Comparison of two automated pupillometry indices in patients with acute brain injury: the prospective multicentre LYNX cohort study

Abstract

Background

Automated pupillometry quantifies the pupillary light reflex in acute brain injury. The neurological pupil index (NPi) is the established composite index and carries prognostic value. A second index from a different device, the quantitative pupil index (QPi), has recently become available, but whether it conveys equivalent information in non-anoxic acute brain injury is unknown.

Methods

LYNX was a prospective cohort study in three European intensive care units (September 2022–July 2024). Consecutive adults with traumatic brain injury, subarachnoid or intracerebral haemorrhage, or ischaemic stroke ventilated for more than 12 h were enrolled and followed for 6 months. Paired NPi (NPi-300) and QPi (NeuroLight) measurements were obtained at least every 6 h to day 7. Ancillary cohorts of healthy volunteers (dark vs. bright light) and surgical patients (before and after sufentanil and propofol) were also studied. The primary outcome was the correlation between the two indices; secondary outcomes were their association with intracranial hypertension and unfavourable 6-month outcome (Glasgow Outcome Scale–Extended ≤ 4). Values below 3 were considered abnormal, with QPi thresholds of 2 and 4 as sensitivity analyses; multivariable models were adjusted for age, sex, admission Glasgow Coma Scale score and focal deficit.

Results

Among 108 patients (median age 57 years [IQR 44–67], 60.2% male), 7089 paired measurements were obtained. The two indices were moderately to strongly correlated (r = 0.717, 95% CI 0.705–0.730). Each 10-percentage-point increase in abnormal values independently predicted unfavourable outcome (NPi: OR 1.49, 95% CI 1.09–2.04, p = 0.013; QPi: OR 1.76, 95% CI 1.31–2.37, p < 0.001). QPi was associated with intracranial hypertension (OR 1.38, 95% CI 1.04–1.82, p = 0.025), whereas NPi was not. Ambient lighting and contralateral eye opening had no meaningful effect in 75 volunteers. In 20 sedated non–brain-injured patients, propofol reduced QPi to a median of 1 (IQR 1–3) and NPi to 4.0 (IQR 3.6–4.3).

Conclusions

NPi and QPi were correlated and both predicted 6-month outcome, but QPi was far more sensitive to anaesthetics, which probably confounds its association with intracranial hypertension. The indices are not interchangeable in sedated critically ill patients.

Trial registration

ClinicalTrials.gov, NCT05567978. Registered 5 October 2022.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s13054-026-06360-8.

Background

Assessment of pupillary light reactivity (PLR) is an integral component of daily clinical practice in patients with acute brain injury (ABI) [1] and provides valuable diagnostic and prognostic information [2, 3]. Traditionally, PLR has been evaluated using a handheld light source (e.g. penlight or flashlight), resulting in a qualitative and non-standardised assessment and providing measurements biased by inter-observer variability and inconsistencies in the technique used [4]. Automated pupillometry has addressed many of these limitations by enabling standardised, quantitative and reproducible measurement of PLR and associated dynamic variables, including constriction amplitude, velocity, dilation velocity and latency [5, 6]. These parameters can be integrated into composite indices, such as the neurological pupil index (NPi), a scalar index ranging from 0 to 4.9 that reflects overall pupillary reactivity and midbrain function [7]. Importantly, NPi values remain within the normal range at doses of sedative and analgesic agents that substantially alter other pupillary parameters, such as pupil diameter and pupillary unrest. The NPi is therefore relatively resistant to, rather than unaffected by, sedation and analgesia [8–10]. It has also demonstrated consistent associations with neurological outcomes across several clinical settings; indeed, while large cohort studies showed that low NPi was highly specific to predict unfavourable neurological outcome in patients with hypoxic–ischaemic brain injury [11–13], a recent multicentre study reported that persistent low NPi values had significant prognostic information in non-anoxic ABI, such as traumatic brain injury (TBI), subarachnoid haemorrhage (SAH) and intracerebral haemorrhage (ICH) [14].

