slope of 0.80 %/mm Hg for the slope of the CBF-MABP regression with a 95% confidence interval of −0.6 to 2.0 %/mm Hg. In the same study, a subgroup of patients with severe intraventricular hemorrhage had a mean slope of 4.0 %/mm Hg, which is suggestive of an impaired autoregulation. Another observational study yielded slopes in the range of −1.66 to 2.6 %/mm Hg with a mean value of 1.0 %/mm Hg for regressions between CBFV in the carotid artery and MABP.19
Because of the uncertainty regarding the critical slope of the CBFV-MABP linear regression, we conducted a sensitivity analysis, also taking into consideration the minimum change in MABP (ΔPmin) that should be required to accept a certain regression. Our results indicate that the classification of intact versus impaired autoregulation is not affected for slopes ranging from 0.0 to 1.0 %/mm Hg (Table 1⇑). However, there is a marked change in the number of records classified as showing absence of autoregulation for slopes ≥1.5 %/mm Hg. This value seems to be in good agreement with the studies mentioned above. In particular, the study of Menke et al,19 which was based on a group of 16 neonates with gestational ages similar to those of our population, would have classified 4 of the 16 as having an impaired autoregulation if the value of 1.5 %/mm Hg was used as a threshold. It should be emphasized that slope values are meaningless for regressions that are not significant. In this case, it is not possible to characterize the absence of autoregulation, and the corresponding records are more appropriately classified as showing an active autoregulation.
Most studies using linear regression to classify the status of autoregulation do not specify the minimum value of ΔP that was adopted to accept a certain segment of data into the regression analysis. Ahmann et al9 adopted ΔPmin=5 mm Hg, and we have looked into the effect of changes in ΔPmin around this value. Menke et al19 introduced a similar condition, accepting only records with a coefficient of variation of 5% or more for MABP. In studies involving spontaneous changes in BP, the choice of ΔPmin can have a major influence: low values of this parameter lead to unreliable estimates for the slope, while high values can drastically reduce the number of records available for analysis. This trade-off is reflected in our Table 1⇑. For a value of ΔPmin=3 mm Hg, 77 records can be used in the regression analysis, but the agreement with the coherent-average classification (κ coefficient) is the lowest, independent of SLmin. Increasing ΔPmin increases κ. With the exception of the case of SLmin=1.5 %/mm Hg, κ is maximum for ΔPmin=5 mm Hg, and we suggest that this value be adopted for any future similar work.
Application of coherent averaging of CBFV and BP to automatically detected BP transients produced encouraging results, particularly as some recordings yielded only a few spontaneous BP transients for averaging. The method allowed the classification of individual recordings as showing the presence or absence of autoregulation, and the classification thus obtained was in good agreement with the alternative method using linear regression of CBFV on BP. Furthermore, coherent averaging allowed the time course of CBFV response to be studied. Figs 3⇑ and 4⇑ suggest onset of autoregulation within 1 to 2 seconds in this group of preterm neonates. There are very few previous reports of the dynamics of autoregulation in infants. Anthony et al7 observed a biphasic CBFV response to tilt and associated this with the presence of autoregulation. However, they have not given an indication of the time course of the autoregulatory response. Our observation of a relatively fast autoregulatory response is in general agreement with the results obtained by Aaslid et al12 in adults and from animal studies.20 21 22 Precise comparison of different studies is difficult; none of the Doppler CBFV methods represents a “gold standard” for comparison, although we chose linear regression because of the existence of previously published work in the newborn. Aaslid et al12 studied a negative step change in BP, whereas our positive changes were not perfect steps. However, there is still very good agreement between the results. Coherent averaging has several potential advantages: the method seems likely to give a result on a much larger number of recordings than linear regression, which requires a relatively large change in BP before it can be applied, and coherent averaging gives additional information about the time course. Furthermore, the use of coherent averaging eliminates interference produced by the background oscillations in CBFV, which we and others have described.10 11 As a limitation, however, coherent averaging cannot be used in recordings without spontaneous BP transients.
