Keywords

Introduction

Previously, we have studied the feasibility of distinguishing ICP B waves (0.5–2.0 cycles/min) of any shape and strength from flat ICP recordings based on ICP pulse morphology and mean value of ICP [9]. The present study was designed to investigate whether ICP B waves of different shapes are associated with different ICP pulse waveform morphology. Mean values of ICP were excluded from this study in order to solely investigate changes in ICP pulse shape. We hypothesised that ICP waveform morphology reflects cerebrovascular changes [10]. Given that ICP B waves are thought to arise from vasomotor instability and ICP, waveform morphology associated with B waves should be unique. The study aimed to investigate whether or not the shape of the ICP pulse waveforms changes in association with the ICP B wave occurrence and to identify the discriminating power of metrics describing ICP pulse waveform morphology to differentiate between different patterns of ICP B wave activities in overnight monitoring of hydrocephalus.

Materials and Methods

A total number of 276 episodes of four clearly distinctive ICP slow wave activities (Fig. 1a–d) were selected from retrospectively analysed pre-shunt operation, overnight intracranial pressure (ICP) recordings performed in 44 patients (18 female and 26 male; mean age: 68  ±  17 years, range 32–94 years) hospitalised at the UCLA Adult Hydrocephalus Centre. The preliminary diagnosis (based on brain scans and clinical picture) for all patients was hydrocephalus. One hundred and thirty-one selections of flat ICP patterns containing no episodes of slow waves (NW), 74 selections of symmetrical slow waves (SW), 45 selections of asymmetrical slow waves (AS) and 26 selections of slow waves with plateau phase (PW) were included into this study. Mean duration of selection was 9 min 58 s  ±  4 min 15 s. An ICP intraparen­chymal microsensor (Codman and Schurtleff, Raynaud, MA, USA) was inserted into the right frontal lobe. Simultaneous recordings of ICP and ECG signals were performed using either a PowerLab™ SP-16 acquisition system (ADInstruments, Colorado Springs, CO, USA) connected to the analogue outputs of the Codman ICP Express Box and the GE bedside monitor or the BedMaster™ system, which acquire data from bedside monitors. A different sampling rate was used by the acquisition systems: PowerLab™ – 400 Hz and BedMaster™ – 240 Hz. An insertion of the ICP sensor and data acquisition were approved by the local IRB committee and informed consent was obtained from each patient or from their family member.

Fig. 1
figure 1figure 1

Four patterns of intracranial pressure (ICP) oscillations and corresponding dominant ICP pulses: (a) NW – no episodes of ICP B waves, (b) SW – symmetrical ICP B waves, (c) AS – asymmetrical ICP B waves, (d) PW – slow ICP B with plateau phase. Each dominant pulse was extracted from 30 s of data

Full size image

The morphological clustering and analysis of ICP pulse (MOCAIP) algorithm was utilized to measure the set of ICP pulse metrics based on the location of sub-peaks in an ICP pulse: systolic peak (P1), tidal peak (P2) and dicrotic peak (P3). Details of this algorithm were precisely described in another publication [8]. The set of 10 ICP pulse metrics was calculated to describe the morphology of ICP pulsation: amplitudes of pulse peaks (dP1, dP2, dP3), latency (LT) between the peak of the QRS complex of ECG and the corresponding onset of the ICP pulse, latencies between the onset of ICP pulse and the location of three peaks (L1, L2, L3) and curvatures of the three sub-waves of ICP pulse wave (Curv1, Curv2, Curv3). The average of the MOCAIP features quantitatively estimates the ICP pulse wave shape, although it ignores the changes during slow wave occurrence (Fig. 1). Therefore, besides mean values of 10 MOCAIP metrics, we also calculated their standard deviations (S.D.). Hence, a total number of metrics analysed increased to 20 (Table 1).

