Feasibility and Variability of Automated Pupillometry Among Stroke Patients and Healthy Participants: Potential Implications for Clinical Practice.
Serial pupillary assessment is a cornerstone of acute neurological monitoring. In the hyperacute setting, it can sometimes be the only detectable sign of END because sedation, intubation, and medications can compromise full neurological assessment. (4) The two main components of the manual pupillary assessment are pupillary light reactivity (PLR) and pupil size. However, these assessments may be confounded by multiple factors, for example, observer bias, subjectivity of terminology, imprecise measuring tools, and various external confounders such as ambient light, visual acuity, torch luminosity, and iris pigmentation. Moreover, traditional manual methods have poor interrater reliability and thus make clinical interpretation of PLR unreliable. (5-7)
Automation of the pupillary assessment using an automated pupillometer (NPi-200; NeurOptics,) offers a potential solution to this important clinical problem. The main benefit of automated pupillometry is to produce an objective and standardized assessment of PLR and pupil size. The NPi-200 also provides a Neurological Pupil index (NPi), which produces a scalar value of PLR from 0.0 to 5.0, with a score less than 3.0 being abnormal. This eliminates the subjectivity of traditional PLR assessment and may make trends more clinically meaningful. Because automated PLR measures are now being used in a number of settings where neurological assessment is key to detecting END, the aim of this study was to explore the acceptability, feasibility, and safety of automated pupillometry after acute stroke. In addition, the interrater reliability of automated pupillometry was compared with that of the traditional manual method.
The study was divided into 2 phases. In the first phase (assessment of feasibility, acceptability, and safety), automated and manual pupillary assessments were recorded, alongside routine neurological observations, in stroke patients considered at a high risk of END within the first 72 hours of admission to the HASU at Sheffield Teaching Hospitals, United Kingdom. Eligible patients included those with any of the following risk factors for END: National Institute of Health Stroke Scale (NIHSS) score of > 5, large vessel occlusion, intracerebral hemorrhage, diabetes, atrial fibrillation, and hemorrhagic transformation of infarction or cerebral edema on initial computed tomography. (2,8,9)
The automated pupillometer would be deemed feasible if at least 80% of readings were completed as intended. The device would be acceptable if all participants and nursing staff rated the assessment positively (1, 2, or 3 out of 5) for comfort and ease of use, respectively, this was evaluated using a Likert scale as follows: 1, very comfortable/easy; 2, comfortable/ easy; 3, borderline; 4, some discomfort/difficulty; and 5, uncomfortable/very difficult. Safety was defined as the absence of any device-related serious adverse event.
In the second phase (assessment of variability), recordings of paired automated and manual assessments were made to analyze variability and interrater agreement for pupil size, detection of anisocoria, and PLR between 2 independent observers. Agreement between observers for pupil size was defined as a difference of < 1 mm between each observer's measurements of the same pupil. Agreement for PLR was defined as both observers recording PLR in the same manual or NPi category as follows: "nonreactive" (NPi, 0.0), "sluggish" (NPi, 0.1-2.9), and "brisk" (NPi, 3.0-5.0). Neurological Pupil index scores are not directly equivalent to speed of pupillary responses. Manual PLR assessment was compared with NPi, in addition to constriction velocity (CV), because NPi scores incorporate multiple additional variables (such as maximum and minimum pupil sizes, percent change in size, dilatation velocity, and latency of constriction) that may better represent the complexity of the pupillary response. (10) In this study, anisocoria was defined as a difference of 1 mm or greater between pupils; this is consistent with other studies and enabled a direct comparison between automated assessment and the whole numbers measured manually. (11) The paired assessments were made within 15 minutes of each other with observers blind to one another's measurements. Paired measurements were also recorded from healthy volunteers (healthcare staff) to improve the sample size.
Tukey boxplots were used to demonstrate the range of measures for PLR and pupil size for manual and automated methods. Percentages of interrater agreement for all 3 pupillary measures (pupil size, anisocoria, and PLR) were calculated between observers. Cohen's k coefficient values were calculated to ascertain interrater agreement of anisocoria and PLR and were interpreted as follows: poor (=0.00), slight (0.00-0.20), fair (0.21-0.40), moderate (0.41-0.60), substantial (0.61-0.80), and near perfect (0.81-1.00). Spearman correlation coefficient was used for pupil size because of the continuous nature of automated values for pupil size.
