Keywords

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Introduction

In this study in preterm infants, NIRS is used to measure cerebral autoregulation over long periods, reflecting static autoregulation. The use of NIRS for this purpose was described by Tsuji et al. [1]. Changes in HbD, rSO2, or TOI reflect changes in CBF and the correlation between MABP and HbD/rSO2/TOI is a reflection of autoregulation. A good correlation was found between autoregulation and outcome, i.e. frequency of severe intraventricular bleeds. rSO2 and TOI are both absolute values, they are less prone to movement artefacts than HbD, and easier to measure in clinical practice.

We studied the concordance between HbD/rSO2/TOI and MABP by means of the coherence (COH, measuring the degree of linear dependence between the frequency spectra of two signals), and partial coherence (PCOH) [2] coefficients. The latter allows the elimination of the linear influence of one signal on another one. This means that, in contrast to COH, PCOH can also be applied in periods of fluctuating SaO2 thereby improving the automation and the use of the method. We developed and investigated four PCOH algorithms fixing the physiological interactions between SaO2, MABP, and HbD/rSO2/TOI. We studied the PCOH properties, in particular during periods of fluctuating SaO2. For this purpose we used parameters that synthesize patient level of autoregulation: the mean score (mCOH and mPCOH) [1], the pressure-passive index (PPI) [3] and the critical percentage of the recording time (CPRT) [4]. Finally, we compared these parameters with the infant clinical characteristics and outcomes: infant post menstrual age (PMA, in weeks), birth weight (BW, in g), Bayley's psychomotor (PDI) and mental (MDI) developmental indices after 9, 18, and 24 months (for the Leuven, Zurich, and Utrecht data respectively), Griffith developmental index (combination of a mental and psychomotor test) after 24 months, and APGAR score at birth and 5 min after birth.

2 Datasets

Thirty premature infants with need for intensive care were monitored, of whom 10 were from the University Hospital Zurich (Switzerland), 10 from the University Medical Centre Utrecht (The Netherlands), and 10 from the University Hospital Leuven (Belgium). SaO2 was measured continuously by pulse oximetry, and MABP by an indwelling arterial catheter. Transcranial NIRS signals HbD (measured by the Critikon Cerebral Oxygenation Monitor 2001), rSO2 (INVOS4100, Somanetics Corp.), and TOI (NIRO300, Hamamatsu) were measured for non-invasive monitoring of cerebral oxygenation. The signals were measured simultaneously in the first days of life. For the Zurich data, the babies were characterized by a mean PMA of 28 1/7 weeks (std = 2 1/7) and a mean BW of 1198 g (std = 439). For the Utrecht data, the babies were characterized by a mean PMA of 29 2/7 weeks (std = 1 2/7) and a mean BW of 1130.67 g (std = 311.36). For the Leuven data, the babies were characterized by a mean PMA of 28 5/7 weeks (std = 3 2/7) and a mean BW of 1125 g (std = 503.76). HbD, rSO2, and TOI were recorded digitally on a personal computer at a sampling frequency of 1.677, 1, and 10 Hz for the Zurich, Utrecht, and Leuven data respectively. Afterwards the signals from all datasets were down-sampled to the smallest frequency multiple of the recording frequencies i.e. 0.333 Hz (periodicity: 3 s) to ensure comparability. A pre-processing algorithm was applied to the recordings of all centres to remove signal artefacts. Each artefact point was simply deleted from the recording for each signal [3]. Among the pre-processing operations, we kept all variables within normal ranges, in particular SaO2 in the range 80–100%.

3 Methods

Since the possible concordance between MABP and the NIRS signals varies with time, we computed COH and PCOH over successive half-overlapping epochs of duration 10, 15, and 12.5 min for the Zurich, Utrecht, and Leuven data respectively. The average of COH and PCOH over the frequency band 0.0033–0.04 Hz (corresponding to phenomena of duration in the range 25–300 s) [3] was used as score for the considered epoch. The epoch durations were computed from the calibration of the mean COH (for all patients of each centre) on the mean absolute-valued correlation coefficient (COR), to be sure that the score value of 0.5 could be considered as critical (it suggests a relation between the signals based on 50% shared variance) [5]. If mCOH or mPCOH was higher than this critical score value (CSV), the infant was said to have an impaired cerebral autoregulation. We built the PCOH algorithms from the supposed physiological models fixing the interactions between the measured signals as follows:

  • SaO2 = i(SaO2) + f(MABP)

  • MABP = i(MABP) + f(SaO2)

  • NIRS = i(NIRS) + f(MABP) + f(SaO2)

where i(…) represents the independent part of the signal, f(…) stands for is a function of, and NIRS represents HbD, rSO2, or TOI. The algorithms are:

  • PCOH1 = COH(MABP–SaO2, NIRS–SaO2)

  • PCOH2 = COH(MABP, NIRS–SaO2)

  • PCOH3 = COH(MABP–i(SaO2), NIRS–i(SaO2))

  • PCOH4 = COH(MABP, NIRS–i(SaO2))

where COH(… , …) is the coherence computed between both signals. For all patients we studied the performances of PCOH compared to COH on a global basis, but we also looked in detail at epochs with fluctuating SaO2 and compared the results when using extra raw (non pre-processed) data.