Another quantitative index derived from automated pupillometry has recently been proposed, the quantitative pupil index (QPi), designed to quantify the pupillary constriction response to standardised light stimuli. While NPi is increasingly used as a reference tool in neurocritical care, it remains unclear whether QPi captures similar or distinct physiological information and whether it offers incremental clinical value. One recent study showed that abnormal QPi values had comparable specificity and a higher sensitivity than NPi in predicting unfavourable neurological outcome in patients with hypoxic–ischaemic brain injury [15]; however, no data regarding the clinical significance and prognostic performance of QPi in non-anoxic ABI patients are available. The influence of opioids on the NPi and on the individual dynamic parameters of the pupillary light reflex has, by contrast, been characterised in volunteers and in surgical patients [8–10]. The behaviour of the QPi has not been systematically explored in other conditions, namely in healthy volunteers, under different ambient lighting conditions, and during exposure to opioids and hypnotics, which are known to affect pupillary dynamics.

Therefore, the objectives of this study were three-fold: (i) to evaluate the agreement and potential differences between QPi and NPi across a range of clinical conditions; (ii) to explore their association with clinical outcomes in patients with ABI; and (iii) to assess their respective profiles in healthy volunteers and in surgical patients receiving sedation and opioid-based analgesia.

Methods

Study design and setting

The LYNX study (pupiLlometrY iNdeX in acute brain injury; registered as NCT05567978) was a prospective, observational, multicentre, international cohort study, reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [16] (see Additional file 2). Ethics approval and informed consent were obtained in accordance with European regulations. ABI patients were enrolled between September 2022 and July 2024 in three European centres where both NPi and QPi were used in routine neuromonitoring, with follow-up to 6 months. Eligible patients were adults (≥ 18 years) admitted to the intensive care unit (ICU) with acute TBI, ICH, SAH or ischaemic stroke requiring endotracheal intubation and mechanical ventilation for more than 12 h due to their neurological condition. Automated pupillometry using both devices was also required as part of standard clinical evaluation. Patients with significant facial or ocular trauma that could interfere with reliable pupillometric assessment were excluded. Subarachnoid or intraparenchymal haemorrhage of traumatic origin was classified in the TBI category, so that the SAH and ICH categories comprised non-traumatic cases exclusively. The aneurysmal or non-aneurysmal origin of subarachnoid haemorrhage, and the modality of aneurysm securing, were not recorded in the case report form. Seventy-five healthy volunteers were included at the Hôpital Universitaire de Bruxelles (HUB), Brussels, Belgium, and 20 patients requiring opioids during surgical procedures at the Department of Anaesthesia, Grenoble-Alpes University Hospital, Grenoble, France, within the same study period. Healthy volunteers were assessed in a windowless 9 m² room under two standardised ambient lighting conditions: a light condition, in which illuminance at the level of the eye was 300 lx, and a dark condition, in which illuminance was below the detection threshold of the meter and recorded as 0 lx. Illuminance was measured with a smartphone-based luxmeter application rather than with a calibrated photometer, so these values should be regarded as approximate. Twenty-five additional volunteers were assessed in the same room with the contralateral eye both open and occluded. In brain-injured patients, measurements were obtained under routine intensive care unit lighting, which was not standardised.

Demographics, pupillometry data and outcomes

Data were prospectively recorded in an electronic case report form (eCRF) developed using REDCap and hosted at Grenoble-Alpes University. Data extracted from medical records included demographic characteristics (age, sex), comorbidities, reason for ICU admission, initial clinical presentation, severity of acute brain injury, and ICU management, including therapeutic interventions, intracranial pressure (ICP; when available), therapeutic intensity level and neuroimaging findings.

The NPi was obtained using the NPi-300 device (NeurOptics, Irvine, CA, USA), while the QPi was obtained via the NeuroLight device (ID-MED, Marseille, France), as previously reported [14, 15]. The two devices differ in their acquisition geometry: the NeuroLight is applied with an eyecup that occludes the measured eye, whereas the NPi-300 is not, so that the measured eye remains exposed to ambient light during acquisition. Because the QPi is derived from the amplitude of constriction expressed relative to the resting pupil diameter, this difference is expected to influence baseline pupil diameter and hence the computed index. Values of NPi and QPi were immediately transferred into the eCRF. Pupillary assessments were performed at least every 6 h for both eyes, from ICU admission up to day 7. QPi and NPi values < 3 were considered abnormal. It should be noted that this threshold was derived from validation studies of NPi; no device-specific threshold has been formally established for QPi. Moreover, QPi is recorded as an integer value (no decimal), which limits threshold precision. In the sedated non–brain-injured cohort, resting pupil diameter was recorded at each of the three anaesthetic steps together with the two indices; pupil diameter was not available for brain-injured patients or for healthy volunteers.

