Physiological Signal Entropy in Pediatric Traumatic Brain Injury: Looking Beyond the Obvious: A STARSHIP Study

Abstract

Objectives: Multimodality monitoring based prognostication in pediatric traumatic brain injury (TBI) relies heavily on the evaluation of instantaneous and easily interpretable monitoring values. Entropy quantifies the level of disorder within a system reflecting overall activity of sensitive closed-loop feedback homeostatic mechanisms. Multiscale entropy (MSE) assesses entropy across different time scales to examine various physiologic systems and processes that operate across different time scales. The current understanding of MSE suggests that low entropy reflects increased rigidity of the various homeostatic control systems, reflecting underperformance of mechanisms such as cerebral autoregulation. This hypothesis-generating retrospective study explores the value of MSE for prognostication after pediatric TBI.

Design: Retrospective analysis of data from an observational multicenter database.

Setting: Ten PICUs across the United Kingdom.

Patients: One hundred thirty-five children with severe pediatric TBI receiving invasive neuromonitoring between 2018 and 2022.

Interventions: None.

Measurements and main results: MSE was calculated based on 10-second time trends of different biosignals (incl. blood pressure, heart rate, intracranial pressure [ICP]). MSE metrics were assessed using univariable and multivariable (logistic regression with backward stepwise elimination and sliding dichotomy) methods. Last, correlation coefficients between MSE and clinical or monitoring metrics were assessed. Decreased MSE of physiologic biosignals were associated with worse outcomes and remained associated with outcomes when added to multivariable analyses. Within multivariable logistic regression analyses (covariates: Injury Severity Score [ISS], Rotterdam score, ICP, pressure reactivity index [PRx]), the resulting odds ratios (ORs) were: MSE arterial blood pressure (abp: OR, 0.83; p = 0.014), MSE cerebral perfusion pressure (cpp: OR, 0.86; p = 0.024), and MSE icp (OR, 0.87; p = 0.029). MSE displayed weak associations with clinical parameters reflecting higher TBI severity (i.e., ISS, Abbreviated Injury Scale, Glasgow Coma Scale, etc.) but moderate correlations with PRx (correlation coefficients: MSE abp, -0.47; MSE cpp, -0.55) and ICP (MSE abp, -0.3).

Conclusions: Biosignal complexity is a promising tool for improving individualized prognostication after pediatric TBI. Our results further underpin the association between MSE and the function of physiologic autoregulatory mechanisms.

Keywords: entropy; multimodality neuromonitoring; neurocritical care; prognostication; traumatic brain injury.

Figure 1.

Multiscale entropy (MSE) metrics stratified by outcome and univariable/multivariable receiver operating characteristic curves. A, MSE stratified by mortality (Glasgow Outcome Scale-Extended Pediatric Version [GOSE-Peds] 1–7 vs. 8; left) or outcome (GOSE-Peds 1–4 vs. 5–8; right) using boxplots. The corresponding statistical analysis is described in Table 2. B, Area under the receiver operating characteristic curves (AUCs). On the left, the receiver operating characteristic curves of the different relevant MSE metrics as well as ICP and CPP. The resulting AUC values were 0.74 (95% CI, 0.65–0.84) for MSE diastolic blood pressure (dbp), 0.70 (95% CI, 0.60–0.79) for MSE icp, 0.70 (95% CI, 0.60–0.80) for MSE cpp, 0.64 (95% CI, 0.54–9.74) for ICP, and 0.56 (95% CI, 0.45–0.67) for CPP, respectively. On the right, the different AUCs of the multivariable models are displayed, which included either just the traditional prognostic parameters (i.e., Injury Severity Score, Rotterdam score, ICP, and CPP) resulting in the base model, or in addition either one of the relevant MSE metrics (i.e., MSE dbp, MSE icp, MSE cpp, respectively). Overall, the models yielded AUCs of 0.74 (95% CI, 0.65–0.83) for the base model, 0.78 (95% CI, 0.70–0.87) for the base model including MSE dbp, 0.76 (95% CI, 0.67–0.85) for the base model including MSE icp, and 0.75 (95% CI, 0.66–0.84) for the base model including MSE cpp. abp = arterial blood pressure, amp = intracranial pressure amplitude, sbp = systolic blood pressure.

Figure 2.

Intercorrelation between clinical and multimodality monitoring metrics. Correlation between multiscale entropy (MSE) and clinical (A) and multimodal monitoring (B) metrics are shown. Clinical metrics were assessed against MSE metrics using Spearman correlation. Multimodality monitoring metrics were assessed against MSE metrics using Pearson correlation. The correlation coefficients are shown as values with the color reflecting the direction and the intensity reflecting the strength of the correlation. abp = arterial blood pressure, AIS = Abbreviated Injury Scale, amp = intracranial pressure amplitude, cpp = cerebral perfusion pressure, CPPopt = optimal cerebral perfusion pressure, dbp = diastolic blood pressure, GCS = Glasgow Coma Scale, HR = heart rate, icp = intracranial pressure, ISS = Injury Severity Score, PAx = pulse amplitude index, PRx = pressure reactivity index, sbp = systolic blood pressure.