Pain pupillary index to prognosticate outcome in comatose cardiac arrest patients

Background: The Neurologic Pupil Index (NPi), derived from automated pupillometry, can accurately predict poor neurological outcome in post-anoxic brain injury. The prognostic role of another index, called Pupillary Pain Index (PPI), remains unknown in this setting. Methods: Monocentric study in adult comatose cardiac arrest (CA) patients. Quantitative PPI and NPi were concomitantly recorded on day 1 and day 2 after CA. The primary study outcome was to assess the prognostic value of PPI to predict 3-month unfavourable outcome (UO, dened as Cerebral Performance Category of 3–5). Secondary outcome was the agreement between PPI and NPi to predict unfavourable outcome. Results: A total of 102 patients were included; 69 patients (68%) patients had UO. Patients with UO showed a lower NPi (4.2 [3.5-4.5] vs. 4.6 [4.3-4.7]; p<0.01 on day 1 - 4.3 [3.8-4.7] vs 4.6 [4.3-4.8] on day 2), and PPI (3 [1-6] vs. 6 [3-7]; p<0.01 on day 1 - 3 [1-6] vs 6 [4-8]; p<0.01 on day 2) than others. PPI and NPi had similar predictive accuracy on day 1 and day 2 for UO. A PPI=1 on day 2 showed a sensitivity of 26 [95% CI 16-38]%, a specicity of 100 [95% CI 89-100]%, a NPV of 39 [95% CI 36-43]% and a PPV of 100% [95% CI 100-100]% to predict UO. On day 2, a total of 6 patients had concomitant PPI=1 and NPi ≤ 2, while 12 showed NPi>2 and PPI=1; the coecient of agreement was 0.42. Moreover, NPi and PPI values showed a moderate correlation both on day 1 and day 2. Conclusions: In this study, PPI=1 day 2 could accurately predict UO in comatose CA patients, with 100% specicity and a higher sensitivity than NPi. The agreement between PPI and NPi values was moderate.


Introduction
Accurate neuro-prognostication in unconscious patients following cardiac arrest (CA) is mandatory to avoid inappropriate withdrawal of life support therapies (WLST) or to provide futile treatment in patients with irreversible brain damage, as well as to effectively communicate with patients' relatives [1]. As such, recent international guidelines updated the prognostic algorithm for these patients and recommended the concomitant presence of at least two speci c patterns (i.e. highly malignant patterns at electroencephalography [EEG]; neuron-speci c enolase [NSE] >60 mcg/L; absence of cortical response at short-latency somatosensory evoked potentials [SSEPs] at day 3; onset of status myoclonus within 72h; extensive brain injury at brain imaging; absence of pupillary and/or corneal re ex at > 72h) to predict unfavorable neurological outcome (UO) [2].
In particular, for pupillary reactivity, the most commonly method used at bedside remains manual pupillary light re ex (PLR) performed by clinicians and/or nurses. Nevertheless, this approach is hampered by a reduced inter-observer reproducibility and incorrect identi cation of unreactive pupils, especially among patients with small pupil size [3]. Some studies have therefore assessed the role of automated pupillometry to increase the accuracy of altered PLR to prognosticate UO in cardiac arrest patients [4][5][6]. A prospective multicentric study also demonstrated that the best parameter derived from automated pupillometry was the Neurological Pupil Index (NPi); when this quantitative score was ≤ 2 at either day 1, 2 or 3 after CA, 100% of patients presented UO, and its prognostic accuracy was higher than manual pupillary examination or quantitative pupillary constriction [7]. Interestingly, no study has assessed the prognostic role of another index derived by the automated pupillometry, the Pupillary Pain Index (PPI), a score derived from pupillary dilation to nociceptive stimulation, which is currently used to monitor the adequacy of analgesia during surgical procedures [8][9][10].
The aim of this study was therefore to assess the prognostic value of PPI to predict UO in unconscious CA patients and to compare PPI with NPi in this setting.