Neurological outcome was assessed at 6 months using the Glasgow Outcome Scale–Extended (GOSE), obtained through structured telephone interviews with patients and/or their relatives using a validated questionnaire [17]. Unfavourable neurological outcome was defined as a GOSE score ≤ 4. The date and cause of death were also recorded. Intracranial hypertension was defined as an ICP value > 20 mmHg for at least 5 min [18]. Data were securely stored at Grenoble-Alpes University, in compliance with the European Union General Data Protection Regulation (EU 2016/679).

Study outcomes

The primary outcome was the correlation between QPi and NPi in patients with non-anoxic brain injury. Secondary outcomes were the association of both indices with (a) intracranial hypertension and (b) unfavourable neurological outcome; (c) their comparison in healthy volunteers across different light conditions; and (d) their comparison after sufentanil/propofol administration in patients undergoing extracerebral surgery.

Statistical analysis

No sample size calculation was performed, as this was an exploratory study conducted in the absence of prior reference data. Analyses used available data without imputation of missing values, and the primary analysis population comprised all enrolled brain-injured patients. Categorical variables are summarised as counts and percentages, and continuous variables as means with standard deviations or medians with interquartile ranges, as appropriate. Distributional assumptions, including normality and heteroscedasticity, were assessed graphically. The significance threshold was set at 0.05, with no adjustment for multiplicity, as all secondary analyses were pre-specified as exploratory and hypothesis-generating. No interim analysis was performed.

The linear association between NPi and QPi was quantified using Pearson’s correlation coefficient with its 95% confidence interval (CI), with Spearman’s rank correlation as a sensitivity analysis if normality assumptions were violated. If the correlation and its confidence interval exceeded 0.5, a mixed-effects linear regression model was fitted to evaluate the slope of agreement between the two indices, with patient-level variability as a random effect and a sensitivity analysis including time as a fixed effect, with or without a time interaction term. Agreement was explored graphically using a Bland–Altman plot for repeated observations [19].

For secondary objectives, associations between pupillometry indices and intracranial hypertension, neurological outcome or survival were assessed in brain-injured patients with at least five pupillometry measurements. The percentage of values strictly below 3 was related to the occurrence of intracranial hypertension using logistic regression and to the number of episodes using negative binomial regression; unfavourable neurological outcome was analysed using logistic regression with the same independent variable. Mortality was analysed using Cox proportional hazards models incorporating pupillometry (lowest value < 3 vs. ≥ 3) as a time-dependent covariate; when proportional hazards assumptions were not met, logistic regression for 6-month mortality was performed instead. Multivariable logistic regression models for unfavourable neurological outcome included age, sex, Glasgow Coma Scale score at admission and the presence of a focal neurological deficit as covariates, selected a priori on clinical grounds and constrained in number by the number of events; sedation depth could not be included as a covariate, because neither sedation scores nor cumulative drug doses were recorded with sufficient precision in the case report form. Because the threshold of 3 used to define an abnormal QPi was transposed from the NPi rather than derived for the QPi itself, all analyses based on the proportion of abnormal QPi values were repeated with thresholds of 2 and 4 as sensitivity analyses.

In sedated non–brain-injured surgical patients, measurements obtained immediately before and after opioid administration were analysed using mixed-effects linear models, with time and sedation status as fixed effects and patient identity as a random effect. Similar models, incorporating patient group and time as fixed effects and an interaction term when relevant, compared healthy volunteers with brain-injured patients. Additional exploratory mixed-effects linear and logistic models examined the influence of light exposure among healthy volunteers. All analyses were performed separately for NPi and QPi.

Results

Brain injury population

Among 666 eligible patients with non-anoxic acute brain injury, a total of 108 consecutive patients were included (34 TBI, 35 SAH, 29 ICH, 10 ischaemic strokes; see study flowchart in Additional file 1: Fig. S1). The reasons for non-inclusion of the remaining 558 patients are detailed in the flowchart: 258 patients were not intubated and 27 were intubated for less than 24 h; 76 had treatment limitations or died within 24 h; 75 were screened more than 24 h after admission; a pupillometer was unavailable for 47; 24 were aged 18 years or younger; 15 had no brain injury; 15 had facial trauma; 13 were not covered by health insurance; 3 had previous ocular disease or surgery; 3 patients or families declined participation; 1 was a legally protected adult; and 1 was already enrolled in another study. Median age was 57 years (IQR 44–67) and 60.2% were male. On admission, 38.5% had severe impairment of consciousness (GCS 3–5), the pupillary light reflex was bilaterally reactive in 83.7%, and 45.2% presented with a focal neurological deficit (Table 1).

