Personalizing the Pressure Reactivity Index for Neurocritical Care Decision Support

Abstract

AbstractNeurocritical care patients may benefit from personalized treatment based on their cerebral autoregulatory function. The pressure reactivity index is an important, prevalent metric used to estimate the state of a patient’s cerebral autoregulation and guide clinical decision-making. However, the pressure reactivity index is a noisy metric, highly sensitive to hyperparameter choices, and is associated with uncertainty that may be influenced by patient demographic. In this manuscript, we develop a methodology that increases robustness and reduces the uncertainty of the pressure reactivity index. Using data from traumatic brain injury patients and novel simulated data, we first show that error in the pressure reactivity index is sensitive to the averaging and correlation window sizes used in its calculation. We then show that this error is patient and heartrate specific. The heartrate-sensitive error has vital implications for extending the use of the pressure reactivity index to patients with different regular heart rates, such as pediatric populations. To address this sensitivity, we develop a new algorithm that personalizes the pressure reactivity index calculation by adjusting for patient heart rate at resolutions of single heartbeats. Implementing our heartbeat-based pressure reactivity index methodology decreases error, uncertainty, and sensitivity, allowing the pressure reactivity index to be more clinically useful and robust to variability across patient populations. We also leverage our data and analysis to identify optimal averaging windows of 9-10 seconds and correlation windows from 40-55 samples in the standard method.