Study design
This pilot retrospective monocentric study was conducted in the Department of Intensive Care at Erasme Hospital, Brussels (Belgium), between 1 st October 2019 and 31 th August 2021. Eligible patients were adult (≥ 18 years old) patients being unconscious (i.e. Glasgow Coma Scale < 9) at hospital admission after resuscitated CA and who survived for at least 24 hours. Patients with pupillary disease (i.e. Adie's pupil, Argyll Robertson pupil), multiple sclerosis, severe periorbital edema and recent ocular surgery were excluded from the study.
The study protocol was approved by the local Ethical Committee (P2019/650), which waived the need for an informed consent because both indices were routinely used as standard monitoring in unconscious brain injured patients. The design and methodology of the study is in accordance with STARD guidelines regarding the diagnostic accuracy of the reported data [11] and the Standards for Studies of Neurological Prognostication in Comatose Survivors of Cardiac Arrest edited by American Heart Association [12].

Post-resuscitation care
In all unconscious patients with post-anoxic brain injury, a standardized therapeutic protocol was used.
Regardless of CA location and initial rhythm, patients were treated with targeted temperature management (TTM) for 24 h with a target temperature of 33°C. Cooling was started immediately after ICU admission with a combination of cold uid bolus (15-20 ml/kg of a crystalloid solution in 30 min) and a cooling pads or intravascular catheter with a temperature-feedback system. Analgesia and sedation were provided using propofol and sufentanil, while cisatracurium was administered to control shivering in the induction phase and, if needed, as a continuous infusion thereafter. More information on post-resuscitation care have already been previously reported [13].
Automated pupillometry assessment Automated pupillometry was performed using two different automated pupillometer. The rst measurement was performed using the NeurOptics NPi-200 device (Neuroptics, Irvine, CA, USA), which uses an infrared camera that integrates a calibrated light stimulation of xed intensity (1000 Lux) and duration (3.2 s) to compute the NPi, based on an integrated algorithm, for each eye. A NPi score ≤ 2 de nes the threshold that has been identi ed to have 100% speci city to predict UO in CA patients [7]. In our ICU, NPi was assessed every 8 hours.
After the NPi assessment, the Algiscan pupillometer (NeuroLight, ID-MED, Marseille, France) was used; brie y, an electrical stimulation with gradual and stepwise intensity (increasing from 10 mA to a maximum of 60 mA) was applied on the left forearm of the patients through two electrodes connected with a cable to pupillometer. This assessment was performed once daily and only in unconscious patients in our ICU. Baseline pupil size (mm), pupillary dilation to pain (i.e., the difference between poststimulation and baseline pupil size, expressed as a %) and PPI were recorded during the stimulation for each eye. PPI measures pupil dilation in response to the increasing electric stimulus; once the threshold of 13% in pupil dilation is detected, electrical stimulation is automatically interrupted and PPI is calculated, ranging from 1 (pupillary re ex dilation < 5% to the maximal stimulation intensity of 60 mA) to 9 (pupillary re ex dilation > 13% with 10 mA stimulus) [9]. For this study, we collected NPi and PPI that were performed on each eye within 24 hours (day 1) and between 24 and 48 hours from CA (day 2).

Data Collection
Patients' demographics, medical and clinical history as well as main biological data were collected; initial rhythm, categorized as shockable (i.e. ventricular brillation or tachycardia) vs. non-shockable (asystole or pulseless electrical activity) and duration of CA were also collected. The use of drugs that might interfere with pupillary activity (i.e. sedatives or opioids) and their doses were also reported. To assess the severity of brain damage, we also collected the highest NSE value within the 72 hours after CA, the absence of cortical N20 waves at SSEP performed within 72 hours from arrest and the presence of highly malignant patterns (HMPs) at continuous EEG (i.e. presence of burst-suppression and persistent suppression), according to recent de nitions [14].
Neurological outcome was assessed at 3 months after CA, using the Glasgow-Pittsburgh Cerebral Performance Categories (CPC) [15], categorized as favorable outcome (CPC 1 = full recovery and 2 = moderate disability, returned home) or UO (including CPC 3 = severe disability, at rehabilitation facility, 4 = vegetative state and 5 = death). The outcome was evaluated during follow-up visits and/or by telephone interview.