Twelve stroke patients were recruited during phase 1 of the study (mean age, 67.8 years), of whom two-thirds were male. Seventy-five percent of the participants had experienced an ischemic stroke with a median (interquartile range) NIHSS of 3 (5.75). Diabetes (41.7% of participants) was the most prevalent END risk factor, followed by NIHSS of > 5 (33.3%), atrial fibrillation (25.0%), cerebral edema (25.0%), and intracerebral hemorrhage (16.7%). No participants had large vessel occlusion or hemorrhagic transformation on initial computed tomography.
Only 68.4% of intended measurements were completed during the first 72 hours. However, subdividing this result into day and night shifts reveals that 92.7% of daytime measurements (9 AM to 7 PM) were completed as intended compared with 30.8% during the night shift (9 PM to 7 AM). The device was acceptable to both patients (average Likert, 1.4/5) and healthcare staff (average Likert, 2.4/5), and no device-related serious adverse event occurred.
For analysis of variability and interrater agreement, 132 paired measurements of individual pupils were recorded from a total of 52 participants (42 paired readings from stroke patients and 90 from healthy volunteers). As detection of anisocoria requires comparing both pupils, there were therefore 66 paired measurements for the detection of anisocoria.
For automated assessment of pupil size, the interrater agreement was 99.2%, whereas the Spearman correlation coefficient was 0.949 (95% confidence interval [CI], 0.929-0.969). Using manual methods, the interrater agreement of pupil size was 61.4% and the Spearman correlation coefficient was 0.633 (95% CI, 0.531-0.735). Figure 1 demonstrates the spread of automated measurements per manually measured pupil size, thus giving an indication of agreement between automated and manual methods for assessing pupil size. The vast majority of manual measures were recorded as 3 or 4 mm (84.1%); however, the corresponding automated values varied considerably from 1.9 to 6.1 mm.
Using the pupillometer, interrater agreement for the detection of anisocoria was substantial (98.5%, [kappa] = 0.660; 95% CI, 0.039-1.00). For manual assessment, interrater agreement was fair (89.4%, [kappa] = 0.306; 95% CI, -0.078 to 0.690). Interrater agreement among unequal pupils only (n = 14) was considerably less at 66.7% and 36.4% for both automated and manual methods, respectively. Agreement for the detection of anisocoria between automated and manual methods was poor (87.9%, [kappa] = -0.027; 95% CI, -0.074 to 0.020).
Neither manual nor automated assessments recorded an absent PLR in this sample. Interrater agreement of automated assessment of PLR was fair (97.7%, [kappa] = 0.389; 95% CI, -0.160 to 0.938), whereas the respective interrater agreement for manual assessment was poor (92.4%, [kappa] = -0.039; 95% CI, -0.063 to -0.015). Figure 2 demonstrates the lack of agreement between methods for the assessment of abnormal PLR. All manually measured sluggish responses were normal when measured using the automated pupillometer ([greater than or equal to] 3.0), and all abnormal NPi scores were marked as brisk manually ([kappa] = -0.026; 95% CI, -0.042 to -0.010). However, NPi is an indexed function of a number of pupillary measures and does not compare reaction speeds directly. Mean (standard deviation) CV among pupils manually reported as sluggish was 1.60 (1.08) m/s, significantly lower than the CV for those reportedly brisk at 2.51 (0.84) m/s (P = .001). However, raw agreement between observers for which pupils were reportedly sluggish was 0%. Using a definition of abnormal CVas less than 0.8 m/s, (10) automated measurements of CV showed that only 2 of the 10 manually reported sluggish pupils had abnormal CV implying only slight agreement between manual measurement of PLR and automated assessment of CV (20.0%, [kappa] = 0.006; 95% CL -0.004 to 0.016). In contrast, interrater agreement for automated assessment of CV was perfect (100.0%, [kappa] = 1.00; 95% CI, 1.00-1.00).