The possibility that the results obtained were artifacts generated by the averaging technique is unlikely. Similar distinct patterns could be observed in single BP transients of large amplitude in recordings without large background oscillations as shown in Fig 1A⇑ and 1B⇑. Unfortunately, direct analysis of single peaks could not be generalized because the CBFV beat-to-beat signal presents considerable variability, which frequently dominates the temporal pattern and obscures the autoregulatory response to small BP changes.10 11 17 To reject large oscillations of this type, we computed the correlation coefficient between BP and CBFV in the 3 seconds preceding the point of synchronism (t=0 in Fig 3⇑) and only accepted transients with r>.3 into the average. This condition for r was particularly important when less than 30 waveforms were available for the coherent average. Since slow cycling of CBFV is not strictly random noise, it is possible that it still affects the final CBFV coherent average. In Fig 3B⇑ (and to a minor extent Fig 4B⇑ also), the CBFV average shows a secondary rise for t >5 seconds despite the fact that the BP average is returning to baseline. The reasons for this delayed increase in CBFV are not clear, although they have also been observed in some individual transients. One possibility could be the superposition of BP peaks separated by a short interval (≈6 seconds), but we recalculated the coherent averages, rejecting all peaks separated by intervals of 12 seconds or less, and patterns very similar to those in Figs 3⇑ and 4⇑ are still observed. Finally, the results obtained with random alignment rather than those that used the position of the maximum derivative for synchronization confirmed that the responses seen in Figs 3⇑ and 4⇑ are a true reflection of the dynamic relationship between CBFV and BP.
Other factors that could have influenced the results described in the previous section, particularly the agreement between the two independent methods of classification, could not be identified by the analysis of the sample studied (Table 2⇑). The lower MABP and gestational age observed for the group B records support the view that prematurity might be a primary cause of loss of autoregulation.5 7 MABP is known to increase with birth weight,23 and for the 51 recordings of Table 2⇑ there is a very significant correlation between MABP and gestational age (r=.425, P=.002). Furthermore, a plot of MABP versus gestational age (not shown here) indicates clustering of group B records for low values of MABP and gestational age, partly explaining the differences expressed in Table 2⇑. For MABP <40 mm Hg, there are 10 records in group B and only 2 in group A (χ2=6.6, P=.01). It is possible that the lower limit of autoregulation is closer to 40 mm Hg for these patients than to the value of 30 mm Hg as previously estimated by De Bor and Walther.1 The difference observed for Po2 in Table 2⇑, however, is more difficult to explain. We believe it results from the occurrence of an outlying value of 150 mm Hg, which is more than 3 SD above the mean of group B. If this value is removed, the t test result changes from P=.045 to P=.09, and the difference observed for Po2 ceases to be significant. Finally, the observation that ΔP is higher in group A compared with group B (Table 2⇑) suggests that the regression method might have a bias toward classifying records with smaller values of ΔP as having a loss of autoregulation. However, the results in Table 1⇑ show that this is not the case because values of ΔPmin of 3 mm Hg give a relative frequency of classification for group B that is lower or approximately the same as those given by ΔPmin of 7 mm Hg.
In summary, there are no “gold standards” for classification of cerebral autoregulation. The approach most commonly seen in the literature is an assessment of the CBF and CBFV changes that follow changes in MABP. This “static” method intrinsically assumes that if autoregulation is active then the CBF-MABP relation is described by the classic autoregulation curve with a plateau of CBF for a wide range of MABP. This method has been used with very wide criteria and, to our knowledge, we are the first group to carry out a thorough sensitivity analysis of the influence of ΔPmin and SLmin. On the other hand, Aaslid et al12 have shown that in normal subjects a step change in MABP leads to a CBFV transient followed by fast recovery to baseline value. They have assumed that this pattern characterizes the dynamics of autoregulation, and we have extended this concept to the neonatal circulation using coherent averaging of spontaneous transients of MABP. Because of multiple problems associated with the practicality and reproducibility of classification of autoregulation based on steady-state changes in MABP (static method), it is more likely that in the future a standard method of classification of autoregulation will be found with the dynamic approach, as proposed by Aaslid and ourselves, than with the less controllable static method.