Table 1 Intracranial pressure (ICP) pulse wave metrics
Full size table

Statistical Methods

With regard to statistical analysis, the forward step-wise discriminant analysis (DA) [12] was used to investigate which features are the most significantly altered by ICP B wave occurrence. At each step of analysis 20 MOCAIP metrics were reviewed and evaluated to determine which one will contribute most to the discrimination between four patterns of ICP B wave activity (NW, SW, AS and PW). The step-wise procedure was controlled by F to enter value (set at 4.0) and minimum tolerance value (1 − R 2 set at 0.01). We repeated the analysis 44 times; in every run the samples from one patient were excluded and the selections from the remaining 43 patients were used to build the discriminant model. Only the metrics that were selected most often (more than 22 times) were included in the final analysis. The significance of the discriminant functions obtained was tested based on a multivariate F value (Wilks’ lambda statistics). Standardised beta coefficients were utilised for estimation of the unique power of discrimination of the variables. To identify between which of the groups the respective function discriminates we looked at the means of canonical variables.

Results

Mean values and standard deviations of ICP pulse metrics along with the number indicating how many times a particular metric was chosen by DA were summarised in Table 2. Nine out of 20 ICP pulse metrics were selected more than 22 times. Eight of them (highlighted in grey in Table 2) were accepted (based on the p value for the F statistic <0.05) into the final model (Wilks’ lambda: 0.31, F =  16.2, p  <  0.0001). Three discriminant functions (the maximal possible number) were statistically significant. The first function (Function 1, p  <  0.000001) differentiates between two groups: combined PW and AS vs combined SW and NW. It explains 83.6% of discriminating power while the second and the third ones account for 16.4% of total power. The standardized coefficients for the two variables SDdP2 and SDdP3, demonstrated the greatest values; therefore, their contributions to discrimination between PW  +  AS vs SW  +  NW are the most significant (Eq. 1).

Table 2 Mean (±SD) and standard deviation (±SD) of the ICP pulse metrics for four patterns of ICP fluctuations
Full size table
$$ \begin{array}{l}2.3SDd{P}_{2}\text-\text1.3SDd{P}_{3}\text-\text0.{\text{4SDCurv}}_{1}\text+\text0.{\text{3SDL}}_{1}\\ +\text0.{\text{3SDL}}_{3}\text+\text0.{\text{2Curv}}_{2}\text-\text0.{\text{2SDCurv}}_{2}\text-\text0.{\text{1Curv}}_{3}\end{array}$$
(1)

The Function 2 (p  <  0.000001) discriminates between PW and AS groups mostly based on inverse changes in mean Curv2 and Curv3 (Eq. 2).

$$ \begin{array}{l}1.2Cur{v}_{2}\text-\text1.1Cur{v}_{3}\text-\text0.{\text{6SDCurv}}_{2}\text-\text0.{\text{4SDL}}_{3}\\ +\text0.{\text{2SDL}}_{1}\text+\text0.{\text{2SDCurv}}_{1}\text+\text0.{\text{2SDdP}}_{3}\text+\text0.0{\text{SDdP}}_{2}\end{array}$$
(2)

The Function 3 (p  <  0.002) separates between PW and SW and the variables SDdP2 and SDdP3 have the biggest impact on this discrimination (Eq. 3)

$$\begin{array}{l}2.1SDd{P}_{2}\text-\text2.0SDd{P}_{3}\text-\text0.{\text{7SDCurv}}_{2}\text-\text0.{\text{7SDCurv}}_{1}\\ +\text0.{\text{3Curv}}_{3}\text+\text0.{\text{2SDL}}_{1}\text-\text0.{\text{1Curv}}_{2}\text+\text0.{\text{1SDL}}_{3}\end{array}$$
(3)

To visualise how the two functions discriminate between groups the individual scores for the two functions (Function 1 vs Function 2) were plotted on Fig. 2.