Automated pupillometry is acceptable and safe, reflecting similar experiences in neurocritical care environments. (12-14) Although feasibility criteria were not achieved, this can largely be attributable to poor handover of study information to night staff. Improving the use of automated pupillometry on the HASU can be easily undertaken through improved communication and training of nursing staff.
Automated pupillometry seemed more reliable than manual pupillary assessment in every domain evaluated. Interrater agreement was higher with automated methods for pupil size, anisocoria, and PLR. By excluding normal findings to analyze only the occurrences of detected anisocoria and abnormal PLR, automated pupillometry still outperformed manual assessment. However, as most phase 2 participants were healthy volunteers, there were relatively few abnormal readings that inhibited more meaningful interpretation. Nonetheless, these findings suggest that automated pupillometry is likely to be more reliable than manual methods of pupillary assessment. This is consistent with previous studies evaluating agreement and reliability between automated and manual pupillary assessments in differing neurological populations. (5,12,15)
There was a substantial discrepancy between assessment methods as shown in Figures 1 and 2. This disparity implies that 1 method may be less accurate than the other. Meeker et al12 (2005) found that automated pupillometry could detect changes to pupillary dynamics much earlier than manual methods in neurosurgical inpatients. Furthermore, the NPi-200 has been shown to correlate closely to changes in concurrent intracranial pressure (ICP) and may even predict rises in ICP before they are directly measured. (13,14) However, numerous factors other than ICP can influence pupillary dynamics. The superior accuracy of the NPi-200 is therefore a major assumption that has not been definitively proven.
Although manual PLR is not directly comparable with NPi, the fact that none of the 10 sluggishly reported pupils had abnormal NPi scores and only 2 had an abnormal CV suggests that manually reporting sluggish pupils may offer little useful clinical information.
This is especially true when considering the use of tracking changes in reactivity from one observer to another as happens during shift handovers of clinicians. Moreover, the NPi-200, as a more precise and comprehensive monitoring tool, is likely to be more sensitive to change, and thus tracking changes may help predict END earlier. In contrast, changes in manual pupillary assessment are likely to be large before clinicians can detect changes to act upon. All of this evidence suggests that manual pupillary assessment could be an unreliable and inaccurate monitoring tool, an imperative notion given its central role within neurological observations.
The findings of this study are directly relevant to clinical practice. Automated pupillometry can be used among more alert and conscious stroke patients outside the intensive care setting. Its use was comfortable, easy to undertake, and devoid of adverse events. Our study suggests that the NPi-200 is a more precise and reliable monitoring tool compared with standard manual assessment. Earlier detection of neurological deterioration may enhance the clinical outcomes for time-dependent interventions (such as decompressive craniotomy) after stroke.
There are disadvantages to automated pupillometry that also need to be considered. The NPi-200 is monocular and therefore unable to detect problems in the consensual pupil such as relative afferent pupillary defects. It was also sometimes difficult to perform the assessment in patients unable to follow instructions. This may be a novel limitation of automated pupillometry among more alert and conscious patients (as opposed to previous research populations involving more sedated patients). Furthermore, the NPi-200 is considerably more expensive than manual assessment. However, if it was able to accurately differentiate neurological from non-neurological causes of deterioration, this may prevent unnecessary and costly neuroimaging. In addition, if automated pupillometry enabled earlier detection of END and referral for lifesaving, time-dependent treatments, it would be hard not to advocate its value.
There were notable limitations of this study. Primarily, the analysis was not statistically powered to prove superiority of one assessment method over the other. Observer bias may also have influenced manual recordings of pupillary assessment. In addition, the relatively high numbers of healthy participants and, consequently, limited amount of abnormal measurements mean readers should interpret results with caution.
The NPi-200 seems to be a more reliable and sensitive evaluation of pupillary responses compared with manual assessment among acute stroke patients and healthy volunteers. The automated pupillometer was being considered for routine clinical use in this HASU. The findings of this study provide underpinning evidence for the integration of automated pupillometry within neurological monitoring after stroke.
Questions or comments about this article may be directed to Matthew Marshall, MSc, at email@example.com. He is a Medical Student, King's College, London, England.
Ritesh Deo, University of Sheffield, Western Bank, Sheffield S10 2TN, England.
