Temperature variability analysis using wavelets in critically ill patients during systemic inflammation

Abstract

Objectives: The aim of this study was to investigate the longitudinal changes of temperature variability using wavelets transformation of the continuously monitored temperature signal, in a mixed population of critically ill patients, during the course of a suspected intensive care unit–acquired infection. Methods: Continuous temperature recordings were performed in 12 patients with ventilator-associated pneumonia, 10 with bloodstream infections, and 7 with systemic inflammatory response syndrome (SIRS). Ten patients developed septic shock, and 12 were considered having sepsis. Temperature monitoring was performed with a thermistor sensor (Datalogger Spectrum 1000; Veriteq Instruments, Richmond, BC, Canada), with a sampling frequency of 1 sample per 10 seconds. Continuous wavelet transformation was applied to the temperature signal, aiming at decomposing it to different frequency components (0.008-0.02 and 0.02-0.05 Hz, indicating metabolic and neurogenic inputs upon local temperature control, respectively) during different time scales, using software available from MATLAB. Different features indicating variability of the signal, such as wavelet coefficients, wavelet, and sample entropy measures per frequency band, were computed and correlated with daily measured specific organ failure assessment score of severity of illness. Multiple analyses of variance and cluster analysis using Mahalanobis distance evaluated differences between groups of patients (SIRS, sepsis, and septic shock). Multiple regression analyses assessed whether different features from continuous wavelet transformation were independent predictors of specific organ failure assessment score and development of septic shock. Results: Average sample entropy of the whole signal was significantly increased in patients with SIRS vs sepsis and septic shock (3.7 ± 0.2 vs 2.8 ± 0.12 vs 2.2 ± 0.18, P b .001, respectively, for all comparisons). Entropy/metabolic band was significantly increased in the SIRS group (0.07 ± 0.002) related to septic shock group (0.02 ± 0.0016, P b .001). The percentage of cross-correlation greater than 0.95 between 30-second distant time slots was 0.85 vs 0.05 in patients with septic shock vs SIRS (P b .001), indicating more steady and periodic behavior of temperature during severe infection. Entropy/metabolic band and crosscorrelation between adjacent timeslots proved to be independent predictors of severity of illness (β slope = −1.087, P = .022 and .88 and P = .013, respectively). Conclusions: Reduced variability of temperature fluctuations seems to correlate with severity of illness, reflecting different thermoregulatory mechanisms, and can be of prognostic value in patients with systemic inflammation.