Abstract
Intracranial pressure (ICP) monitoring remains a cornerstone in the management of severe traumatic brain injury (TBI), yet its utility as a dynamic predictor of outcomes continues to evolve. We aimed to examine the role of serial ICP measurements as a potential predictor of outcomes after TBI, to combine ICP data with cerebrovascular reactivity metrics, and to highlight emerging trends in ICP modeling such as machine learning-based predictive models. We conducted a rigorous scoping review following Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews guidelines to investigate the utility of ICP monitoring as a dynamic predictor of outcomes following TBI. A systematic search of major databases identified relevant studies published between January 1, 1998, and August 1, 2024. Two reviewers identified relevant articles, and conflicts were adjudicated by a third. Data from the included studies were abstracted and synthesized. Analysis of 29 studies (N = 5,743 patients) revealed significant associations between specific ICP patterns and clinical outcomes. Key findings included threshold-dependent mortality predictions, the value of early monitoring patterns (i.e., patterns observed within the first 72 h post-injury), and the enhancement of predictive accuracy through integration with cerebrovascular reactivity indices. Many studies now explore ICP as a multidimensional metric rather than a straightforward number, but overarching conclusions are limited by inter-study variability in analysis. The integration of advanced monitoring techniques, the use of features capturing the temporal complexity of ICP, and machine learning approaches show promise in enhancing the predictive value of ICP monitoring as a new form of precision medicine. These findings support strong associations between specific ICP dynamic patterns and mortality and functional outcomes. Standardization of protocols and validation in diverse populations remain important challenges to address in future studies.
Introduction
Traumatic brain injury (TBI) affects an estimated 69 million individuals annually throughout the world, posing a significant burden on patients and health care systems while also driving a global health care burden exceeding $400 billion per year. 1 Since the 1950s, intracranial pressure (ICP) monitoring has been a cornerstone of TBI management in the critical care setting, offering critical insights into secondary injury mechanisms and guiding therapeutic interventions. However, emerging evidence suggests that ICP values alone may not fully capture the multifaceted nature of TBI pathophysiology. For instance, the pathophysiology of secondary brain injury involves a complex interplay between cerebral blood flow (CBF), oxygen delivery, and substrate utilization, which is inadequately monitored by ICP.2,3 Moreover, simply normalizing ICP parameters does not necessarily translate into improved patient outcomes.4,5 These observations underscore the need for a more comprehensive approach that integrates ICP monitoring with physiological and clinical parameters to optimize care and decision-making.5–8 Recent advances in monitoring technology and analytical methods have expanded our understanding beyond simple threshold-based approaches, suggesting that the relationship between ICP patterns and patient outcomes may be more nuanced than previously recognized. 9
Current Brain Trauma Foundation (BTF) guidelines recommend ICP monitoring in patients with severe TBI (Glasgow Coma Scale [GCS] <8) and abnormal computed tomography findings (level II B recommendation), and the Seattle International Severe Traumatic Brain Injury Consensus Conference (SIBICC) offered additional guidance for managing ICP crises, emphasizing a tiered approach to treatment. 8 Nevertheless, in spite of widespread acknowledgment of the importance of ICP monitoring, substantial variability exists in these monitoring protocols, thresholds for intervention, and ICP-based prognostication across studies and institutions.10–12
Part of this variability in ICP-guided monitoring and treatment may be due to the continued evolution of our understanding around how ICP is related to TBI progression and treatment. Recent innovations in ICP measurement—such as refined frequency sampling protocols and artificial intelligence-driven methods for artifact detection and minimization—have the potential to enhance both data fidelity and clinical decision-making. However, evidence supporting these emerging approaches remains constrained by a relatively small number of individual studies with relatively few participants. 13 The objective of this scoping review is to examine the current evidence for serial or near-continuous ICP measurements to predict outcomes after moderate/severe TBI, incorporating emerging methodologies and novel analytical approaches.