4 Results

When considering all patients, we saw a trend of higher mPCOHs as compared to mCOH. In addition the CPRTs and PPIs of PCOH were generally higher as compared to the CPRT and PPIs of COH. PCOH3 shows a higher mPCOH, CPRT, and PPI than the other PCOH algorithms, and than PCOH2 which shows the lowest values. For more details we refer to Tables 1, 2 and 3. We also considered patients for whom the oxygen fraction of the inspired air has been intentionally modified. We particularly concentrated on epochs with a high variance in SaO2. In these epochs, mean scores were significantly higher with PCOH3 as the highest. More details are given in Table 4 and Fig. 1

Table 1 Mean score, standard deviation (std), critical percentage of the recording time (CPRT), and pressure-passive index (PPI) with confidence level α = 0.1 of the Leuven data. The numbers below are averages on the ten infants from Leuven
Full size table
Table 2 Mean score, standard deviation, CPRT, and PPI10 of the Utrecht data. The numbers below are averages on the ten infants from Utrecht
Full size table
Table 3 Mean score, standard deviation, CPRT, and PPI10 of the Zurich data. The numbers below are averages on the ten infants from Zurich
Full size table
Table 4 Local analysis: the oxygen fraction inhibited by the infant has, in some Zurich patients, intentionally been modified to create locally a high variance in SaO2. The table contains the overall score means of such a patient, and the means related to the epoch of high SaO2 variance (20–40 min). Please see also Fig. 1
Full size table
Fig. 1
figure 31_1_117031_1_En

Local analysis: the oxygen fraction inhibited by the infant has, in some Zurich patients, intentionally been modified to create locally a high variance in SaO2. Please see also Table 4

Full size image

mPCOH3 detects a few more patients with impaired autoregulation than do mCOH and the other mPCOHs. The CPRT and PPI10 (with confidence level α = 0.1) detects many more patients with impaired autoregulation than mCOH and mPCOHs, and approximately twice as many patients as PPI5 (α = 0.05).

Patients with mCOH and mPCOHs > 0.5 have a slightly lower mean PMA and BW than the overall average. High PCOH values (m > 0.5, CPRT > 0, PPI > 0) are better indicators of poor clinical outcome than COH (MDI < 84, PDI < 84, Apgar < 7). CPRT and PPI10 are better indicators of poor clinical outcome than mean score values.

5 Discussion

It is important to remember that COH and PCOH are measurements of impaired cerebral autoregulation, and that the considered clinical patient characteristics suggest possible brain damage. An evident correlation between impaired autoregulation and brain damage in neonates has been described in the literature and this is what we assume in this study. We looked for the method that best fits the occurrences of brain damage. Our results indicate that (1) the PCOH score is more accurate in detecting infants with brain malfunctions as compared to the COH method; (2) PCOH3 has the highest accuracy in detecting impaired autoregulation.

In particular, the CPRT and PPI10 – computed from the COH and PCOH scores – detect more than 50% of all infants with poor clinical outcomes. These observations indicate rather that PCOH highlights more cases of impaired autoregulation as compared to COH, and does not necessarily mean that PCOH indicates a better fit between patients with impaired autoregulation and patients with poor clinical outcome. However, we should remark that in this study the patient characteristics and outcomes are not available for all patients whereas further statistical analysis on larger multicentre datasets are needed

We expected the PCOH4 model to be more realistic than the PCOH3 method, because SaO2 was seen to have an influence on MABP only in those recordings with changes in SaO2 which were deliberately provoked by modifying the inspired oxygen fraction. Nevertheless, this could not be confirmed by our results. Furthermore, a lack of concordance with patient neurological outcomes could also be explained by the fact that COH (and consequently PCOH) only measures the linear and stationary concordance between MABP and the NIRS-measured signals, or by the fact that PCOH assumes a linear dependency between the considered signals.

In conclusion, as HbD/rSO2/TOI are often considered as surrogate for CBF, impaired autoregulation can be assessed by quantifying the concordance between these signals and MABP. As it is able to eliminate the influence of other signals such as SaO2, the partial coherence method was shown to perform better than the classical coherence method.