Progress in Reverse-engineering the Neurological Pupil Index

Introduction: The pupillary light reflex (PLR) is used to determine neurologic status. The PLR is often obtained quantitatively in the neurocritical care setting to monitor for worsening in neurological status. We studied the ability of the PLR parameters to predict the Neurological Pupil Index (NPi), which is a quantitative predictor of neurological status.

Methods: PLR recordings were taken with a digital infrared pupillometer that reports both PLR parameters (latency, maximum diameter, minimum diameter, percent change in diameter, mean constriction velocity, maximum constriction velocity, mean dilation velocity) and the NPi score. These data were reviewed retrospectively from a cohort that includes healthy persons and persons hospitalized for traumatic brain injury, aneurysmal subarachnoid hemorrhage, and ischemic stroke. A multiple linear regression analysis was performed using k-fold (k=10) cross-validation on all possible combinations of the seven PLR parameters reported by the device with the NPi score as the dependent variable.

Results: A total of 1734 measurements of the NPi along with the seven associated parameters of the PLR were collected. Average values across the cohort: NPi=4.2, latency=0.24s, maximum diameter=3.69mm, minimum diameter=2.59mm, percent change=28.3, mean constriction velocity=1.87mm/s, maximum constriction velocity=2.99mm/s, mean dilation velocity=0.79mm/s. Parameter combinations that best predicted the NPi were: latency/minimum/maximum/mean dilation velocity, all except percent change, and all except maximum constriction velocity (all with the same results, R2: 0.67, mean absolute error: 0.24, mean squared error: 0.14).

Conclusion : These results suggest that the prediction of NPi is nonlinear with respect to associated PLR parameters. Further investigation with more sophisticated modeling is necessary.