Estimating intracranial pressure via low-dimensional models: toward a practical tool for clinical decision support at multi-hour timescales

Abstract

AbstractBroad clinical application of non-invasive intracranial pressure (ICP) monitoring using computational models requires a method of modeling ICP on the basis of easily measured patient data such as radial or brachial arterial blood pressure (ABP). These models may be highly complex, rendering them too slow for clinical and operational use, or may rely on data that is not consistently available. Coupling these models to an upstream vasculature component model decreases data requirements. For the purposes of clinical decision support at multi-hour timescales, two natural choices for model development are to increase intracranial model complexity or to include feedback mechanisms between ICP and vascular model components. We compare the performance of these two approaches by evaluating model estimates against observed ICP in the case of a slow hypertensive event from a publically available dataset. The simpler model with bi-directional feedback requires minimal identifiability and is sufficiently accurate over these timescales, while a more complex is difficult and expensive to identify well enough to be accurate. Furthermore, the bi-directional simple model operates orders of magnitude faster than the more anatomically accurate model when driven by high-resolution ABP. It may also be configured to use lower resolution ABP summary data that is consistently clinically available. The simpler models are fast enough to support future developments such as patient-specific parametrization and assimilation of other clinical data streams which are illustrated during the case of a complex ICP regime for a different patient. We present model comparisons to highlight the advantages of the incorporated simple model and its possible predictive power with further optimization.