Materials and Methods
Protocol and guidance
This study was registered with the Open Science Framework portal (https://osf.io/wpf36/?view_only=8d171152a57143afb91e85b653cb5fac). This review was conducted according to Arksey and O’Malley’s framework for scoping reviews and adhered to Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) guidelines. 14 Ethical approval by our institutional review board was not required.
Selection criteria
We conducted a systematic review of peer-reviewed literature, selecting studies that met predefined inclusion criteria. Eligible studies were published in English (or had an English translation) in peer-reviewed journals and focused on adult patients (age ≥18 years) in hospital settings with a primary diagnosis of TBI. These studies reported the association between ICP and outcomes after TBI, including the Glasgow Outcome Scale (GOS) or the Glasgow Outcome Scale Extended (GOSE) at any time post-injury. Secondary outcomes of interest included hospital length of stay (LOS), intensive care unit (ICU) LOS, and mortality. Studies were required to specify ICP monitoring type and sampling frequency categorized as either “intermittent” or “continuous.” Intermittent sampling refers to measurements taken at intervals of 1 h or less, while continuous sampling involves more frequent measurements, typically recorded on a minute-to-minute basis or at waveform resolution. We excluded pediatric studies, nonhospital settings, conference abstracts, and studies lacking details on monitoring protocols.
Search strategy
We searched PubMed, CINAHL, Cochrane Library, and Embase from 1998 to 2024 without language restrictions. PubMed search terms were defined as: (“Traumatic Brain Injury” OR TBI) AND (prediction OR prognosis) AND (“ICP” OR “Intracranial hypertension” OR Longitudinal OR Trajectory OR “Time series” OR dynamic OR GCS) AND (humans[Filter]) AND (English[Filter]) AND (“critical care”) AND (hospital) NOT (pedi* OR child*). CINAHL search terms were defined as: (“Traumatic Brain Injury” OR TBI) AND (prediction OR prognosis) AND (“ICP” OR “Intracranial hypertension”) AND (Longitudinal OR Trajectory) AND (“Time series” OR dynamic OR GCS) AND (human) AND (“critical care” OR hospital) NOT (pedi* OR child*). Cochrane Library search terms were defined as: (“Traumatic Brain Injury” OR TBI) AND (prediction OR prognosis) AND (“ICP” OR “Intracranial hypertension” OR Longitudinal OR Trajectory OR “Time series” OR dynamic OR GCS) AND (humans) AND (“critical care”). Embase search terms were defined as: “traumatic brain injury”/exp AND “intracranial hypertension”/exp AND “time series” AND human AND hospital AND [adult]/lim AND [english]/lim. Our search was performed between August 22, 2024, and September 3, 2024.
Selection process
According to guidelines, two reviewers (J.H.K. and R.C.O.) excluded publications that were not eligible or duplicate based on title and abstract review. Then, full-text articles were reviewed independently. Conflicts in study selection were resolved by a third independent reviewer (H.E.H.).
Data extraction and analysis
Data were extracted by the same two reviewers and entered into a shared database. Data extraction focused on study characteristics, monitoring protocols, outcome measures, prediction models, and methodological approaches. We specifically examined population characteristics and injury severity (GCS), ICP monitoring protocols, devices used, outcome measures and their definitions, statistical and analytical methods, novel approaches to ICP data interpretation, validation methods where applicable, and emerging areas of research. Disagreements were resolved through discussion and consensus of both reviewers. If consensus could not be reached, the third reviewer could adjudicate. Findings are summarized in Table 1.
Results
Eligible study characteristics
Our selection process is detailed in Figure 1. Initial literature search identified 319 articles eligible for review. Seven duplicate records were removed. After screening titles and abstracts, 240 studies were excluded, while 72 articles were sought for retrieval and reviewed in their entirety for eligibility. Forty-three studies were excluded during full-text review, leaving 29 articles for inclusion (Table 1). These studies contained sample sizes ranging from 17 to 651 patients, with a median of 174 patients, which together totaled 5,743 participants. Most studies (96.5%) included severe TBI (GCS 3–[6 or 8]), with some (48.2%) also including moderate injuries (GCS 9–13). Of all studies reporting GCS, 34.5% included mild injuries (GCS 14–15). Settings varied from single-center (68.9%) to large multicenter studies (31.1%), representing 47 centers in 18 countries. Of the 29 studies, 20 (68.9%) were conducted retrospectively, with only 9 performing a prospective study.