Study Outcomes
The primary endpoint of this study was to examine the prognostic accuracy of PPI in predicting UO in comatose patients after CA. Secondary outcome included the agreement between PPI and NPi to predict UO in this patients' population.

Statistical Analysis
Discrete variables were expressed as counts (percentage) and continuous variables as means ± standard deviation (SD) or median [25 th to 75 th percentiles], as appropriate. The Kolmogorov-Smirnov test was used, and histograms and normal-quartile plots examined to verify the normality of distribution of continuous variables. Differences between groups was assessed using a χ-square or Fisher's exact test for categorical variables, as appropriate, a t-test or a Mann-Whitney-U test for comparison of continuous variables between two groups and an ANOVA test or Kruskal-Wallis test for comparison of continuous independent variables among more the 2 groups. In the latter case, to assess the pairwise comparison, a Bonferroni correction was used. To assess the predictive value of the NPi and PPI, we calculated the area under the receiver operating characteristics curve (AUROC); comparison between AUROCs was performed using the DeLong test. The prognostic performance of each predictor was analyzed calculating the speci city, sensitivity, positive predictive value (PPV), negative predictive value (NPV). False positive rate (FPR) for each tool was calculated as False Positive / FO. We therefore used an NPi ≤2 as the most speci c threshold to predict UO (as previously described) [7] and identi ed the PPI threshold showing 100% speci city; the agreement between the two indices was evaluated by comparing the number of patients with "predicted UO" with the two pupillometries, using the Cohen's kappa coe cient. Correlation between PPI and NPi was performed using Pearson's correlation analysis; the degree of correlation was de ned according to the coe cient of correlation as strong (higher than 0.6), moderate (between 0.4 and 0.6) or weak (below 0.4). A p<0.05 was considered statistically signi cant. Statistical analyses were performed using IBM SPSS Statistics 25.0 for Macintosh. We considered this study as a pilot study and no sample size calculation was performed, therefore a "convenient" sample size of 100 patients was selected.

Agreement and correlation between PPI and NPi values
On day 1, a total of 7 patients had concomitant PPI=1 and NPi ≤2, while 11 showed NPi >2 and PPI=1. Among the remaining 83 patients, 1 had NPI≤2 with PPI>1 and 82 had PPI >1 with NPi >2. Thus, the Cohen's k coe cient was 0.48. On day 2, a total of 6 patients had concomitant PPI=1 and NPi ≤2, while 12 showed NPi>2 and PPI=1; the coe cient of agreement was 0.42. Moreover, NPi and PPI showed a moderate correlation both on day 1 and day 2 ( Figure 2).

Characteristics of the patients according to predictors of UO at 3 months
On day 2, 6 patients showed both NPi ≤2 and PPI=1, 13 patients showed either NPi ≤2 or PPI=1 and 83 had no pathological NPi and PPI. Patients with both NPi ≤2 and PPI=1 received a higher dose of epinephrine during resuscitation attempts and had higher lactate on admission when compared to patients with no pathological NPi and PPI. Moreover, patients with no pathological NPi and PPI showed lower NSE levels and less frequently HMPs on EEG than others (Supplemental Table 2).

