Quantifying methodological uncertainties and navigating feasibilities in monitoring of cerebral autoregulation - Towards individualised management of TBI.

Abstract

Traumatic brain injury (TBI) remains a critical medical challenge due to its heterogeneous nature and complex pathophysiological mechanisms. Individualised management of TBI patients, taking advantage of such technologies like cerebral autoregulation (CA) monitoring, has the potential to improve outcomes by tailoring interventions to the unique physiological responses of each patient. However, a significant barrier to the widespread clinical use of CA-based metrics is the current lack of reliability and standardisation in their assessment and interpretation. This thesis addresses several core aspects of this problem. Part I presents the extensive groundwork necessary for state-of-the-art data analysis of high-resolution physiological measurements. Conducting reliable, reproducible, high-quality, and ethical research comes with specific requirements. To meet these, I focused on several key aspects: obtaining ethical approval for the Brain Physics Database, creating pipelines for the curation of large retrospective datasets, establishing data flow processes for prospective data collection and curation, and fine-tuning the IT infrastructure, including workstations and analytical tools based on the ICM+ framework. Although this might be viewed as foundational work, it constituted a substantial part of my PhD and was crucial not only for my own methodological and analytical research but also for providing a framework utilised by a wider group of researchers. Therefore, this work warrants a dedicated chapter in my thesis. Part II focuses on developing and evaluating methodological approaches to quantify uncertainty in CA metrics. Various techniques were investigated, including simple summary statistics, parametric bootstrap, stationary block bootstrap, and Bayesian inference. Among these, the parametric bootstrap for Phase Shift (PS) metric of CA proved the most promising, offering robust estimates of uncertainty. Additionally, deep learning methods were explored to mark and reject periods of data where the pressure reactivity (PRx) metric calculations were invalid. The reliability of automated PRx-derived optimal cerebral perfusion pressure (CPPopt) estimates was also addressed, with a special focus on prospective bedside application. Finally, Part III examines the current clinical relevance of CA-derived metrics. This section confirmed the clinical impact of the PRx-derived lower limit of reactivity (LLR), showing its association with patient outcomes. Further, the characteristics of the PRx and cerebral perfusion pressure (CPP) relationship were evaluated to determine the effect of targeting CPP at CPPopt. Data from a prospective trial confirmed that lung-protective ventilation strategies do not compromise PRx, making them compatible with autoregulation-guided management in patients without lung injury. Lastly, similar autoregulation metrics were examined in neurosurgical settings, demonstrating their broader potential beyond TBI. This thesis represents a significant contribution to both the methodological and clinical understanding of monitoring of cerebral autoregulation and derived pressure targets, particularly in traumatic brain injury.