PRISMA diagram of cohort. PRISMA, Preferred Reporting Items for Systematic reviews and Meta-Analyses.
Included Papers
AIS, abbreviated injury scale; CBF, cerebral blood flow; CCG, critical care guide; CPP, cerebral perfusion pressure; CT, computed tomography; DC, discharge; DIICP, disproportionate increase in intracranial pressure; EVD, external ventricular drain; GC, Granger causality; GCS, Glasgow Coma Scale; GOS, Glasgow Outcome Scale; GOSE, Glasgow Outcome Scale Extended; ICP, intracranial pressure; IH, intracranial hypertension; LCAI, longest cerebrovascular autoregulation impairment; LLR, lower limit of reactivity; MAP, mean arterial pressure; NR, not reported; PAx, pulse amplitude index; PbtO2, partial pressure of brain oxygen; PI, pulsatility index; PRx, pressure reactivity index; PTD, pressure time dose; RAC, the correlation [R] between AMP [A] and CPP [C]; RICH, refractory intracranial hypertension; SD, standard deviation; TBI, traumatic brain injury; TCD, transcranial doppler; ONSD, optic nerve sheath diameter; US ultrasound.
Monitoring protocols
Although 62.1% of the included studies explicitly cited the BTF guidelines and expressed a commitment to following them, the details revealed significant variability in actual implementation. Among these studies, 48.1% employed continuous ICP monitoring—most commonly using intraparenchymal devices (55%) or external ventricular drains (EVDs; 6.8%), and 27.5% utilized a combination. Notably, 10% did not specify the device type. ICP thresholds were frequently set at 20 mmHg (58.6%), while some studies (20.7%) explored dynamic thresholds above 20 mmHg based on cerebrovascular reactivity metrics, and 20.7% did not report an ICP threshold at all. Sampling frequencies ranged from continuous recording at 100 Hz to intermittent hourly measurements. Two (6.8%) did not report whether monitoring was continuous or intermittent. This inconsistency underscores the challenges of achieving uniform conformance to BTF guidelines despite direct references to them in the study protocols.
Outcome measures
Primary outcomes typically included GOS or GOSE at 3, 6, or 12 months post-injury (55%). Secondary outcomes encompassed mortality, ICU LOS, and various physiological parameters. Ten of the 29 studies did not define a secondary outcome. Studies consistently defined poor outcomes as GOS 1–3 or GOSE 1–4, with favorable outcomes as GOS 4–5 or GOSE 5–8.
Predictive models
Out of the 29 studies, 25 assessed the predictive performance of ICP- or cerebral perfusion pressure (CPP)-derived features on outcomes either as primary or secondary objectives. Logistic regression as a model for prediction dominated (n = 21), while a few studies incorporated machine learning (n = 2). There was limited out-of-sample validation (6.9% of the studies), hindering generalizability across most studies.
Key findings
Traditional ICP metrics
Sustained ICP elevation >20 mmHg showed strong correlation with mortality (Table 2). Time spent above 20 mmHg emerged as a significant predictor of unfavorable outcomes. Early ICP patterns (<72 h after admission) demonstrated prognostic value, and age-specific thresholds suggest younger patients (<45 years) may tolerate higher ICPs. Across multiple studies, ICP-related metrics (e.g., average ICP, pulse amplitude index [PAx], pressure reactivity index [PRx]) were consistently found to be significant predictors of unfavorable GOSE at 3–12 months. Higher ICP was more frequently linked to mortality. Additionally, these findings primarily pertain to patients with severe TBI, though some studies included moderate injuries.
Specific Findings Demonstrating ICP Elevation Association with Poor Outcomes
CPP, cerebral perfusion pressure; GOSE, Glasgow Outcome Scale Extended; ICP, intracranial pressure; ICU, intensive care unit; PTD, pressure time dose.
