Manuel Cabeleira

High-resolution, waveform-level data from bedside monitors carry important information about a patient’s physiology but is also polluted with artefactual data. Manual mark-up is the standard practice for detecting and eliminating artefacts, but it is time-consuming, prone to errors, biased and not suitable for real-time processing. In this paper we present a novel automatic artefact detection technique based on a Symbolic Aggregate approXimation (SAX) technique which makes it possible to represent individual pulses as ‘words’. It does that by coding each pulse with a specified number of letters (here six) from a predefined alphabet of characters (here six). The word is then fed to a support vector machine (SVM) and classified as artefactual or physiological. To define the universe of acceptable pulses, the arterial blood pressure from 50 patients was analysed, and acceptable pulses were manually chosen by looking at the average pulse that each ‘word’ generated. This was then used to train a SVM classifier. To test this algorithm, a dataset with a balanced ratio of clean and artefactual pulses was built, classified and independently evaluated by two observers achieving a sensitivity of 0.972 and 0.954 and a specificity of 0.837 and 0.837 respectively.

Background: To date, the cerebral physiologic consequences of persistently elevated intracranial pressure (ICP) have been based on either low-resolution physiologic data or retrospective high-frequency data from single centers. The goal of this study was to provide a descriptive multi-center analysis of the cerebral physiologic consequences of ICP, comparing those with normal ICP to those with elevated ICP. Methods: The Collaborative European NeuroTrauma Effectiveness Research in Traumatic Brain Injury (CENTER-TBI) High-Resolution Intensive Care Unit (HR-ICU) sub-study cohort was utilized. The first 3 days of physiologic recording were analyzed, evaluating and comparing those patients with mean ICP < 15 mmHg versus those with mean ICP > 20 mmHg. Various cerebral physiologic parameters were derived and evaluated, including ICP, brain tissue oxygen (PbtO₂), cerebral perfusion pressure (CPP), pulse amplitude of ICP (AMP), cerebrovascular reactivity, and cerebral compensatory reserve. The percentage time and dose above/below thresholds were also assessed. Basic descriptive statistics were employed in comparing the two cohorts. Results: 185 patients were included, with 157 displaying a mean ICP below 15 mmHg and 28 having a mean ICP above 20 mmHg. For admission demographics, only admission Marshall and Rotterdam CT scores were statistically different between groups (p = 0.017 and p = 0.030, respectively). The high ICP group displayed statistically worse CPP, PbtO₂, cerebrovascular reactivity, and compensatory reserve. The high ICP group displayed worse 6-month mortality (p < 0.0001) and poor outcome (p = 0.014), based on the Extended Glasgow Outcome Score. Conclusions: Low versus high ICP during the first 72 h after moderate/severe TBI is associated with significant disparities in CPP, AMP, cerebrovascular reactivity, cerebral compensatory reserve, and brain tissue oxygenation metrics. Such ICP extremes appear to be strongly related to 6-month patient outcomes, in keeping with previous literature. This work provides multi-center validation for previously described single-center retrospective results.

Brain tissue oxygen (PbtO₂) monitoring in traumatic brain injury (TBI) has demonstrated strong associations with global outcome. Additionally, PbtO₂ signals have been used to derive indices thought to be associated with cerebrovascular reactivity in TBI. However, their true relationship to slow-wave vasogenic fluctuations associated with cerebral autoregulation remains unclear. The goal of this study was to investigate the relationship between slow-wave fluctuations of intracranial pressure (ICP), mean arterial pressure (MAP) and PbtO₂ over time. Using the Collaborative European NeuroTrauma Effectiveness Research in Traumatic Brain Injury (CENTER-TBI) high resolution ICU sub-study cohort, we evaluated those patients with recorded high-frequency digital intra-parenchymal ICP and PbtO₂ monitoring data of a minimum of 6 h in duration. Digital physiologic signals were processed for ICP, MAP, and PbtO₂ slow-waves using a moving average filter to decimate the high-frequency signal. The first 5 days of recording were analyzed. The relationship between ICP, MAP and PbtO₂ slow-waves over time were assessed using autoregressive integrative moving average (ARIMA) and vector autoregressive integrative moving average (VARIMA) modelling, as well as Granger causality testing. A total of 47 patients were included. The ARIMA structure of ICP and MAP were similar in time, where PbtO₂ displayed different optimal structure. VARIMA modelling and IRF plots confirmed the strong directional relationship between MAP and ICP, demonstrating an ICP response to MAP impulse. PbtO₂ slow-waves, however, failed to demonstrate a definite response to ICP and MAP slow-wave impulses. These results raise questions as to the utility of PbtO₂ in the derivation of cerebrovascular reactivity measures in TBI. There is a reproducible relationship between slow-wave fluctuations of ICP and MAP, as demonstrated across various time-series analytic techniques. PbtO₂ does not appear to reliably respond in time to slow-wave fluctuations in MAP, as demonstrated on various VARIMA models across all patients. These findings suggest that PbtO₂ should not be utilized in the derivation of cerebrovascular reactivity metrics in TBI, as it does not appear to be responsive to changes in MAP in the slow-waves. These findings corroborate previous results regarding PbtO₂ based cerebrovascular reactivity indices.