Table 1.

Characteristics of patients with acute brain injury at admission

	All participants N = 108	Traumatic brain injury N = 34	Subarachnoid haemorrhage N = 35	Intracerebral haemorrhage N = 29	Ischaemic stroke N = 10
Age, median [IQR], years	57 [44–66.5]	43 [31–63]	60 [54–68]	57 [46–66]	67.5 [61–69]
Male, n (%)	65/108 (60.2)	23/34 (67.7)	11/35 (31.4)	22/29 (75.9)	9/10 (90.0)
GCS score at admission, n (%)
3–5	40/104 (38.5)	17/34 (50.0)	10/33 (30.3)	9/28 (32.1)	4/9 (44.4)
6–8	20/104 (19.2)	5/34 (14.7)	9/33 (27.3)	5/28 (17.9)	1/9 (11.1)
9–15	44/104 (42.3)	12/34 (35.3)	14/33 (42.4)	14/28 (50.0)	4/9 (44.4)
Pupillary reactivity, n (%)
Both pupils reacting	87/104 (83.7)	29/33 (87.9)	31/34 (91.2)	19/29 (65.5)	8/8 (100.0)
One pupil reacting	12/104 (11.5)	1/33 (3.0)	3/34 (8.8)	8/29 (27.6)	0/8 (0.0)
No pupils reacting	5/104 (4.8)	3/33 (9.1)	0/34 (0.0)	2/29 (6.9)	0/8 (0.0)
Focal neurological deficit, n (%)	48/106 (45.2)	8/33 (24.2)	15/35 (42.9)	16/28 (57.1)	9/10 (90.0)
Cardiovascular history, n (%)	49/108 (45.4)	14/34 (41.2)	13/35 (37.1)	15/29 (51.7)	7/10 (70.0)
Neurological history, n (%)	13/108 (12.0)	1/34 (2.9)	5/35 (14.3)	6/29 (20.7)	1/10 (10.0)
Respiratory history, n (%)	10/107 (9.4)	4/34 (11.8)	3/35 (8.6)	3/28 (10.7)	0/10 (0.0)
Liver and kidney history, n (%)	6/107 (5.6)	1/34 (2.9)	1/35 (2.9)	3/28 (10.7)	1/10 (10.0)
Endocrinological history, n (%)	16/108 (14.8)	6/34 (17.7)	4/35 (11.4)	4/29 (13.8)	2/10 (20.0)
Rate of NPi < 3, median [IQR]a	0.012 [0.0; 0.156]	0.014 [0.0; 0.156]	0.010 [0.0; 0.114]	0.017 [0.0; 0.227]	0.021 [0.0; 0.208]
Rate of QPi < 3, median [IQR]a	0.127 [0.022; 0.500]	0.165 [0.030; 0.513]	0.095 [0.022; 0.302]	0.159 [0.0; 0.758]	0.202 [0.042; 0.475]
Number of intracranial hypertension episodes, median [IQR]b	6 [1; 17]	8 [0; 17]	3 [1; 11]	6 [1; 19]	3 [1; 14]
At least one intracranial hypertension episode, n (%)b	62 (77.5)	21 (72.4)	23 (79.3)	14 (82.4)	4 (80.0)
GOSE at 6 months, n (%)c
1	34/98 (34.7)	9/31 (29.0)	10/33 (30.3)	9/25 (36.0)	6/9 (66.7)
2	2/98 (2.0)	1/31 (3.2)	1/33 (3.0)	0/25 (0.0)	0/9 (0.0)
3	5/98 (5.1)	4/31 (12.9)	0/33 (0.0)	1/25 (4.0)	0/9 (0.0)
4	13/98 (13.3)	1/31 (3.2)	5/33 (15.2)	5/25 (20.0)	2/9 (22.2)
5	12/98 (12.2)	4/31 (12.9)	6/33 (18.2)	1/25 (4.0)	1/9 (11.1)
6	14/98 (14.3)	5/31 (16.1)	4/33 (12.1)	5/25 (20.0)	0/9 (0.0)
7	13/98 (13.3)	4/31 (12.9)	6/33 (18.2)	3/25 (12.0)	0/9 (0.0)
8	5/98 (5.1)	3/31 (9.7)	1/33 (3.0)	1/25 (4.0)	0/9 (0.0)

