Age Related Thresholds for Management of Severe Head Injury in Paediatrics: An Extension of STARSHIP

1. Background Neuro-morbidity associated with TBI in children is high and there is limited available evidence for strategies to improve outcomes1. The last international guideline for management of severe TBI in children were mainly based on level III evidence due to lack of robust evidence for age related thresholds for treatment targets in paediatric intensive care (PICU) like intracranial pressure (ICP) and cerebral perfusion pressure (CPP), or availability of methods to individualise these targets, like indices of cerebral autoregulation (CA)2. As the paediatric age range encompasses a developmental trajectory, combined with the disease heterogeneity, single treatment thresholds are insufficient for the best outcomes in paediatric TBI (pTBI). Given the ethical challenges and difficulties with doing research in children and poor yield from randomized controlled trials, there is increasing recognition of using big data to answer research questions.

   Investigators have recently completed the multicentre prospective research database study, STARSHIP including 135 pTBI patients with promising results confirming PRx association with outcome irrespective of the initial disease severity, ICP and CPP, and centre specific differences in management3. Investigators have also identified dynamic temporal trends of PRx and its thresholds for outcome. In this process, investigators have also created a rich research database with time-synchronized physiological information with clinical and outcome data which has been used to further explore the thresholds for ICP and CPP related to outcome. However, the small number of patients in each age category does not allow identification of age-appropriate thresholds. STARSHIP has ethics approval as a research database which allows for the database to be combined with other databases. There are similar datasets that exists in individual centers, (for eg, Cambridge pTBI dataset prior to STARSHIP, Birmingham Children's Hospital in the UK; Phoenix Children's Hospital in the United States of America; Uppsala, Leuven and Rotterdam in Europe, to name a few). Given the small numbers of pTBI, all these centers have come together to collaborate to make a meaningful impact by combining datasets.

   Through this project, Investigators propose to combine similar existing datasets from other UK and international centres to create a larger cohort (350- 500 patients) of pTBI patients. This will facilitate statistically meaningful analysis of age-stratified thresholds, namely ICP, CPP and PRx using the same analytical methods as used for STARSHIP3,4. The larger dataset analysis may help stratify CA guided treatment targets for neuroprotection across the paediatric age range.

2. Rationale \& Theoretical Framework Investigators will use the combined dataset to identify age-appropriate treatment thresholds for neuroprotection, i.e., ICP and CPP in severe pTBI for improved outcomes as well test if the PRx thresholds are different in different age ranges. The results of this study would be crucial to fill some of the gaps in current evidence base for managing children with severe TBI. This will help with planning future prospective interventional studies to evaluate if the age-appropriate and CA guided treatment thresholds can improve outcomes in children with TBI.

   The study will help stratification of age-based thresholds for treatment targets used in managing severe pTBI with the hypothesis that there would be different thresholds for ICP and CPP depending on the patients age and autoregulatory status. The results should inform the international evidence-based guidelines as well as future research by creating foundation for phase II/III studies to test the hypothesis in prospective clinical studies.

3. Research Question/Aims and Objectives:

   To identify age-stratified thresholds of treatment targets for neuroprotection, namely intracranial pressure (ICP), cerebral perfusion pressure (CPP) and cerebral autoregulation (PRx) in children with traumatic brain injury.

4. Study Design/ Methods:

   Investigators will use the combined dataset to identify age-appropriate treatment thresholds for neuroprotection, i.e., ICP and CPP in severe pTBI for improved outcomes as well test if the PRx thresholds are different in different age ranges. The results of this study would be crucial to fill some of the gaps in current evidence base for managing children with severe TBI. This will help with planning future prospective interventional studies to evaluate if the age-appropriate and CA guided treatment thresholds can improve outcomes in children with TBI. The study will help stratification of age-based thresholds for treatment targets used in managing severe pTBI with the hypothesis that there would be different thresholds for ICP and CPP depending on the patients age and autoregulatory status. The results should inform the international evidence-based guidelines as well as future research by creating foundation for phase II/III studies to test the hypothesis in prospective clinical studies.