Charmaine Childs, PhD MPhil BNurs, Sheffield Hallam University, Collegiate Campus, Sheffield, England.
Ali Ali, MBChB FRCP MSc, Sheffield Teaching Hospitals and University of Sheffield, Sheffield, England.
The authors declare no conflicts of interest.
(1.) Roquer J, Rodriguez-Campello A, Gomis M, et al. Acute stroke unit care and early neurological deterioration in ischemic stroke. J Neurol. 2008;255(7):1012-1017.
(2.) Kwan J, Hand R Early neurological deterioration in acute stroke: clinical characteristics and impact on outcome. Q J Med. 2006;99(9):625-633.
(3.) Weimar C, Mieck T, Buchthal J, et al. Neurologic worsening during the acute phase of ischemic stroke. Arch Neurol. 2005; 62(3):393-397.
(4.) Berlin T. Advanced clinical concepts in pupil assessment. 2016.
(5.) Olson DM, Stutzman S, Saju C, Wilson M, Zhao W, Aiyagari V. Interrater reliability of pupillary assessments. Neurocrit Care. 2016;24(2):251-257.
(6.) Couret D, Boumaza D, Grisotto C, et al. Reliability of standard pupillometry practice in neurocritical care: an observational, double-blinded study. Crit Care. 2016;20:99.
(7.) Omburo L, Stutzman S, Supnet C, Choate M, Olson DM. High variance in pupillary examination findings among postanesthesia care unit nurses. J Perianesth Nurs. 2017;32(3): 219-224.
(8.) Thanvi B, Treadwell S, Robinson T. Early neurological deterioration in acute ischaemic stroke: predictors, mechanisms and management. Postgrad Med J. 2008;84(994): 412-417.
(9.) DeGraba TJ, Hallenbeck JM, Pettigrew KD, Dutka AJ, Kelly BJ. Progression in acute stroke: value of the initial NIH stroke scale score on patient stratification in future trials. Stroke. 1999;30(6): 1208-1212.
(10.) Shoyombo I, Aiyagari V, Stutzman SE, et al. Understanding the relationship between the Neurologic Pupil index and constriction velocity values. Sci Rep. 2018;8(1):6992.
(11.) Olson DM, Fishel M. The use of automated pupillometry in critical care. Crit Care Nurs Clin North Am. 2016;28(1):101-107.
(12.) Meeker M, Du R, Bacchetti P, et al. Pupil examination: validity and clinical utility of an automated pupillometer. J Neurosci Nurs. 2005;37(1):34-40.
(13.) McNettM, Moran C, Janki C, Gianakis A. Correlations between hourly pupillometer readings and intracranial pressure values. J Neurosci Nurs. 2017;49(4):229-234.
(14.) Chen JW, Gombart ZJ, Rogers S, Gardiner SK, Cecil S, Bullock RM. Pupillary reactivity as an early indicator of increased intracranial pressure: the introduction of the neurological pupil index. Surg Neurol Int. 2011 ;2:82.
(15.) Zhao W, Stutzman S, DaiWai O, Saju C, Wilson M, Aiyagari V. Inter-device reliability of the NPi-100 pupillometer. J Clin Neurosci. 2016;33:79-82.
Caption: FIGURE 1 Boxplots of Automatically Measured Pupil Sizes (Using the NPi-200) According to Each Manually Measured Pupil Size
Caption: FIGURE 2 Boxplots of NPi Scores According to Manually Measured Sluggish or Brisk PLR Responses (The Dotted Line at 3.0 Indicates Minimum Nornal NPi Score)
|Printer friendly Cite/link Email Feedback|
|Author:||Marshall, Matthew; Deo, Ritesh; Childs, Charmaine; Ali, Ali|
|Publication:||Journal of Neuroscience Nursing|
|Date:||Apr 1, 2019|
|Previous Article:||Triage of Mild Head-Injured Intoxicated Patients Could Be Aided by Use of an Electroencephalogram-based Biomarker.|
|Next Article:||Direct-Acting Oral Anticoagulants and Warfarin-Associated Intracerebral Hemorrhage Protocol Reduces Timing of Door to Correction Interventions.|