Data are presented as median [interquartile range] or n (%). ^aAnalyses were restricted to the 106 patients with at least five measurements. ^bData were available for the 80 patients with ICP monitoring; one patient was excluded because of extreme values. ^cGOSE was available for 98 patients, as 8 patients were lost to follow-up. Subarachnoid haemorrhage and intracerebral haemorrhage were non-traumatic in all cases; the aneurysmal or non-aneurysmal origin of subarachnoid haemorrhage and the modality of aneurysm securing were not recorded. The 13 patients with a neurological history comprised previous ischaemic stroke or transient ischaemic attack in six, epilepsy in four, a cerebrovascular malformation or a previously treated aneurysm in three (moyamoya disease, thalamic arteriovenous malformation, embolised anterior communicating artery aneurysm), previous traumatic brain injury in one, previous cerebral malaria in one and an anxiety disorder in one; several patients had more than one condition. Previous ocular disease or surgery was an exclusion criterion, on the basis of which three screened patients were not enrolled (Fig. S1); among those enrolled, one had blindness of the left eye following a previous stroke and one had undergone decompression of the optic pathways. GCS, Glasgow Coma Scale; GOSE, Glasgow Outcome Scale–Extended; ICP, intracranial pressure; IQR, interquartile range; NPi, neurological pupil index; QPi, quantitative pupil index

Correlation between NPi and QPi in brain-injured patients

Across the acute brain injury cohort, 7339 NPi and 7154 QPi measurements were recorded, of which 7089 were paired. Distributions of all measurements are shown in Additional file 1: Figs. S2A and S3A. NPi and QPi demonstrated a significant positive association across the study population; Pearson’s correlation coefficient was 0.717 (95% CI 0.705–0.730), while Spearman’s rho was 0.688 (95% CI 0.675–0.702), indicating a moderate-to-strong correlation between the two indices (Fig. 1A).

Fig. 1.

Correlation and agreement between NPi and QPi in patients with acute brain injury. A Scatterplot of paired NPi and QPi measurements (N = 7089) with regression line; Pearson’s r = 0.717. B Bland–Altman plot of NPi versus QPi adjusted for repeated measures (method of Myles and Cui). The mean difference and 95% limits of agreement are shown. NPi, neurological pupil index; QPi, quantitative pupil index

In a mixed-effects model accounting for repeated measurements within patients, QPi was significantly associated with NPi (β = 0.51, 95% CI 0.48–0.53; p < 0.001). Neither time (β = 0.008; p = 0.509) nor its interaction with QPi (β = −0.006; p = 0.064) was significant, suggesting a stable relationship between the two indices over the observation period. Bland–Altman analysis adjusted for repeated measures confirmed good agreement across the measurement range (Fig. 1B).

Association with intracranial hypertension

The median proportion of abnormal measurements was 0.012 for NPi and 0.127 for QPi among the 106 patients with at least five measurements (Table 1; distribution of measurements in Additional file 1: Figs. S4 and S5). Among the 80 patients with ICP monitoring, 77% experienced at least one episode of intracranial hypertension (Table 1; the number of episodes per patient is shown in Additional file 1: Fig. S6). One patient with traumatic brain injury, who was in intracranial hypertension almost continuously and whose measurements were abnormal in 52% of NPi and 96% of QPi assessments, recorded 1162 episodes of intracranial hypertension, whereas the highest count among all other patients was 67; this patient was therefore identified as an extreme outlier and excluded from the primary analyses. In a sensitivity analysis retaining this patient, the negative binomial coefficients per 10-percentage-point increase in abnormal values were 0.47 (95% CI 0.29–0.65; p < 0.001) for NPi and 0.30 (95% CI 0.20–0.39; p < 0.001) for QPi, both associations becoming statistically significant. In univariate logistic regression analyses, a 10-percentage-point increase in the proportion of QPi values < 3 was significantly associated with the occurrence of intracranial hypertension (OR 1.38, 95% CI 1.04–1.82; p = 0.025). In contrast, the proportion of NPi values < 3 was not significantly associated with intracranial hypertension (OR 1.42, 95% CI 0.89–2.24; p = 0.142). Negative binomial regression accounting for the number of episodes gave similar findings; abnormal QPi measurements remained associated with a greater burden of intracranial hypertension (β = 0.15, 95% CI 0.05–0.24; p = 0.004), whereas the association for NPi did not reach statistical significance (β = 0.15, 95% CI − 0.01 to 0.31; p = 0.062). Analyses using alternative QPi thresholds were concordant. For the occurrence of at least one episode, the odds ratio per 10-percentage-point increase in abnormal values was 1.60 (95% CI 0.98–2.63; p = 0.063) with a threshold of 2 and 1.23 (95% CI 1.03–1.45; p = 0.020) with a threshold of 4, and the corresponding negative binomial coefficients were 0.18 (95% CI 0.05–0.31; p = 0.008) and 0.11 (95% CI 0.03–0.20; p = 0.010) (Additional file 1: Table S1).