5. Study setting: Inclusion criteria- Datasets containing at least 15 pTBI patients upto 18 years of age with advanced neuromonitoring data, consisting of high-resolution (>100Hz) waveform data with minimum 6 hours of recording along with clinical and demographic details defining severity of pTBI and 6 months follow-up post-ictus will be included.

   i) Full-resolution physiological monitored data in PICU (mainly ICP and arterial blood pressure, ABP; if available end-tidal carbon dioxide (ETCO2), electrocardiogram (ECG), Temperature) ii) Clinical details- demographic data (age, sex, weight, injury details including type, severity and mechanism, post resuscitation Glasgow coma scale (GCS), pre-hospital cardio-pulmonary resuscitation (CPR), pre-hospital hypoxia, hypotension, hyperthermia, admission pupillary size and response], disease and injury severity scores, treatment for traumatic brain injury (medical \& surgical), head computerized tomography (CT) scan at admission). If available, ventilatory parameters (oxygenation and ventilation parameters), laboratory analyses (Haemoglobin, sodium, glucose, lactate etc.) will also be collected.

   iii) Outcome follow-up at least upto 6 months with Glasgow Outcome Score extended for Pediatrics (GOSEP) or GOS or Pediatric Cerebral Performance Category (PCPC) and if available upto 12 months outcome.

6. Sample \& Statistical Methods: After completing the data sharing agreements and processes required to combine the datasets in the 1st quarter, investigators will complete harmonization of physiological measurements as well as clinical and follow-up data from the different datasets, including time synchronization of the clinical annotations with the physiological recordings. Investigators will collect the pre-processed data from each centre which would have undergone the artefact cleaning and preliminary analysis; depending on the data sharing agreement, will either get the hdf files or the processed data from the ICM+ (Intensive Care Monitoring) with harmonized configurations. For the unprocessed data, after pre-processing including harmonization and careful, standardised, curation, time-averaged values of ICP, CPP, PRx and CPPopt will be calculated using waveform time integration over 60-sec intervals. This will be followed by general data audit and evaluation of per-centre differences. Investigators will also take account of the missingness of the data/variables in the datasets and use valid imputation methods where applicable. This will lead to creation of a single, homogeneous, data set for further processing. Finally, a set of statistical models will be applied to the data in age-stratified groups. Data will be hosted in the Brain Physics Lab, University of Cambridge server and the analysis will be performed under supervision of the lab which has 3 decades of experience in data curation, handling and analysis. The calculated ICP and CPP values would be used to identify the thresholds of each against the 6- and 12-months outcome in 6 different age categories: \<1, 1-3, 3-5, 5-8, 8-11, 11-16 years, respectively and receiver operating curves created for individual age categories. Previously validated methods will be used to calculate ICP, CPP and PRx dose response above or below these identified thresholds to further understand the strength of their association with outcome. There will be further exploratory analysis of ICP/CPP targets in relation to the state of CA.

Statistical methods: Appropriate statistical methods will be employed to account for initial disease severity (propensity score calculation and sliding dichotomy, propensity score matching), and centre related differences (mixed effects model) which have been used for analysing STARSHIP results. Bias reducing techniques, like 5-fold cross-validation, will be applied, where appropriate, for age dependent thresholds derivations. Continuous variables will be assessed for normality and expressed as mean (standard deviation) or median (interquartile range), depending on the underlying distribution of the data. Categorical variables will be reported as counts and proportions. Differences in physiological values between survivors and non-survivors will be interrogated with the Mann-Whitney U-test for non-normally distributed continuous variables, Student t-test for normally distributed continuous variables and Fisher exact test for categorical variables. All tests will be two-tailed and unadjusted for multiple comparisons. Spearman rank correlation coefficient (ρ) will be used to see correlations between variables. Identification of critical threshold values of PRx for predicting unfavorable outcome will be performed using iterative chi-square tests, as per the method used in STARSHIP analysis. To account for any possible confounding effect of the differing periods of monitoring and day post admission on which monitoring starts on outcome associations, ordinal logistic regression analysis will be performed. The independent effect of each continuously monitored variable on the outcome will be assessed by including the length of monitoring and monitoring commencement day post admission as covariates in the regression models. For those predictors that are concentrated around 0 (e.g., PRx and percentages of monitoring time expressed as fractions), the raw values will be multiplied by 10 to enable a meaningful interpretation of odds ratios in the regression analyses. The effect of time on PRx will be assessed using repeated-measures analysis of variance with each patient treated as a random effect.

Risks/Benefits: There are no risks as the project only involves retrospective analysis of previously collected anonymised datasets from the routinely monitored and collected data from the bedside monitors and clinical notes. The datasets will not include any patient identifiable information including even year of birth, investigators will only collect the patient's age at the time of ictus. The project does not accrue any direct benefit to the participants but will provide the scientific community with much needed information and resource to answer important questions which may help in improving management and outcome of children with severe TBI in the future.