A novel algorithm for the continuous determination of individualized intracranial pressure (iICP) thresholds using a multi-window weighted approach.

backgroundIndividualized intracranial pressure (iICP) thresholds, defined as the function intersectionality between intracranial pressure (ICP) and cerebrovascular reactivity (CVR), represent a promising potential approach to personalized medicine in neurocritical care. However, current iICP derivation methods fail to account for the variable quality of iICP estimates and rely on entire recording periods, limiting their use to retrospective, post-hoc calculations and thereby limiting clinical applicability. Therefore, the goal of this study was to develop an automated, continuous iICP derivation algorithm that provides accompanying quality metrics, evaluate the algorithm’s performance, and identify patient-related factors influencing derivation yields.

methodsA custom algorithm employing a multi-window weighted approach was developed for the continuously updating derivation of iICP. The algorithm was designed to concurrently generate quality metrics alongside iICP outputs, which grade key characteristics of the underlying ICP-CVR curves. The algorithm was tested on a cohort of 131 moderate-to-severe traumatic brain injury (TBI) patients from the Winnipeg Acute TBI Database. Multiple iterations of the algorithm, with varying parameter settings, were tested to assess algorithm performance. Subgroup analyses were performed to identify demographic-, admission-, and treatment-related factors that may influence derivation yields.

resultsAlgorithm performance varied significantly with parameter selection, including modeling method, CVR index-threshold combination, and update frequency. The best performing iteration achieved a median derivation yield of 66.1%. The use of modeling methods in conjunction, the use of the pressure reactivity index and thresholds of + 0.20 and + 0.25, and the use of higher update frequencies were shown to be associated with superior algorithm performance. Time-based subgroup analyses revealed lower yields during the early phases post-injury, while demographic-, admission-, and treatment-related factors showed minimal influence on algorithm performance.

conclusionThis study demonstrates the feasibility of deriving iICP thresholds in a continuously updating fashion. While promising, further optimization is needed to improve derivation yields and enable bedside implementation. Moreover, comprehensive outcome and physiologic analyses are needed to clarify the relationships of iICP with long-term outcomes and cerebral physiologic insult burden.