Dynamic cerebral autoregulation (AR) assessments , such as the pressure reactivity index (PRx), can be used to determine a theoretically optimum CPP value (CPPopt) to be targeted in brain injured patients . Presented here are two methodologies to improve the standard CPPOpt algorithm.
Method 1: The standard method, by Aries , “bins” the PRx values over a CPP range and summarises these via a mean. The binned data can be skewed (Figure 1). Therefore, we propose the use of a median summary value to fit a robust linear model (RLM).
Method 2: The standard method uses data reduction steps to enable the squared polynomial fit. It is proposed that no binning of data is performed and only outliers be removed, then the data fitted with a simple generalised additive model (GAM) and finally the minimum turning point for the fit found.
10 patients from neuro-ICU had CPPOpt calculated via each method, which were compared via repeated measures ANOVA.
The standard model fails to calculate CPPopt in 3 patients and the RLM in 4 patients. The GAM method finds a CPPOpt in all cases with no statistical difference in values (p = 0.0716).
 Panerai RB. Assessment of cerebral pressure autoregulation in humans--a review of measurement methods. Physiological Measurements. 1998; 19:305-338
 Aries MJ, et al. Continuous determination of optimal cerebral perfusion pressure in traumatic brain injury. Crit Care Med 2012; 40: 2456-2463
Figure1: Patient CPPOpt method fits with the CPPOpt shown as a vertical dashed line.
|Group 1, n= 63||Group 2 = 17||Group 3 = 15|
|Age (median, [IQR])||39 [20-58]||32 [24-40]||26 [18-33]m|
|GCS Motor Score||5||2||2|
|Left Pupil Reacting (%)||100||0||100|
|Right Pupil Reacting (%)||97||29||100|
|Hypoxia Present (%)||0||53||66|
|Hypotension Present (%)||2||41||33|
|Hematocrit (%, median [IQR])||38 [31-42]||34 [30-36]||34 [30-35]|
|Glucose (mmol/l, median [IQR])||8.6 [6.5-9.9]||6.8 [6.1-8.3]||7.0 [5.7-8.4]|
|GOSe Code (median)||6||3||5|
TABLE 1 – cluster groups based on physiological data. Median GOSe was calculated for each group after clustering.
|Group 1, n= 94||Group 2 = 77||Group 3 = 16|
|SpO2 (%)||100 [99-100]||100 [98-100]||100 [99-100]|
|HR (bpm)||76 [65-90]||80 [67-95]||75 [65-88]|
|MAP (mmHg)||77 [70-84]||88 [81-95]||104 [96-114]|
|ICP (mmHg)||12 [8-17]||14 [8-19]||20 [14-26]|
|CPP (mmHg)||64 [57-71]||75 [67-82]||80 [68-94]|
|GOSe Code (median)||5||5||6|
TABLE 2 – cluster groups based on physiological data. Physiological values are median [IQR]. Median GOSe was calculated for each group after clustering.
We describe the use of cluster analysis, a form of unsupervised learning, to identify similar groups of TBI patients within the multi-centre BrainIT database  using: A) admission data and B) physiological variables recorded every minute during their NICU admission.
Data used for calculation of the IMPACT prognostic model  was selected from the Admission data table. The Physiological data table was cleaned, summarised and the physiological variables of SpO2, HR, MAP, ICP and CPP selected. Agglomerative hierarchical cluster analysis was then performed separately on both data sets to identify 3 distinct groups of patients. Each patient’s group allocations were then compared using similarity tests of cluster analysis output.
There was sufficient data to analyse 95 patients based on admission data and 187 patients on physiological data. Selected group characteristics and Glasgow Outcome Scores are outlined in Tables 1 & 2. There was no tendency for patients to cluster together in both analyses.
Cluster analysis can automatically detect groups of TBI patients with differing outcomes based on both admission and continuous physiological data. Patients do not tend to stay within the same clusters between analyses and it will be of interest to investigate how their clinical management varied.
In patients being investigated for CSF abnormality there have been anecdotal reports that more periods of raised ICP or B-waves are observed if overnight monitoring is extended beyond 24 hours. This study compares a series of metrics of raised ICP and B-wave activity across 24 Vs 48 hour recordings of ICP.
Twenty Nine patients with CSF abnormalities (NPH-2,Chiari I–5,IIH–3,Tumour/Cyst–3,other etiology–15) were ICP monitored for a minimum of 48 hours. Metrics of raised ICP (mean, median, max, min, %Time ICP > 15 mmHg & B-wave activity) were compared across Day 1 Vs Day 2, Overnight (ON) 1 Vs Overnight (ON) 2. Data was analysed using repeated measures ANOVA.
Summary Statistics (see Fig 1) include: Day1 vs Day2 mean resp: 6.3 mmHg (-11-22) Vs 6.7 mmHg (-10-19). ON1 vs ON 2 mean resp: 8.9 mmHg (-9-22) Vs 9.2mmHg (-7-20). Day1 vs Day2 Bwaves resp: 4 (0-10) Vs 4 (0-10). ON1 vs ON 2 Bwaves resp: 2 (-9-5) Vs 2 (0-5). Repeated Measures ANOVA showed significant differences between Day1 and ON2 for mean (p = 0.037), median (p = 0.0213) and minimum diastolic ICP (p = 0.0155). There were no other statistically significant differences found although in many instances there was a trend towards higher calculated metrics on the second overnight recording.
At least Overnight recording of ICP is recommended with a suggestion that more events may be detected on the 2nd overnight recording. Further study is warranted in particular to include admission pathology to the model when study recruitment is large enough.
Fig1. Box Plot of Mean ICP Vs Factor where Factor 2 = Day1, 3 = Day2, 4 = ON1, 5 = ON2, 6 = Day1-ON1, 7 = Day2-ON2
This is a collaborative project between Aridhia and NHS Greater Glasgow & Clyde (NHSGGC). The aim of the project is to develop software which will enable clinically important physiological models and analyses to be implemented more quickly into clinical practice. This will provide the basis for better treatment and more cost-effective and sustainable healthcare by closing the loop between clinical research and practice.
This will be achieved by: 1) enabling Aridhia’s software to automatically connect with in-hospital patient monitoring devices via collaboration with Philips Medical on their Intensive Care Unit (ICU) eRecord system; 2) linking Aridhia’s existing platforms: AnalytiXagility and Cloud Foundry, to provide storage of high frequency data, rapid application of clinical analysis algorithms and the integration of "apps" to allow clinicians to control the analysis; and 3) presenting the results of analyses back at the patient bedside. Ultimately, we aim to collaboratively develop a prototype of this system. The project will deliver the designs, the prototype and the clinical evidence for market clearance that would add to Aridhia's existing Software as a Service offering.