Advanced analytical approaches
The integration of cerebrovascular reactivity indices (e.g., PRx and PAx) enhanced predictive accuracy when added to models incorporating early static predictors (e.g., age, GCS score). Signal complexity analysis revealed associations between ICP waveform characteristics and outcomes. For example, Gao et al., 2016 and 2020 showed that approximate entropy for ICP, mean arterial pressure (MAP), and heart rate, a measure of waveform irregularity and unpredictability of fluctuations, improved prediction performance for outcomes. Guiza et al. extracted more than 1,000 features from minute-by-minute MAP and ICP and showed that they could predict episodes of increased pressure 30 min ahead of an event. Machine learning models incorporating multiple ICP-derived features of complexity—such as waveform analysis, correlation between ICP amplitude and mean ICP, PRx, and CPP—and accounting for the dynamics and nonlinearity showed promise in predicting adverse events. Furthermore, a novel visual aid, the ICEBERG score, demonstrated utility in real-time decision-making using six variables from multimodal monitoring and treatment-related criteria: CPP, ICP, body temperature, sedation depth, arterial partial pressure of CO2 (PaCO2), and blood osmolarity. 39
Discussion
Our review highlights the evolution of ICP monitoring in TBI, moving from a reliance on simple, threshold-based intermittent assessments to more sophisticated analyses that treat ICP as a continuous metric with multiple characteristics, complexity summaries (e.g., approximate entropy), and time-dose (PTD) methods that capture dynamic patterns. Several papers provide evidence, albeit limited, that treatment approaches based on sustained ICP values relate to better clinical outcomes. Studies assessing “sustained ICP” use varied thresholds and durations, ranging from continuous elevation over hours to cumulative “pressure-time dose” thresholds (e.g., 20 mmHg for >30 min). Although more robust data are needed, this finding underscores the importance of ICP not only as a biomarker for predicting outcomes but also as a potential therapeutic target and a marker for monitoring therapy compliance.
This continuous approach to ICP monitoring can offer greater granularity and real-time clinical insight, but it also complicates predictive modeling. Rather than relying on a single value or discrete time points, modeling a dynamic ICP signal involves curating large-volume time-series datasets, accounting for fluctuations and artifacts, and implementing advanced computational methods. Dynamic prediction of elevated ICP events up to 30 min in advance has already been demonstrated, suggesting the possibility of implementing early warning signals for preventive management. However, the choice of monitoring device itself may influence predictive accuracy and patient outcomes. For instance, EVDs, although providing both ICP monitoring and therapeutic CSF drainage, do not offer truly continuous monitoring if they are intermittently open for drainage. In contrast, bolt-anchored ICP monitors allow for continuous monitoring but lack therapeutic capabilities. Additionally, each device type carries different risk profiles; notably, EVDs are associated with higher complication rates—infection and hemorrhage—when compared with intraparenchymal monitors. The differences in monitoring methodology—intermittent versus continuous—may further influence these outcomes. Studies utilizing intermittent monitoring, typically hourly measurements, may assess a different patient population compared with those using continuous monitoring. It would be valuable to explore whether findings in cohorts with continuous monitoring more consistently show a relationship between ICP and outcomes compared with intermittent monitoring. These factors must be weighed carefully when selecting a monitoring modality.
Despite these challenges, incorporating continuous ICP metrics—particularly those capturing cerebrovascular reactivity or waveform morphology—into established prognostic models has yielded promising results. Indeed, adding ICP and CPP features to static models such as the International Mission for Prognosis and Analysis of Clinical Trials (IMPACT) in TBI and the Medical Research Council Corticosteroid Randomization of Significant Head Injury (MRC CRASH) models appears to improve their predictive performance. 40 These models were originally developed using more traditional, single time-point predictor variables (e.g., age, GCS score, pupillary reactivity) and have undergone both internal and external validations. The observed performance enhancements point toward the potential of precision-driven TBI care, provided that new data integration methods, standardized protocols, and additional external validations are undertaken.40–46
The addition of multimodal and systemic time-series variables, including cardiac vital signs and even point-of-care laboratory values, appears to add prognostic value as well.47,48 This lends support to the idea that tracking dynamic physiology, including ICP, is a key to individualized critical care management.
