Estimation of subject specific ICP dynamic models using prospective clinical data

Abstract

We used a prospective clinical trial to generate physiologic data to create subject-specific in silico (computer simulation) models of intracranial pressure dynamics in children with severe traumatic brain injury. The trial included a physiologic challenge protocol with changes in head-of-bed elevation and minute ventilation, applied over multiple iterations to three subjects. Physiologic signals (electrocardiogram, respiration, arterial blood pressure, intracranial pressure [ICP], oxygen saturation) were recorded continuously, along with clinical annotations indicating the precise timing of physiologic challenges. Several parameters within the model of ICP dynamics were estimated for each subject based on the ICP response to the challenges. Estimation was done using a standard optimization algorithm to minimize the difference between the ICP trajectory predicted by the model and the actual ICP data. The ICP trajectory predicted by the model was similar to the actual ICP data in all cases, and the mean absolute error varied between 0.5-2.8 mmHg (mean = 1.4mmHg). These results demonstrate the potential for using clinically annotated prospective data to create subject-specific computer simulation models. Future research will focus on improvements in the logic for cerebral autoregulatory mechanisms and physiologic adaptation.

Assessing the prediction potential of an in silico computer model of intracranial pressure dynamics*

“…Consistent with prior investigators (15,17), autoregulation is modeled as a feedback loop . We have previously shown that the behavior of our model was qualitatively correct (11,12), and we were able to manually calibrate the model so that it replicated retrospective clinical data from two prior subject-specific cases (13). Preliminary parameter estimation results were reported in (12).…”

“…Some parameter estimates derived from the laboratory data and/or clinical data have been reported, but much of the necessary data remains difficult to obtain. Models have been calibrated to fit data recorded for specific patients (7,8,(11)(12)(13)(17)(18)(19)(20)(21) and excellent results have been reported in some cases (8,21). However, to our knowledge actual prediction of patient-specific ICP response to therapies or other interventions has not been reported.…”

“…The model used in the present research is similar in many ways to previously reported dynamic ICP models by our research group in that it considers the cranial vault to contain the brain parenchyma plus several fluid "compartments" that together are constrained not to exceed the total cranial volume (11)(12)(13). The fluid compartments include the arterial blood volume, capillary blood volume, venous blood volume, CSF volume, the brain volume, and "other" volume that may or may not be present in a particular patient (e.g.…”

“…The model of ICP dynamics used in this study was patterned after models previously reported by our research group (11)(12)(13). Since the underlying physiology is very complex and not completely understood, the mathematical model is necessarily highly simplified.…”