Association with neurological outcome and mortality

Six-month neurological outcome was available for 98 patients; an unfavourable outcome was observed in 34.7% of the cohort (Additional file 1: Fig. S7; Table 1). In univariate analyses, each 10-percentage-point increase in the proportion of abnormal measurements increased the odds of unfavourable outcome by 39% for NPi (OR 1.39, 95% CI 1.07–1.81; p = 0.015) and by 40% for QPi (OR 1.40, 95% CI 1.17–1.68; p < 0.001). In the multivariable models, each 10-percentage-point increase in abnormal NPi measurements was associated with a 49% increase in the odds of unfavourable outcome (OR 1.49, 95% CI 1.09–2.04; p = 0.013), whereas the corresponding increase for QPi was 76% (OR 1.76, 95% CI 1.31–2.37; p < 0.001). Sensitivity analyses varying the QPi threshold were concordant: each 10-percentage-point increase in the proportion of QPi values below 2 was associated with an odds ratio of 2.08 (95% CI 1.34–3.21; p = 0.001) for unfavourable outcome, and each 10-percentage-point increase in the proportion of values below 4 with an odds ratio of 1.48 (95% CI 1.22–1.79; p < 0.001) (n = 93; Additional file 1: Table S1).

Vital status at 6 months was available for all 98 patients with completed follow-up. One patient was excluded from the time-to-event analysis because the date of death could not be ascertained, leaving 97 patients for the survival models. After adjustment for relevant covariates, each 10-percentage-point increase in abnormal NPi values was associated with a nearly three-fold increase in mortality risk (adjusted HR 2.78, 95% CI 1.15–6.70; p = 0.023). In contrast, although a similar trend was observed for QPi, the association did not reach statistical significance (adjusted HR 3.56, 95% CI 0.82–15.40; p = 0.089). Kaplan–Meier survival curves further supported these findings (Fig. 2).

Fig. 2.

Kaplan–Meier survival curves according to abnormal pupillometry values. A Survival according to the presence of NPi < 3 at any measurement (N = 97). B Survival according to the presence of QPi < 3 at any measurement (N = 97). NPi, neurological pupil index; QPi, quantitative pupil index

Healthy volunteers

Fifty healthy volunteers (age 32 [29–40] years, 46% male) had higher and less variable NPi and QPi values compared with brain-injured patients (median NPi 4.5 vs. 4.3; median QPi 5 vs. 3; see Additional file 1: Figs. S2 and S3). Regression models confirmed lower pupillometry indices in brain-injured patients after adjustment for age and sex.

Analysis of lighting conditions in volunteers showed a small but significant increase in NPi in dark versus light environments (β = 0.13; 95% CI 0.09–0.17; p < 0.001), while QPi was not significantly affected (β = 0.03; 95% CI − 0.01 to 0.07; p = 0.172) (Additional file 1: Figs. S8 and S9). Similar findings were observed regardless of whether the contralateral eye was occluded during pupillometric assessment (extra 25 healthy volunteers; NPi: coefficient 0.07, 95% CI − 0.01 to 0.14, p = 0.075; QPi: coefficient 0.02, 95% CI − 0.03 to 0.07, p = 0.120). No volunteers exhibited indices < 3. Ambient illuminance was 300 lx in the light condition and below the detection threshold of the meter (0 lx) in the dark condition.

