Transcranial Doppler Monitoring of Intracranial Pressure Plateau Waves

Acute intracranial hypertension (ICH) is a recurrent cause of secondary injury in patients under neurocritical care. Intracranial pressure (ICP) represents an essential monitoring modality in the correct clinical management of several neurological diseases carrying risk of potentially lethal ICH. ICP is a complex variable, consisting of four components modulated by different physiological mechanisms: (1) inflow and volume of arterial blood, (2) venous blood outflow, (3) cerebrospinal fluid (CSF) circulation, and (4) brain parenchymal or lesion volume. The importance of ICP is not always associated with its absolute value, but with monitoring of its dynamics in time and with identifying which of the above mentioned components are responsible for the observed pattern of ICH [1]. This is essential, as every component that elevates ICP requires different countermeasures (like short-term hyperventilation to control 1; head elevation to control 2; extra ventricular drainage to control 3; osmotherapy or craniectomy to control component 4).

First described by Janny [2] in 1950 and later by Lundberg [3], plateau waves of ICP (or Lundberg A waves) [3] are characterized by sudden and relevant increases in ICP (generally above 40 mmHg [4]), related to increased volume of arterial blood in response to arterial vasodilation stimulus. Such phenomena may develop in patients presenting intact cerebral autoregulation [4] and low cerebrospinal compensatory reserve [5], suffering from a wide range of cerebral pathological conditions including traumatic brain injury (TBI) [4], idiopathic intracranial hypertension [6], subarachnoid hemorrhage [7], brain tumors, hydrocephalus [8], and craniosynostosis [9].

During the occurrence of plateau waves, ICP increases from normal or slightly elevated to uncompensated, acute ICH, with relatively stable arterial blood pressure (ABP) [10]. The mechanism driving plateau waves can be described as a “vasodilatory cascade,” initiated by a vasodilatory stimulation (reduction in ABP during intact autoregulation, for instance) [4]. Following this, a rapid increase in cerebral blood volume (CBV) leads to a cyclic rise in ICP, decrease in cerebral perfusion pressure (CPP), further vasodilation and further rise of ICP. This cycle is maintained until the cerebral vasculature reaches a state of maximum vasodilation [10]. The reverse of this positive feedback loop occurs when a vasoconstrictive stimulus initiates a vasoconstrictive cascade, decreasing CBV with a consequent drop in ICP toward normal levels [10, 11].

Transcranial Doppler ultrasonography (TCD), owing to its specificity to detect changes in cerebrovascular dynamics [12], offers non-quantitative measurements of cerebral blood flow (CBF). CBF is the blood supply to the brain in a given period of time, and global changes in this parameter can be monitored continuously and noninvasively using TCD-derived cerebral blood flow velocity (FV) [13]. CBV and CBF are usually correlated, as CBV equals CBF multiplied by the cerebral vascular mean transit time [14]. Some studies have demonstrated that specific patterns of FV waveform reflect inadequate cerebral perfusion caused by a decrease in CPP [15, 16] (as it occurs during plateau waves, for instance). In such cases, ICP increases lead to decreases in CPP by closing arterioles at higher diastolic pressures, thus causing remarkable drops in diastolic flow velocity (FV_d). Consequently, there is a generalized decrease in mean flow velocity (FV_m). These characteristics observed in the FV waveform pattern can be used as indicators of cerebral perfusion derangements and have been applied as variables for noninvasive ICP estimations [17]. However, the inherent difficulty of ICP estimation in such a scenario is the lack of reliable individual calibration of estimators in appropriate units (mmHg).

Although it has been shown that different TCD-based nICP estimators present with different measures of accuracy [18], there are reports suggesting that these methods are more accurate when changes of ICP are related to vasogenic phenomena [19, 20]. Considering these assumptions, the aim of this study was to assess a set of four previously described TCD-based nICP methods in a cohort of patients who presented changes in ICP purely of vasogenic origin (specifically plateau wave increases of ICP) to verify whether these methods reliably predict ICP under such conditions.