Background Impaired cerebrovascular reactivity in adult traumatic brain injury (TBI) is known to be associated with poor outcome. However, there has yet to be an analysis of the association between the comprehensively assessed intracranial hypertension therapeutic intensity level (TIL) and cerebrovascular reactivity. Methods Using the Collaborative European Neuro Trauma Effectiveness Research in TBI (CENTER-TBI) high-resolution intensive care unit (ICU) cohort, we derived pressure reactivity index (PRx) as the moving correlation coefficient between slow-wave in ICP and mean arterial pressure, updated every minute. Mean daily PRx, and daily % time above PRx of 0 were calculated for the first 7 days of injury and ICU stay. This data was linked with the daily TIL-Intermediate scores, including total and individual treatment sub-scores. Daily mean PRx variable values were compared for each TIL treatment score via mean, standard deviation, and the Mann U test (Bonferroni correction for multiple comparisons). General fixed effects and mixed effects models for total TIL versus PRx were created to display the relation between TIL and cerebrovascular reactivity. Results A total of 249 patients with 1230 ICU days of high frequency physiology matched with daily TIL, were assessed. Total TIL was unrelated to daily PRx. Most TIL sub-scores failed to display a significant relationship with the PRx variables. Mild hyperventilation (p < 0.0001), mild hypothermia (p = 0.0001), high levels of sedation for ICP control (p = 0.0001), and use vasopressors for CPP management (p < 0.0001) were found to be associated with only a modest decrease in mean daily PRx or % time with PRx above 0. Conclusions Cerebrovascular reactivity remains relatively independent of intracranial hypertension therapeutic intensity, suggesting inadequacy of current TBI therapies in modulating impaired autoregulation. These findings support the need for investigation into the molecular mechanisms involved, or individualized physiologic targets (ICP, CPP, or Co2) in order to treat dysautoregulation actively.

Incorrect family name of Nino Stocchetti.

Background Monitoring cerebrovascular reactivity in adult traumatic brain injury (TBI) has been linked to global patient outcome. Three intra-cranial pressure (ICP)-derived indices have been described. It is unknown which index is superior for outcome association in TBI outside previous single-center evaluations. The goal of this study is to evaluate indices for 6- to 12-month outcome association using uniform data harvested in multiple centers. Methods Using the prospectively collected data from the Collaborative European NeuroTrauma Effectiveness Research in TBI (CENTER-TBI) study, the following indices of cerebrovascular reactivity were derived: PRx (correlation between ICP and mean arterial pressure (MAP)), PAx (correlation between pulse amplitude of ICP (AMP) and MAP), and RAC (correlation between AMP and cerebral perfusion pressure (CPP)). Univariate logistic regression models were created to assess the association between vascular reactivity indices with global dichotomized outcome at 6 to 12 months, as assessed by Glasgow Outcome Score–Extended (GOSE). Models were compared via area under the receiver operating curve (AUC) and Delong’s test. Results Two separate patient groups from this cohort were assessed: the total population with available data (n = 204) and only those without decompressive craniectomy (n = 159), with identical results. PRx, PAx, and RAC perform similar in outcome association for both dichotomized outcomes, alive/dead and favorable/unfavorable, with RAC trending towards higher AUC values. There were statistically higher mean values for the index, % time above threshold, and hourly dose above threshold for each of PRx, PAx, and RAC in those patients with poor outcomes. Conclusions PRx, PAx, and RAC appear similar in their associations with 6- to 12-month outcome in moderate/severe adult TBI, with RAC showing tendency to achieve stronger associations. Further work is required to determine the role for each of these cerebrovascular indices in monitoring of TBI patients.

