Dynamic assessment of signal entropy for prognostication and secondary brain insult detection after traumatic brain injury.

backgroundEntropy quantifies the level of disorder within a system. Low entropy reflects increased rigidity of homeostatic feedback systems possibly reflecting failure of protective physiological mechanisms like cerebral autoregulation. In traumatic brain injury (TBI), low entropy of heart rate and intracranial pressure (ICP) predict unfavorable outcome. Based on the hypothesis that entropy is a dynamically changing process, we explored the origin and value of entropy time trends.

methods232 continuous recordings of arterial blood pressure and ICP of TBI patients with available clinical information and 6-month outcome (Glasgow Outcome Scale) were accessed form the Brain Physics database. Biosignal entropy was estimated as multiscale entropy (MSE) that aggregates entropy at several time scales (20 coarse graining steps starting from 0.1 Hz). MSE was calculated repeatedly for consecutive, overlapping 6 h segments. Percentage monitoring time (ptime) or dosage (duration*level/hour) below different cutoffs were evaluated against outcome using univariable and multivariable analyses, and propensity score matching. Associations to clinical and monitoring metrics were explored using correlation coefficients. Lastly, individual secondary brain insults (deviations in ICP, cerebral perfusion pressure - CPP, or pressure reactivity) were assessed in relation to changes in MSE.

resultsIncreased MSE abp and MSE cpp ptime (OR 1.28 (1.07-1.58) and OR 1.50 (1.16-2.03) for MSE abp and cpp respectively) and dose (OR 1.12 (1.02-1.27) and OR 1.21 (1.06-1.46) for MSE abp and cpp respectively) were associated with poor outcome even after propensity score matching within multivariable models correcting for ICP, CPP, and the pressure reactivity index. MSE trajectories differed significantly dependent on outcome. The entropy metrics displayed weak correlations to clinical parameters. Individual episodes of deranged physiology were associated with decreases in the MSE metrics from both cerebral and systemic biosignals.

conclusionsBiosignal entropy of changes dynamically after TBI. The assessment of these variations augments individualized, dynamic, outcome prognostication and identification of secondary cerebral insults. Additionally, these explorations allow for further exploitation of the extensive physiological data lakes acquired for each TBI patient within an intensive care environment.