In 27 patients, a total of 36 plateau waves were identified (7 patients presented two plateau waves, and one patient presented 3 plateau waves). Table 1 presents correlations (∆R) between ∆ICP and ∆nICP and averaged correlations in time domain.

Correlations in time domain are independent of mean values of ICP or nICP and represent the ability of a nICP method to replicate relative changes observed in direct ICP across time. In average, correlations in time were reasonably good, with R > 0.60 for all methods. nICP_PI, nICP_BB, and nICP_CrCP presented similar averaged correlations in time (R ≥ 0.78). Examples of good and poor correlations between ICP and nICP during plateau waves are shown in Fig. 1.

Correlations between ∆ICP and ∆nICP were better represented by nICP_CrCP and nICP_BB, while the other methods presented inferior and nonsignificant correlations (Fig. 2). All ∆nICP mean values were significantly underestimated in comparison with ∆ICP.

Table 1 also contains the prediction values for each nICP method according to the ROC analysis. The best AUC value for predicting intracranial hypertension (i.e., ICP ≥ 35 mmHg) was presented by nICP_PI. Nevertheless, all methods presented AUC above 0.7, an indication of reasonable prediction abilities for detecting ICH. The DeLong’s test for two correlated ROC curves did not reveal any statistically significant difference between nICP methods. The ICP thresholds yielding the best sensitivity and specificity were 27.41 mmHg (AUC = 0.86), 28.42 mmHg (AUC = 0.79), 23.08 mmHg (AUC = 0.75), and 20.28 mmHg (AUC = 0.84), respectively, for nICP_BB, FVd, CrCP, and PI, respectively.

nICP_CrCP and nICP_PI presented null values for specificity and negative predictive value, indicating that values of these two estimators during plateau waves were always below 35 mmHg (although they reacted to rise in ICP, a remarkable underestimation of real pressure was observed). Nevertheless, all methods presented good positive predictive values (65, 69, 58, and 58% for nICP_BB, FVd, CrCP, and PI, respectively), which otherwise would indicate that many of the positive results were false positives.

Table 2 presents the median values (IQR) for the physiological variables evaluated during baseline and plateau phases. During plateau waves, ICP (and ICP pulse amplitude—ICP Amp) increased and CPP decreased significantly, whereas ABP remained unchanged. FV_m and FV_d presented significant decreases, whereas FV_s increased significantly. This resulted in significant increase in PI and decrease in CVR. As the only nICP method provides ICP waveform, nICP_BB presented increased pulse amplitude (nICP_BB Amp), even with replication of the characteristic triangular waveform shape observed during plateau waves [30] (Fig. 3). The correlation between ∆ICP Amp and ∆nICP_BB Amp was significant (R = 0.50, p < 0.05); however, nICP_BB Amp was significantly underestimated in comparison with ICP Amp during plateau phase (Bias = −2.88 mmHg, p < 0.05).

In this work, we present the results of an assessment of a set of TCD-based nICP methods in a cohort of patients with changes in ICP purely of vasogenic origin (i.e., plateau waves of ICP). Within the evaluated methods, only nICP_BB and nICP_CrCP presented significant correlations with ∆ICP. On the other hand, changes in ICP recorded in time  were confidently replicated by all methods.

Changes of ICP in time domain have been assessed in other clinical conditions, and it appears to be highly accurate when changes in ICP are of vasogenic origin. For instance, during CSF infusion tests a controlled and artificial increase in CSF circulation causes an elevation in ICP similar to the pattern observed during plateau waves. Nevertheless, the correlations in time domain obtained with the same nICP methods were considerably weaker [19] in comparison with those observed here, where ICP changes were specifically related to increases in cerebral arterial blood volume. Moreover, such findings were not better replicated when another TBI cohort was analyzed [31], in which changes in ICP across time were not always related to vasogenic waves.

