Evaluating the Reliability of Neurological Pupillary Index as a Prognostic Measurement of Neurological Function in Critical Care Patients

Background Neurological pupil index (NPi) is a novel method of assessing pupillary size and reactivity using pupillometry to reduce human subjectivity. This paper aims to evaluate the use of NPi as a potential prognostic tool in a broad population of neurocritical care patients by observing the correlation between NPi, modified Rankin Scale (mRS), and Glasgow Coma Scale (GCS). Methods Our data was collected from 194 patients in the neurosurgical intensive care unit (ICU) at Arrowhead Regional Medical Center (ARMC), as determined by the power calculation. We utilized the Kolmogorov-Smirnova and Shapiro-Wilk normality tests with Lilliefors significance correction. Pearson product-moment correlation was performed between average final NPi and final GCS. Multi-variate linear regression and analysis of variance (ANOVA) were used to evaluate the association and predictive capabilities of NPi on GCS and discharge mRS. Finally, we evaluated whether age, ethnicity, sex, length of stay (LOS), or discharge location were significantly associated with NPi. Results We observed a significant correlation between final GCS and NPi (r=0.609, p<0.001). Our regression analysis revealed that NPi significantly predicted GCS and mRS scores; however, no associations were found between age, ethnicity, sex, LOS, or discharge location. Limitations of our study include a single institutional study with a lack of disease subtyping and the inability to quantify the predictive ability of NPi. Conclusion The analysis revealed a strong correlation between final GCS and average final NPi. NPi was also able to significantly predict GCS and mRS scores. The correlation between NPi and established methods to determine neurological function, such as mRS and GCS, suggests that NPi can be a good prognostication tool for neurological diseases.


Introduction
Pupillary size and reactivity are among the major non-invasive methods of assessing neurological function. During the pupillary reflex exam, the pupil's size and symmetry are measured, and the rate of reactivity is classified as either brisk, sluggish, or non-reactive. Abnormal measurements can indicate diseases such as stroke, tumors, and traumatic brain injuries [1][2][3]. It is a valuable prognostic tool for assessing the patient's neurological health [4]. Current methods of assessing pupillary reflex include the Glasgow Coma Scale (GCS) and modified Rankin Scale (mRS). To date, many variations of these clinical tools have been developed to further refine the prognosticative ability of patient outcomes. However, these measurements are predisposed to inaccuracy due to subjectivity, inexperience, language barriers, iatrogenic barriers (e.g., intubation, sedation), and lack of standardization of the examiner [5][6][7][8][9]. Neurological pupil index (NPi) is a new practice established by NeuroOptics, Inc. (Irvine, USA) that measures the size, latency, and velocity parameters and quantifies them on a scale from zero to five, zero being non-reactive, and a score equal to or above three indicating normal pupil behavior [6]. NPi utilizes automated pupillometry to decrease human subjectivity and minimize administration time to increase the efficiency and accuracy of neurological assessments. Evaluating the change in NPi could potentially serve as a more robust prognostication tool to assess the recovery of the brain in neurocritical care.
With the development of the international curing coma campaign (COME TOGETHER), we have seen increasing interest to improve the assessment of patients with impaired neurological function [7][8][9][10]. Previous studies have investigated the utility of these prognostic measurements in combination and even integrated in prognostic modeling calculators, such as the international mission for prognosis and clinical trials in traumatic brain injury (IMPACT) and corticoid randomization after significant head injury (CRASH) [8,[11][12][13]. These studies found that among a variety of these clinical measurements, GCS, NPi, and mRS were all significant predictors of patient outcome in the context of traumatic brain injury (TBI) [14][15][16][17]. Other studies have corroborated these findings, but many of these analyses hone into specific clinical contexts, such as TBI and stroke [18][19][20]. More studies are needed to compare prognostic capabilities across a broad range of neurocritical diseases to increase the generalizability and feasibility of automated pupillometry.
If we can establish a correlation between NPi, mRS, and GCS, it could support the reliability of NPi as an alternative and potentially more robust prognostic tool in critical care patients. In this study, we used a pupillometer to measure the NPi in 194 subjects in the neurosurgical intensive care unit (ICU) at Arrowhead Regional Medical Center (ARMC) in San Bernardino, California. We hypothesized that NPi and GCS would be positively correlated, mRS and NPi would be negatively correlated, and NPi could significantly predict GCS and mRS scores. Therefore, NPi can be used similarly to or in conjunction with GCS as a tool to assess neurologic function across a varied neurocritical patient population.

