Exploring the temporal statistical relationship between commercial cerebral near-infrared spectroscopy signals and intracranial pressure in traumatic brain injury

Abstract Background Monitoring of intracranial pressure (ICP) is important for not only assessing neurological status of patients following traumatic brain injury (TBI), but also for guiding therapeutic interventions for the prevention of secondary neurological injury. Although current ICP monitoring techniques are highly temporally accurate, they come with some significant limitations. These limitations include the invasive nature, expertise required for placement of these devices, high cost associated with maintenance, and poor spatial resolution. Cerebral near-infrared spectroscopy (NIRS) offers a non-invasive alternative technique for cerebral physiological monitoring. This modality eliminates the need for surgical expertise, as well as decreasing the financial cost associated with monitoring cerebral physiology. Preliminary evidence has pointed to a potential link between the NIRS variable, regional oxygen saturation (rSO 2 ), and invasively obtained ICP. Despite this, there are very few studies that exist that investigate the direct statistical relationship between NIRS variables, such as rSO 2 , and invasively obtained ICP. As such, this study set out to characterize and compare the temporal statistical properties of commercial NIRS-based rSO 2 , ICP, and mean arterial pressure (MAP) signals. Results Utilizing retrospective high-frequency physiologic data from a population of 118 TBI patients across different temporal resolutions and data window sizes, we evaluated independent statistical signal structure of rSO 2 and ICP using variance analysis and ARIMA modelling. Following normalization, variance metrics remained significantly different between all physiologic variable comparison groups, across all temporal resolutions and window sizes examined. In contrast, median optimal ARIMA model orders converged, with no significant differences observed between any of the included physiological variables. Next, Pearson correlation, cross-correlation, Granger causality, and VARIMA IRF analysis continued to show increasing temporal similarity between ICP and rSO 2 as temporal resolution decreased and window size increased. In contrast, throughout the majority of the analyses, the relationship between AMP and rSO 2 remained not only consistently weaker, but also more inconsistent across the majority of the analyses. Conclusion Findings from this study demonstrate convergence in the temporal statistical behaviour of ICP and rSO 2 at lower temporal resolutions, providing a foundation for future predictive non-invasive ICP monitoring approaches. Further investigation into direct statistical relationships and development of reliable non-invasive ICP prediction models using NIRS-derived data remains necessary for clinical utility.