The fact that Doppler ultrasonography does not give a direct measurement of absolute flow but gives the mean spatial velocity (which is equal to flow divided by cross-sectional area) has to be considered in any situation where changes in cross-sectional area might affect the conclusions of a study. Aaslid et al12 have addressed this problem by analyzing the instantaneous spectral power of the reflected Doppler signal, which is proportional to the number of red cells scattering ultrasound and, consequently, proportional to the cross-sectional area. Their conclusion is that the adult middle cerebral artery cross-sectional area does not change significantly during a step decrease of 20 mm Hg in MABP. No similar investigation has been performed in the case of neonates. For the hypothetical situation in which the diameter of the middle cerebral artery is changing in response to BP changes, the speed of such changes has to be considered. If the cross-sectional area were to change with the same speed as the coherent averages in Figs 3⇑ and 4⇑, it is unlikely that these changes would be such as to exactly cancel the observed differences between parts A and B of those figures. On the other hand, for slower changes in diameter, the observed patterns of CBFV would be expected to remain unchanged. In this case, the regression method would be likely to lead to the wrong classification, since mean CBFV and MABP measurements are taken during a 5-minute recording, and changes in diameter throughout the recording period could influence the regression significance and slope value.
There are many situations other than neonatal intensive care in which a bedside technique for assessing autoregulatory capacity would be useful, such as in the care of adults and children with head injury and patients suffering from meningitis, intracranial hemorrhage, and stroke. As yet, we have not applied the method to enough newborns to realize its full potential, but these preliminary results demonstrated that more of the recordings classified as showing absence of autoregulation came from less mature infants and from those with lower MABP. In several babies, the ability to autoregulate came and went at different times; further research may reveal the reasons for this. Possibilities include drug therapy such as morphine or pancuronium, changes in the baseline cerebral perfusion pressure, and recent hypoxia or hypothermia, all of which have been shown to disturb autoregulation in animals. We feel that computerized coherent averaging of CBFV in response to BP transients offers clear advantages over alternative methods for the study of cerebral autoregulation, the most exciting being the ability to characterize the dynamics of autoregulation repeatedly in individual patients.
A.W.R.K. is supported by Action Research. We would like to thank the parents of babies under our care for consenting to the monitoring of their infants.
- Received May 9, 1994.
- Revision received September 12, 1994.
- Accepted September 12, 1994.
- Copyright © 1995 by American Heart Association
Anthony MY, Evans DH, Levene MI. Neonatal cerebral blood flow velocity responses to changes in posture. Arch Dis Child. 1993;69:304-308.
Dellagrammaticas HD, Kapetanakis J, Papadimitriou M, Kowakis G. Effect of body tilting on physiological functions in stable very low birthweight neonates. Arch Dis Child. 1991;66:429-432.
Ahmann PA, Dykes FD, Lazzara A, Holt PJ, Giddens DP, Carrigan TA. Relationship between pressure passivity and subependymal/intraventricular haemorrhage as assessed by pulsed Doppler ultrasound. Pediatrics. 1983;72:665-668.
Anthony MY, Evans DH, Levene MI. Cyclical variations in cerebral blood flow velocity. Arch Dis Child. 1991;66:12-16.
Coughtrey H, Rennie JM, Evans DH. Postnatal evolution of slow variability in cerebral blood flow velocity. Arch Dis Child. 1992;67:412-415.
Aaslid R, Lindegaard KF, Soteberg W, Nornes H. Cerebral autoregulation dynamics in humans. Stroke. 1989;20:45-52.
Aaslid R. Visually evoked dynamic blood flow response of the human cerebral circulation. Stroke. 1987;18:771-775.
Coughtrey H, Rennie JM, Evans DH, Cole TJ. Factors associated with respiration induced variability in cerebral blood flow velocity. Arch Dis Child. 1993;68:312-316.
Symon L, Held K, Dorsch NWC. A study of regional autoregulation in the cerebral circulation to increased perfusion pressure in normocapnia and hypercapnia. Stroke. 1973;4:139-147.
Kontos HA, Wei EP, Navari RM, Levasseur JE, Rosenblum WI, Patterson JL. Responses of cerebral arteries and arterioles to acute hypotension and hypertension. Am J Physiol. 1978;234:H371-H383.
Versmold HT, Kitterman JA, Phibbs RH, Gregory GA, Tooley WH. Aortic blood pressure during the first 12 hours of life in infants with birth weight 610 to 4,220 grams. Pediatrics. 1981;67:607-613.