Materials And Methods
We collected data from patients in the neuro-ICU at ARMC. The following demographic and clinical information were obtained from medical records to describe patient baseline characteristics: age, sex, ethnicity, length of stay (LOS), and discharge disposition/location. Clinical neurological assessments (i.e., GCS, mRS) were obtained using conventional established methods upon admission and at the time of discharge. NPi measurements were obtained at admission, every four hours (every hour for critically unstable patients), and at ICU discharge using the NeuroOptics Pupillometer (Irvine, CA, USA) version 2.00. NPi values greater than 3.0 were characterized as normal, and NPi values less than 3.0 was considered abnormal.
Initially, we conducted a pre-study power calculation using G*Power (version 3.1.9.7; Heinrich Heine University Düsseldorf, Germany) to compute the statistically significant sample size needed for data collection. Our a priori analyses suggested a sample size of 191 patients to achieve a power of 0.80. Next, we evaluated the homogeneity of our data distribution by using the Kolmogorov-Smirnova and Shapiro-Wilk normality tests with Lilliefors significance correction. We then performed Pearson product-moment correlation analyses to identify any statistically significant correlations between our continuous variables average final NPi, Final GCS, and LOS. The obtained Pearson correlation coefficients were categorized as weak (0.00-0.30), moderate (0.31-0.60), and strong (>0.60). The association of NPi, GCS, and mRS and the ability of NPi to predict GCS and mRS, was determined using multiple and ordinal regression analysis. Finally, we determined whether several predictor variables, such as age, ethnicity, sex, LOS, and discharge location, could significantly predict a patient's NPi score using multivariate regression and ANOVA. All statistical analyses were performed using SPSS statistics software V28.0.1.0 (IBM Inc., Armonk, USA).
We conducted this study in compliance with the principles of the Declaration of Helsinki. The study's protocol was reviewed and approved by the Institutional Review Board of Arrowhead Regional Medical Center (#22-21). Informed consent was waived.
Participants of the study were those that were admitted to the neuro ICU at ARMC. Their sex was determined through medical records. This study did not involve an exclusive population. Ethnicity was self-determined by patients upon initial admission to the hospital and was included in this study to determine any patterns between NPi and prognosis in certain groups.

Variable
Value (

Discussion
As we hypothesized, our results identified a strong correlation between average final NPI and final GCS scores. The correlation between these two variables suggests that NPi measurements could be used similarly to GCS as an effective predictor of prognosis. Our subsequent regression model reveals that our utilization of these three measurements is significantly correlated within varying disease contexts present in our patient population. Since GCS and discharge mRS have been shown to effectively predict patient prognosis, NPi's ability to significantly predict these two measurements supports the possibility of using NPi as an alternative predictor of prognosis [21][22]. However, further multi-institutional studies are needed to evaluate the potential superiority of NPi as a predictor of the prognosis within a variety of neurological disease contexts.
Interestingly, although NPi, GCS, mRS displayed significant relationships, we found that only NPi was able to significantly predict discharge location. Although the majority of patients were discharged home, being able to identify relationships between a patient's "measured" neurological health and inevitable discharge location could help improve hospital resource utilization and care management. This highlights the need to explore whether these trends can be seen at other institutions and if better categorizations are needed to better identify significant relationships and predictive capabilities within these variables.
Our study also looked at whether demographic parameters such as age, sex, and ethnicity played a role in predicting our patients' prognoses, as measured by NPi. Overall, we failed to identify any significant associations between these variables, which could be attributed to our predominantly Hispanic patient population. Finding strong correlations between certain demographic groups and the prognostic capabilities of NPi, GCS, and mRS could help make better-informed decisions. However, such strong demographic patterns could also question the external validity of these measurement tools in a varied patient population. These results convey that NPi and GCS can be similarly used in predicting prognosis among a potentially diverse patient demographic, although more studies are needed to confirm these findings.
Our study does have some limitations. First, although our sample size displayed appropriate statistical power, a majority of our patients are Hispanic and over the age of 50. Further analysis is needed within a multi-institutional study with more patients to ensure reproducibility and generalizability to a diverse patient population. Second, since our study aimed to evaluate the reliability of NPi in all neurocritical care patients, we did not stratify our patients based on the clinical context. Hence, we are unable to take into account specific neurological diseases and the potential confounding effects on predicting patient prognosis. Finally, our correlation analyses cannot determine any causations or quantify how well these tools predict long-term patient outcomes and should be interpreted with these points in mind.
Overall, our study offers additional insight into how NPi, which could serve as a more objective and efficient method of evaluating patient prognosis, relates to conventional methodologies like GCS and mRS. Despite their widely accepted use, these tools are inherently slightly more subjective and more prone to potential differences in inter-observer reliability. Our future studies will aim to quantify how well NPi can predict long-term patient outcomes within specific disease subtypes. We plan to replicate this analysis as part of a multi-institutional study and corroborate NPi as a potentially superior prognostic tool for neurocritical care patients.