Background Continuous assessment of cerebral compensatory reserve is possible using the moving correlation between pulse amplitude of intra-cranial pressure (AMP) and intra-cranial pressure (ICP), called RAP. Little is known about the behavior and associations of this index in adult traumatic brain injury (TBI). The goal of this study is to evaluate the association between admission cerebral imaging findings and RAP over the course of the acute intensive care unit stay. Methods We retrospectively reviewed 358 adult TBI patients admitted to the Addenbrooke’s Hospital, University of Cambridge, from March 2005 to December 2016. Only non-craniectomy patients were studied. Using archived high frequency physiologic signals, RAP was derived and analyzed over the first 48 h and first 10 days of recording in each patient, using grand mean, percentage of time above various thresholds, and integrated area under the curve (AUC) of RAP over time. Associations between these values and admission computed tomography (CT) injury characteristics were evaluated. Results The integrated AUC, based on various thresholds of RAP, was statistically associated with admission CT markers of diffuse TBI and cerebral edema. Admission CT findings of cortical gyral effacement, lateral ventricle compression, diffuse cortical subarachnoid hemorrhage (SAH), thickness of cortical SAH, presence of bilateral contusions, and subcortical diffuse axonal injury (DAI) were all associated with AUC of RAP over time. Joncheere-Terpstra testing indicated a statistically significant increase in mean RAP AUC across ordinal categories of the abovementioned associated CT findings. Conclusions RAP is associated with cerebral CT injury patterns of diffuse injury and edema, providing some confirmation of its potential measurement of cerebral compensatory reserve in TBI.

Objectives: Modern neuro-critical care units generate high volumes of data. These data originate from a multitude of devices in various formats and levels of granularity. We present a new data format intended to store these data in an ordered and homogenous way. Material and methods: The adopted data format was based on the hierarchical model, HDF5, which is capable of dealing with a mixture of small and very large datasets with equal ease. It is possible to access and manipulate individual data elements directly within a single file, and this is extensible and versatile. Results: The file structure that was agreed divided the patient data into four different groups: ‘Annotations’ for clinical events and sporadic observations, ‘Numerics’ for all the low-frequency data, ‘Waves’ for all the high-frequency data and ‘Summaries’ for the trend data and calculated parameters. The addition of attributes to every group and dataset makes the file self-described. More than 200 files have been successfully collected and stored using this format. Conclusion: The new file format was implemented in ICM+ software and validated as part of a collaboration with participating centres across Europe.

Objectives: Optimal cerebral perfusion pressure (CPPopt) is a concept that uses the pressure reactivity (PRx)–CPP relationship over a given period to find a value of CPP at which PRx shows best autoregulation. It has been proposed that this relationship be modelled by a U-shaped curve, where the minimum is interpreted as being the CPP value that corresponds to the strongest autoregulation. Owing to the nature of the calculation and the signals involved in it, the occurrence of CPPopt curves generated by non-physiological variations of intracranial pressure (ICP) and arterial blood pressure (ABP), termed here “false positives”, is possible. Such random occurrences would artificially increase the yield of CPPopt values and decrease the reliability of the methodology. In this work, we studied the probability of the random occurrence of false-positives and we compared the effect of the parameters used for CPPopt calculation on this probability. Materials and methods: To simulate the occurrence of false-positives, uncorrelated ICP and ABP time series were generated by destroying the relationship between the waves in real recordings. The CPPopt algorithm was then applied to these new series and the number of false-positives was counted for different values of the algorithm’s parameters. Results: The percentage of CPPopt curves generated from uncorrelated data was demonstrated to be 11.5%. Conclusion: This value can be minimised by tuning some of the calculation parameters, such as increasing the calculation window and increasing the minimum PRx span accepted on the curve.