Discussion
This is the rst study evaluating the prognostic role of PPI to predict unfavorable outcome in comatose CA patients. Our ndings showed that a PPI on day 2 presented a speci city of 100% to predict UO, with a higher sensitivity than NPi. Also, a moderate agreement between PPI and NPi values was observed, suggesting that these two indices are not inter-changeable.
Pupillary size depends on the balance of sympathetic and parasympathetic systems that are connected into the midbrain, and by the neurons that originate in locus coeruleus, colliculi and cingulate cortex [16]. The pupillary dilation re ex is usually considered under the in uence of the sympathetic system. Sympathetic innervation of the radial muscle arises from pre-ganglionic neurons located in the C8-T2 spinal cord (i.e. the so-called ciliospinal center of Budge), and a painful stimulus, in particular when arising from the neck and upper trunk, can activate these groups of cells via a spinal re ex and project to the noradrenergic postganglionic neurons in the superior cervical ganglion, which innervate the dilator pupillae muscle [17].
Although several studies assessed the role of pupillary constriction to light re ex to predict unfavorable outcome after CA [7,18,19], no data are available on pupillary dilation to pain. Interestingly, pupillary dilation to nociceptive stimulation was 100% speci c and more sensitive than other pupillary variables derived from light stimulation, including NPi. Among 19 patients who presented PPI=1 or NPi≤2, only 6 of them had both concomitant present and most presented only a pathological PPI value. A possible explanation of these results may be due to the different location of the structures implicated in the genesis of the re ex. PPI might be considered as a "quanti cation" of the spinocilial re ex. Previous studies showed that such re ex is mediated by supraspinal structures, located mostly in the thalamus [20]. In fact, Yang et al. assessed the role of sympathetic nervous system on spinocilial re ex concluding that its origin is sympathetic in awake subject. Moreover, in brain dead patients, spinocilial re ex was absent after a high-intensity tetanic stimulation of somatic nociceptors [21]. A second explanation would be related to the different ischemic tolerance of cerebral structures involved in pupillary light re ex and spinocilial re ex after CA. In one study, the recovery of cranial nerve re exes occurred in a xed order, irrespective of patients' outcome, and was characterized by a progressive return of pupillary light re exes, which preceded spinocilial re exes [22]. As one patient had a PPI=1 on day 1 and eventually presented a favorable neurological recovery, one may argue that early assessment of PPI, in particular during sedatives and opioids administration, would result in false prediction of neurological outcome. However, the proportion of patients with PPI=1 was similar between those with and without opioids on day 2. Interestingly, PPI of 1 can be observed, although rarely, in the operative room in patients without brain injury, because of analgesic administration during surgical procedures [8, 23] (i.e. it is not always a sign on extended brain injury). However, abnormal NPi in the absence of severe post-anoxic brain damage has also been reported in association with sevo urane and ketamine therapy [24] thus underlying how all prognostic index might have suffer from confounders and be potentially biased with false prediction of poor outcome. As such, future studies should evaluate PPI in larger cohorts of CA patients in the absence of any confounders and correlate PPI ndings with additional predictors of poor outcome in this setting, such as biomarkers or brain imaging. Finally, we were unable to assess whether clinical characteristics of the patients might explain the discrepancies between patients with both abnormal NPi and PPI and those with only one of them, due to the limited number of events.
This study had several limitations to acknowledge. First, this was a pilot study and the number of patients included was relatively small. Second, the study was monocentric and local protocol of postresuscitation care and/or limitation of life-sustaining therapies may differ from other centers, which would potentially limit replicability and generalizability of ndings. However, PPI values were not used as a prognostication tool, therefore reducing the risk of self-ful lling prophecy. Third, we did not speci cally consider the cumulative dose of opioids and sedatives, but only their administration. Fourth, the resting size of the pupil progressively decreases after the fourth decade of life and this aspect may have interfered with pupillary response to light or pain. Finally, no data on brain imaging (CT-scan, MRI) are available, this aspect would have provided a more precise characterization of the brain injury showing the structures of the brain involved.

Conclusions
Our study showed that PPI=1 on day 2 after arrest was 100% speci c to predict unfavorable outcome in comatose cardiac arrest patients, with a higher sensitivity than NPi. Further larger multicentric studies are necessary to better de ne the prognostic role of different pupillometry derived indices in clinical practice.

Declarations
Ethics approval and consent to participate The original study protocol was approved by the ethics committees and the informed written consent was waived as pupillometry, in our institution, is standard of care.

Consent for publication
Not applicable

Availability of data and materials
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests
FST is scienti c advisor for NeurOptics inc. Other authors have no con ict of interest to declare.  Tables   Table 1. Characteristics of study population, according to neurological outcome at 3 months (FO = favorable; UO = unfavorable).
Data are expressed as counts (percentage) and median [Interquartile Range]. NPi = Neurological pupil index; CH = percentage of constriction, PPI = Pupillary pain index