Several emerging approaches support these efforts. Machine-learning techniques applied to multimodal monitoring data, the development of personalized thresholds aligned with individual patient characteristics, and real-time decision support tools such as the ICEBERG score have shown increasing adoption.34,35,39,49,50 Integrating these methods with continuous ICP data may further enhance the predictive power of existing models and enable proactive management strategies.
Priority areas for future research include: (1) standardizing monitoring protocols to facilitate cross-study comparisons, (2) validating novel analytical approaches in diverse populations, (3) incorporating real-time analytics for personalized management, (4) building predictive models that integrate multiple physiological parameters, and (5) defining age-specific, injury-specific, and autoregulatory thresholds.9,35,51 As these methods evolve, more rigorous and large-scale validation studies will be crucial for confirming their clinical utility and reproducibility.
Methodological limitations
Important methodological limitations persist across the studies reviewed, including heterogeneous monitoring durations and protocols, inconsistent reporting of sampling frequencies and time thresholds, limited external validation, inconsistent treatment protocols across centers, and lack of standardization in multimodal monitoring techniques.23,26,28,31,32,34 Furthermore, many studies failed to report denominators clearly, introducing substantial potential for subject selection bias, which further limits the ability to perform rigorous meta-analysis and justifies the use of the scoping review methodology employed here. Moreover, many studies do not detail key methodological aspects, such as the specifics of data cleaning or artifact removal, undermining reproducibility. Our own review is subject to inherent limitations associated with retrospective literature analyses, English-language papers only, and the exclusion of unpublished materials (e.g., conference abstracts), which may introduce publication bias. We could not perform a meta-analysis due to the lack of standardization of data points measured, disparate time windows, and differing analytical techniques in original studies, hindering direct comparisons of various ICP monitoring methods.
Overall, the reporting quality in many of these studies remains relatively low, with incomplete descriptions of protocols and outcome measures. These issues emphasize the need for more rigorous and transparent reporting. Implementing common data elements and standardized protocols for ICP monitoring could enable more meaningful comparisons, facilitate multicenter collaborations, and clarify the complex relationship between ICP dynamics, patient outcomes, and therapeutic interventions. 52 By strengthening the consistency and quality of evidence, the field can progress toward a more comprehensive understanding of how continuous ICP monitoring can be harnessed both as a predictive tool and as a target for personalized treatment strategies.
Conclusion
While ICP monitoring is nearly ubiquitous and clinically useful in TBI management, frequent sampling of ICP as a dynamic predictor of outcomes is evolving. The integration of advanced analytical methods and multimodal monitoring shows promise in enhancing prognostication, though standardization and validation remain important challenges. Future research should focus on developing personalized approaches and validating novel methodologies across diverse populations.
Transparency, Rigor, and Reproducibility Statement
This scoping review was conducted in accordance with the methodological framework proposed by Arksey and O’Malley and adhered to the PRISMA-ScR guidelines. 14 The study protocol was prospectively registered with the Open Science Framework and is publicly accessible at https://osf.io/wpf36/?view_only=8d171152a57143afb91e85b653cb5fac. All search strategies, eligibility criteria, and data extraction methods were defined a priori and applied systematically. As this review involved only publicly available literature and did not include individual patient data, institutional review board approval was not required.
Authors’ Contributions
J.H.K.: Conceptualization, investigation, formal analysis, data curation, writing—original draft, review and editing, project administration. R.C.O.: Formal analysis, data curation, writing—review and editing. A.T.-E.: Conceptualization and formal analysis, writing—review and editing. L.M.D.: Data curation. B.F.: Writing—review and editing. D.O.O.: Writing—review and editing. G.T.M.: Supervision. H.E.H.: Conceptualization, investigation, writing—review and editing, supervision, project administration.
Footnotes
Author Disclosure Statement
All the authors have no competing interest to disclose.
Funding Information
H.E.H. is supported by NINDS 1K23NS110828.