Detrended fluctuation analysis of intracranial pressure predicts outcome following traumatic brain injury

Abstract

Detrended fluctuation analysis (DFA) is a recently developed technique suitable for describing scaling behavior of variability in physiological signals. The purpose of this study is to explore applicability of DFA methods to intracranial pressure (ICP) signals recorded in patients with traumatic brain injury (TBI). In addition to establishing the degree of fit of the power-law scaling model of detrended fluctuations of ICP in TBI patients, we also examined the relationship of DFA coefficients (scaling exponent and intercept) to: 1) measures of initial neurological functioning; 2) measures of functional outcome at six month follow-up; and 3) measures of outcome, controlling for patient characteristics, and initial neurological status. In a sample of 147 moderate-to-severely injured TBI patients, we found that a higher DFA scaling exponent is significantly associated with poorer initial neurological functioning, and that lower DFA intercept and higher DFA scaling exponent jointly predict poorer functional outcome at six month follow-up, even after statistical control for covariates reflecting initial neurological condition. DFA describes properties of ICP signal in TBI patients that are associated with both initial neurological condition and outcome at six months postinjury.

Fig. 1

Average empirical ICP DFA profile for the total sample of TBI patients is depicted by the solid curve. The abscissa represents local model piecewise time scales n, ranging (on a log₁₀ scale) from about 30 s to 24 h. The ordinate represents the log₁₀ rms amplitude F (n) of fluctuations or residuals from the corresponding family of local detrended models of the cumulated zero mean signals. Also shown on the figure are the theoretical DFA profiles corresponding to pure random noise (DFA scaling exponent = 0.5), simple monofractal (scaling exponent = 1.0), and an integrated random walk (scaling exponent = 1.5).

Fig. 2

Examples of mean ICP pressure series (each 8 h in length) for two subjects and corresponding DFA fits. Subject A [see Fig. 2(a)] experienced good functional outcome at six month follow-up, while subject B [see Fig. 2(b)] died before the six month assessment point. DFA lines for subjects A and B, fit to the individual profiles at time scales n between 30 s and 2 h, are presented in Fig. 2(c). The unity slope DFA scaling exponent = 1 line is overplotted on the figure for comparison.

Fig. 3

Plots of means and error bars representing the 95% confidence intervals about the means for ICP DFA intercept [see Fig. 3(a)] and ICP DFA scaling exponent [see Fig. 3(b)] as a function of GCS-M, a clinical rating of neurological function at the time of admission to the ICU after TBI (scale ranges from 1–6, higher values coding better function).

Fig. 4

Plots of means and error bars representing the 95% confidence intervals about the means for ICP DFA intercept [see Fig. 4(a)] and DFA scaling exponent [see Fig. 4(b)] as a function of GOSE, a clinical rating of functional outcome at six month follow-up (scale ranges from 1 = dead to 8 = upper good recovery).

Fig. 5

Error bar plots representing the 95% confidence intervals for the standardized odds ratios for survival until six months, and for favorable outcome at six months (GOSE > 4), based on DFA intercept [see Fig. 5(a)] and DFA scaling exponent [see Fig. 5(b)]. The odds ratios were determined by binary logistic regression, first without covariates, and then, for comparison, with covariate control of age, gender, GCS-M, mean ICP level, and craniectomy.