Personalization and Pragmatism: Pediatric Intracranial Pressure and Cerebral Perfusion Pressure Treatment Thresholds

Neurocritical care after brain injury is designed to prevent secondary injury and create optimal conditions for survival of recoverable brain tissue. The most important component of this approach is to limit global or regional cerebral ischemia. To accomplish that, we attempt to avoid excessive elevation of intracranial pressure (ICP) and maintain adequate cerebral perfusion pressure (CPP). However, the ideal ICP and CPP targets for a given patient are not known.(1)

The ICP treatment threshold of 20 mm Hg recommended by pediatric traumatic brain injury (TBI) treatment guidelines(2) has not changed in two decades (Table 1), despite relatively weak supporting evidence. Recommended pediatric CPP thresholds have decreased and become less age-dependent despite known age-related changes in mean arterial pressure (MAP), cerebral blood flow (CBF), and cerebral metabolic rate of oxygen.

Table 1:

Guideline-Recommended ICP and CPP Treatment Thresholds

Year	2000	2003	2007	2012	2017	2019
Treatment Threshold
ICP
Adult	20-25		20		22
Pediatric		20		20		20
CPP
Adult	70		50-70		60-70
Pediatric		40		40		40
“age-related continuum”		40-65		40-50		40-50

Legend: Intracranial pressure (ICP) and cerebral perfusion pressure (CPP) treatment thresholds, in mmHg, recommended by each edition of the Severe Traumatic Brain Injury Treatment Guidelines for adults and children.

Simultaneously, the field is undergoing a vigorous debate about whether fixed ICP or CPP treatment thresholds should exist.(3, 4) Heterogeneity is a signature of TBI and other forms of brain injury, relationships between CBF, CPP, cerebral oxygen delivery and utilization, and cellular metabolism are complex, and ICP time series have many features other than whether or not a treatment threshold is crossed.(3) Personalized therapy has become the standard in oncology and other medical specialties. Multi-modal invasive monitoring can inform one form of personalized therapy guided by cerebrovascular pressure reactivity, brain tissue oxygenation, cerebral metabolism, and electrophysiology. However, it is invasive, resource-intensive, and not often used in children.

Others have argued for a pragmatic form of ICP-guided therapy designed to limit treatment toxicity: non-invasive ICP monitoring first, less invasive ICP monitoring overall, and more generalized ICP and CPP targets.(4) The distinction between a threshold that should not be breached and a threshold where treatment should begin (indicating some wiggle room) is important. A landmark trial in adults with TBI found no difference in outcome between intensive imaging- and clinical examination-guided (ICE-guided) therapy compared to ICP-guided therapy.(5)

Against this backdrop, Woods et al.(6) report the results of a retrospective cohort study in this issue of Pediatric Critical Care Medicine. The cohort was 262 children who received an ICP monitor for any indication at a single center during 2012-2016. The study aimed to evaluate for associations between each child’s mean hourly ICP and CPP values and subsequent in-hospital mortality. They developed ICP and CPP thresholds to test by generating cut-points based on Youdon’s index. The authors conducted subgroup analyses for children with (n = 87) and without (n = 175) TBI and across pre-specified 3 age groups (<2 years old, 2-8 years old, and ≥8 years old).

The authors found that a mean ICP > 15 mm Hg (non-TBI) and > 18 mm Hg (TBI) was associated with mortality. These threshold values are lower than current guideline recommendations (Table 1) and agree with other pediatric studies.(7) They also agree with a recent analysis of ICP-monitored adults(8) that found that ICP values > 19 mm Hg (lower than the currently recommended adult threshold, Table 1) were associated with mortality.

In addition, children with TBI with mean CPP < 67 mm Hg (much higher than in current guidelines) had higher mortality. No such association was found in the non-TBI group. This is a valuable contribution. ICP and CPP thresholds developed for children with TBI are frequently applied to children without TBI, but few studies to date have explicitly addressed whether this is appropriate or not. The need for comparative studies is particularly acute as the TBI guidelines become more evidence-based over time.

In the combined cohort, the CPP thresholds associated with mortality increased with patient age: 47, 58, and 73 in the 3 age groups. This analysis is a significant strength of the manuscript. Whether or not ICP and CPP thresholds should be age-dependent is a critical knowledge gap in the current pediatric guidelines. Sound physiologic arguments for age-dependent thresholds exist, but the field has lacked clinical evidence of sufficient strength to inform practice. This study in isolation is not enough, but it should inform additional work to answer this important question.