Sedated non–brain-injured patients

Twenty sedated patients without acute brain injury were included in the analysis (Table 2). Both pupillary indices were influenced by sedative exposure, although the magnitude of the effect differed between the two measures. Following propofol administration, NPi values showed a modest reduction, decreasing from a median of 4.5 (IQR 4.1–4.6) during sufentanil-only administration to 4.0 (IQR 3.6–4.3). In contrast, QPi appeared more sensitive to the effects of anaesthesia, with values decreasing to a median of 1 (IQR 1–3) under propofol. Mixed-effects regression analyses accounting for repeated measurements confirmed a significant association between anaesthetic exposure and lower pupillary index values, with both propofol and sufentanil contributing to the observed reductions (Fig. 3). Resting pupil diameter decreased in parallel, from a median of 3.10 mm (IQR 2.57–3.45) before anaesthesia to 2.33 mm (IQR 2.02–2.65) after sufentanil and 2.13 mm (IQR 1.92–2.32) after sufentanil plus propofol. QPi values below 3 were observed almost exclusively at resting diameters under approximately 2.7 mm, whereas NPi values remained at or above 3 across the whole range of diameters (Additional file 1: Fig. S10).

Table 2.

Characteristics of sedated non–brain-injured patients

	All participants N = 20
Age, median [IQR], years	61 [28.5; 80.5]
Male, n (%)	7 (35.0)
ASA score at admission, n (%)
1	7/19 (36.8)
2	8/19 (42.1)
3	3/19 (15.8)
4	1/19 (5.3)
Sufentanil, median [IQR], µg	15 [10; 20]
Propofol, median [IQR], mga	180 [120; 200]

Data are presented as median [interquartile range] or n (%). ^aThree patients received other sedative agents. ASA, American Society of Anesthesiologists; IQR, interquartile range

Fig. 3.

NPi and QPi in sedated non–brain-injured patients according to anaesthetic regimen. Mean values and 95% confidence intervals of NPi and QPi before anaesthesia, after sufentanil, and after sufentanil plus propofol (N = 20). NPi, neurological pupil index; QPi, quantitative pupil index

Discussion

In this prospective multicentre cohort of patients with non-anoxic acute brain injury, QPi and NPi values were moderately to strongly correlated, and low values for both indices were independently associated with unfavourable neurological outcome after adjustment for major confounders. QPi, derived from a distinct automated pupillometry device, therefore captures physiological information closely related to that provided by the established NPi and may represent a clinically viable alternative for quantitative pupillary assessment in neurocritical care.

However, a central and clinically important finding of this study is that QPi was markedly more sensitive to anaesthetic drugs than NPi [8, 20]. In sedated non–brain-injured surgical patients, QPi values fell substantially with propofol administration (median 1 [IQR 1–3]), while NPi values were only modestly affected (median 4.0 [IQR 3.6–4.3]) [8, 9, 21]. This differential susceptibility to sedation has direct implications for the interpretation of QPi in the ICU context, where deep sedation is frequently required. Indeed, the stronger association observed between QPi and the occurrence of intracranial hypertension, compared with NPi, may not solely reflect a superior sensitivity to intracranial pressure elevations [22–25], but may in part be explained by confounding from sedation levels: patients experiencing more intracranial hypertension episodes may have received higher doses of sedative and analgesic agents, thereby lowering QPi independently from brainstem impairment [20, 26, 27]. This hypothesis, although not formally tested in the present study due to the absence of sedation adjustment in primary outcome models, is supported by the pharmacological data obtained in the surgical cohort and warrants careful consideration when interpreting QPi in mechanically ventilated patients. This interpretation is further supported by the expected direction of the pupillary response to raised intracranial pressure: because an increase in intracranial pressure typically produces pupillary dilation, it should, in isolation, tend to increase rather than decrease a percent-reflex-based index, so that the association we observed between abnormal QPi values and intracranial hypertension is difficult to attribute to intracranial pressure itself and is more plausibly explained by the concomitant escalation of sedative and analgesic therapy in these patients. These observations align closely with a recent editorial on the bedside use of the NPi, which argued that the clinical value of the index derives from its composite, normalised structure: by integrating several dynamic features of the pupillary light reflex into a single standardised score, the NPi is rendered less vulnerable to sedation, ambient conditions and technical artefacts than any individual raw pupillometric variable [28]. Our findings provide direct empirical support for this argument, as the NPi remained relatively stable after propofol administration whereas the QPi declined markedly, indicating that composite indices derived from different devices are not equally buffered against pharmacological confounding.