Fig. 2
figure 2

Scatter plot of the canonical scores from the discriminant ­analysis of (•) no ICP B waves, (□) symmetrical ICP B waves, (◊) asymmetrical ICP B waves, (▲) slow ICP B with plateau phase PW

Full size image

The variation of amplitudes of peaks P2 and P3 contributes to the distinction between NW  +  SW vs AS  +  PW (Fig. 3). Mean values of SDdP2 and SDdP3 evolve from NW to PW when different ICP patterns were compared (Fig. 3a). The discriminating power, however, was attributed to the relative changes between SDdP2 and SDdP3: profound variations in dP2 prevailed over the changes in dP3 during AS and PW, but the range of variation in amplitudes of these two peaks was similar during SW and NW (Fig. 3a). The highest discriminating power to distinguish between PW and AS was ascribed to the mean curvatures of the two peaks P2 and P3. A more rounded sub-wave of P2 along with a less sharp sub-wave of P3 seems to be typical for PW (Fig. 3b). We could not find any significant combination of MOCAIP features to distinguish between SW and NW.

Fig. 3
figure 3

Mean  ±  SE (box) ±95% confidence interval (whiskers) of (a) SDdP2 and SDdP3 and (b) Curv2 and Curv3 for four patterns of ICP oscillation: NW – no episodes of ICP B waves, SW – symmetrical ICP B waves, AS – asymmetrical ICP B waves, PW – slow ICP B with the plateau phase

Full size image

Discussion

Based solely upon ICP pulse morphological features (mean ICP values were not taken into account in this study), we were able to demonstrate that different subtypes of ICP B waves are indeed associated with different shaped ICP pulsations and a different range of alterations of MOCAIP metrics. Our finding suggests then that AS and PW might be physiologically distinct from SW and NW. Moreover, AS and PW may also be physiologically different.

The changes in ICP pulse morphological features reflect acute changes in intracranial conditions and changes in the cerebral haemodynamics [1, 2]. The origin of the three peaks in ICP pulse waveform has not yet been established. The P2 component is thought to be dependent upon intracranial compliance [6] whereas pulsations of major arteries and choroid plexus may contribute to the P1 component [2, 13]. Miller noted that venous hypertension from jugular venous or sagittal sinus occlusion causes an increase in the terminal portion of the ICP pulse wave, which suggests the venous pressure dependency of P3 [2, 13]. Recent study in TBI patients showed that the amplitude of ICP was mostly affected by changes in the pulsatility of the cerebral blood volume (CBV), whereas in patients with NPH, who underwent an infusion study, amplitude was more dependent upon mean ICP, which was increased during the test in a controllable manner [3]. In the case of ICP B waves, both factors (mean ICP and increased pulsatility of CBV) play a significant role. As the ICP slow waves are most probably vasogenic in origin, they may be dependent on cerebral blood volume pulsatility [4, 7]. Moreover, the compliance in NPH patients is usually reduced owing to underlying pathological conditions, which may amplify an increase in mean ICP during ICP B waves and provoke proportional changes in ICP pulse amplitude [11, 15].

In our study the distinguishing power was mostly related to variation in the amplitudes of two peaks: P2 and P3, which may reflect more reduced intracranial compliance and increased venous pressure because of an elevation in ICP during AW and PW ICP B waves compared with during NW and SW. It was suggested by recent studies [5] that an increased amplitude of ICP pulsation may have predictive value for selecting shunt responders. It was also discussed in an early work [14], where a five-category classification of B wave was proposed, that only the asymmetrical B wave has a predictive value with regard to outcome after shunting. Therefore, we think that joint analysis of both ICP pulse wave amplitudes and ICP B waves (AS and PW in particular) performed in hydrocephalic patients undergoing overnight monitoring might be ideal for overnight ICP analysis. Further study is needed to establish the usefulness of such a joint approach.

Conclusion

Subtypes of ICP slow wave are associated with different configurations of morphological features describing ICP pulse waves, which may suggest that they are physiologically distinct.