For a better understanding of the results, we compared the accuracy measures of the assessed methods in different clinical conditions related to ICP increase (plateau waves, CSF infusion tests [18], and TBI [31]). The 95% confidence interval for nICP estimation is around 18 mmHg during plateau waves and 15 mmHg in infusion studies, while in TBI it is remarkably smaller, 9.4 mmHg. The black-box model is the best one across these three clinical scenarios. The correlation coefficient in time domain between real and estimated ICP is greatest during plateau waves (around 0.8, except for nICP_FVd [0.63]). For infusion studies, the correlation is much lower (between 0.3 and 0.4 for nICP_BB, FVd, and CrCP). In TBI, the correlation is between 0.5 to 0.6, but only for black-box and PI-based methods.

In respect to ICP accuracy, it is known that even the standard invasive techniques might not present the specified limits for error [32‐34], particularly using intraparenchymal microtransducers. Thus, it is debatable whether these accuracy requirements are realistic for all sorts of ICP monitoring. In view of this, an important concept that should be stressed is ICP not solely “as a number,” once dynamical features of this parameter, such as its waveform and relative changes in time, are fundamental for a proper assessment of the clinical state of the patient [35].

The nICP black-box model had been previously analyzed during plateau waves of ICP [20]. In that study, clinical material from 17 patients was used to construct the ICP simulation model, which produced better accuracy and better correlation between real and estimated values of ICP in comparison with our present study. This discrepancy could be associated with the way nICP_BB was generated in both studies. In Schmidt et al.’s previous study [20], the formation dataset was much more specific for plateau waves, in which 7 patients (41.18% of the dataset) presented such phenomena. On the other hand, the formation dataset used for the present study consisted of a large general TBI cohort (n = 140 patients), as mentioned previously.

In this regard, we could generally infer that nICP methods based on formation datasets specific for certain conditions may present better prediction performances than those based on general datasets. An example supporting this inference can be found in another study [31], which demonstrated that the same methods used here presented smaller 95% CI when analyzing a TBI cohort convergent to the formation datasets, i.e., general in terms of different sources of ICP changes. However, the obvious disadvantage of specific formation datasets is the restriction of applicability of the nICP method. This effect was observed when TBI dataset-generated nICP procedures were used for ICP assessment in an NPH cohort subjected to CSF infusion tests [19]. In that study, the use of non-specific nICP formation datasets resulted in rather large deviations from ICP.

Although the evaluated methods generally presented low correlations considering the magnitude of changes in ICP as an absolute value, not all clinical situations involving ICP monitoring may require high accuracy. Alternatively, TCD allows a noninvasive assessment of cerebral circulation dynamics as ICP changes in time domain, with the possibility to track nICP changes in real time in a variety of clinical settings (emergency rooms, ambulatories, operating theatres). This ability is one of the gains of using TCD for nICP monitoring, and it could potentially be applied as an alternative assessment or screening tool where ICP measurements are not part of a standard clinical protocol, or in situations which invasive ICP monitoring is contraindicated.

Plateau waves affect approximately 25% of TBI patients [10] and might also occur in other conditions such as subarachnoid hemorrhage [7], brain tumors, hydrocephalus [8], and craniosynostosis [9]. The identification of these phenomena could represent an important application of TCD-based noninvasive ICP methods for the conditions aforementioned, as plateau waves have been also reported to be associated with poor outcome in TBI patients, especially in regard to their duration [10]. Thus, identifying plateau waves as nICP changes in time domain could eventually provide to the clinician a better understanding of the clinical state of the patient and guide oriented treatments in situations of sustained plateau waves.

The methods of nICP assessment are remarkably accurate in detecting relative changes in ICP during plateau waves across time. Furthermore, they presented high performance in ruling out intracranial hypertension. These characteristics encourage the application of TCD-based nICP in clinical conditions where the knowledge of absolute values of ICP would not be essentially relevant. However, we could infer that the estimation of ICP absolute values by the nICP methods tested is still limited at the current state of development. Such inaccuracy might be related to the matter of non-specific nICP calibration.