One obstacle to advancing our understanding is the lack of a known normal value for ICP in children. Neither the definition of pediatric intracranial hypertension nor therapeutic guidelines have a useful baseline ICP reference. A recent study in adults(8) identified 8 and 9 mm Hg as the most common ICP values in ICP-monitored adults. No pediatric equivalent is known. It is certainly plausible that normal ICP values would change with cerebrovascular development and be age-dependent.

Another strength of this manuscript is the attempt to leverage dense clinically-collected data to inform clinical practice. As in this manuscript, hourly vital signs in electronic health records (EHRs) are typically bedside monitoring system outputs selected and validated by nurses. These hourly values are already “smoothed” – nurses have typically removed outliers and selected a value representative of the signal during that hour. More granular “unvalidated” signals from continuous monitors are a relatively untapped resource for building clinical decision support systems.

The manuscript also has several weaknesses. One problem is that ICP and CPP values are nearly always collected from children with brain injury who are receiving threshold-guided therapies. Correspondingly, persistently elevated ICP and/or depressed CPP may largely reflect injury severity. We have long known that children whose ICP we can lower have better outcomes than those whose ICP we cannot.(2) Second, in-hospital mortality is a fairly insensitive outcome measure. Assessment using validated outcome scales several months after injury is now the gold-standard. In a related issue, important confounders were not considered in the mortality models. These include known predictors of poor outcome after TBI such as penetrating injury, non-accidental trauma, severe non-head injuries, and cardiac arrest as well as lethal non-TBI conditions such as hemophagocytic lymphohistiocytosis and subarachnoid hemorrhage. The lack of confounding adjustment was largely due to the relatively small single-center sample size, which also limits the study’s generalizability. Given the heterogeneous pathophysiology, separate TBI and non-TBI mortality models would be more appropriate than a TBI covariate in a single model.

The idea that mean, minimum, and maximum ICP and CPP values accurately represent the contribution of ICP and CPP to mortality has several challenges. Averaged data may fail to account for intermittent periods of threshold crossing associated with secondary injury.(9) Among those who ultimately die, maximum ICP may occur at the end of aggressive attempts to support a patient and proximate to a decision to limit or withdraw interventions. Among those who ultimately survive, maximum ICP may occur at a variety of different times. Means are sensitive to outliers. Local practice regarding how long ICP monitors are left in could powerfully influence the observational data available. The relative impact of transient elevations or the amount of time spent with elevated ICP can only be assessed using “pressure-time indices.”(7, 10) Fixed treatment thresholds could lead to harmful under- or over-treatment. Finally, the pathway to translation for models such as those presented in this manuscript is in clinical decision support (CDS) systems. Means, minima, and maxima are not viable prospective predictors in CDS systems because they are not available until all data have accumulated.

Another method to personalize treatment is to use computational physiology models paired with data assimilation methods, which optimize model state and parameters to external observations.(11) Personalized model output could be used to customize treatment on the basis of both estimated and observed ICP as well as other observed patient data. Existing modeling approaches(12, 13) and their extensions deserve attention for potential roles in future TBI management. In particular, such models can be conditioned on monitored quantities such as ICP, MAP, and oxygenation to provide numerical estimates of autoregulatory function and adaptive capacity. Estimates of these mechano-physiological properties should better inform appropriate treatment thresholds for TBI management.

In order to advance generalizable personalized therapy, the field needs multi-center sharing and integration of ICP, MAP, and other signal data from a wide variety of patients. Hourly nurse-validated signal data likely do not have sufficient resolution to represent all of the physiological processes linked to patient outcome. Analysis of ICP waveforms and other signal-derived time series may be one method to target therapeutic guidelines on the basis of empirical injury phenotypes.(14) The Brain-IT (http://www.brainit.org) and Physionet (https://physionet.org) groups have made progress in collecting and sharing such signals, but more remains to be done. One open question is how dense the signals need to be. An every-one-minute sampling frequency may be sufficient for many studies.(8) Other studies may require highly granular signals that allow beat-to-beat resolution.(12, 13)

Pragmatic ICP and CPP treatment thresholds will still be needed. The global burden of TBI still falls heaviest on low- and middle-income countries(15) where ICP monitoring is inconsistently available, let alone multi-modal monitoring and other sophisticated care. Models developed using higher-resolution signal data paired with robust injury and patient descriptors and studies of their application in prospective use will inform a better balance of benefits and risks from ICP-targeted therapies guided by both personalized and pragmatic treatment thresholds.