The differential sensitivity of the two indices to anaesthetic agents has a straightforward mechanistic explanation. The QPi is derived from the percent constriction of the pupil, that is, the difference between the resting and the minimum diameter expressed as a fraction of the resting diameter. The absolute constriction amplitude that the iris sphincter can generate is approximately proportional to the resting diameter, so that the percent reflex, and hence the QPi, declines as the pupil becomes smaller, independently of any change in the strength of the parasympathetic output [9, 10, 29, 30]. Below a resting diameter of approximately 2.5 mm this decline is steep: a 2 mm pupil would have to constrict by 0.66 mm to achieve the same 33% reflex as a 6 mm pupil constricting by 2 mm, which the mechanical properties of the iris preclude. The NPi, being a normalised composite of several dynamic parameters, is comparatively insensitive to baseline pupil diameter. Since opioids and hypnotics produce dose-dependent miosis, this predicts precisely what we observed: a collapse of the QPi to a median of 1 after propofol administration while the NPi remained at 4.0. Our own data support this directly: in the sedated non–brain-injured cohort, resting pupil diameter fell from 3.10 mm before anaesthesia to 2.33 mm after sufentanil and 2.13 mm after sufentanil plus propofol, so that the median diameter lay below 2.5 mm from the first anaesthetic step onwards, and QPi values below 3 were observed almost exclusively at resting diameters under approximately 2.7 mm, whereas the NPi remained at or above 3 across the whole range of diameters. It follows that the QPi should not be interpreted as an index of brainstem reflex integrity in miotic patients without concurrent knowledge of pupil diameter, and that recalibration of the index for small pupil diameters would be required before it could be used interchangeably with the NPi in sedated critically ill patients.

These results extend previous findings from cardiac arrest populations, in which QPi showed comparable or superior prognostic performance relative to NPi after hypoxic–ischaemic brain injury [12, 13, 15]. Our NPi findings replicate those of the ORANGE study, the largest prospective multicentre cohort in non-anoxic acute brain injury (514 patients in 13 hospitals across eight countries), in which abnormal NPi values were independently associated with unfavourable 6-month neurological outcome and with mortality [14]; the magnitude of the association we observed is comparable despite a cohort approximately five times smaller. The two cohorts also agree on what the NPi does not capture. In a secondary analysis of ORANGE restricted to the 318 patients who underwent invasive monitoring, comprising 8692 intracranial pressure measurements, the mean change in NPi was − 0.003 per 1 mmHg increase in intracranial pressure (95% credible interval − 0.006 to 0.000), and neither abnormal (< 3) nor absent (0) NPi values were associated with intracranial pressure (odds ratio 1.01, 95% credible interval 0.99–1.03, and odds ratio 1.03, 95% credible interval 0.99–1.06, respectively); the authors concluded that repeated NPi measurements may not be a sufficient replacement for invasive monitoring [31]. Our own estimate is concordant with that conclusion (OR 1.42, 95% CI 0.89–2.24; p = 0.142), and the two studies taken together indicate that the NPi does not provide an accurate measure of intracranial hypertension, notwithstanding the associations reported in earlier single-centre series [22–25]. This convergence also bears directly on the interpretation of our QPi result: if a normalised composite index shows no detectable relationship with intracranial pressure in a dataset of that size, the association we observed between abnormal QPi values and intracranial hypertension is unlikely to reflect a genuine sensitivity to intracranial pressure, and is more plausibly explained by the sedation received by the patients in whom intracranial hypertension occurs. LYNX further includes patients with ischaemic stroke, a population not represented in ORANGE. To our knowledge, this is the first systematic evaluation of QPi in a non-anoxic ABI cohort including traumatic brain injury, subarachnoid haemorrhage, intracerebral haemorrhage and ischaemic stroke [32, 33]. The moderate rather than perfect correlation across 7089 paired measurements indicates that the two devices do not provide fully redundant information, consistent with the distinct algorithms and light stimulation protocols used by the NPi-300 and NeuroLight systems [21]. Furthermore, the markedly higher proportion of abnormal measurements with QPi (12.7%) compared with NPi (1.2%), using the same threshold of < 3, likely reflects both the distinct measurement properties of the two devices and the limited precision of the QPi scale, which is recorded as an integer without decimal values. Device-specific thresholds, validated against clinical outcomes, may therefore be necessary before QPi can be used interchangeably with NPi. We did not attempt to derive an optimal QPi cut-off by receiver operating characteristic analysis, because the exposure in our models is not the QPi value itself but the proportion of measurements falling below a given value, so that such an analysis would not identify a threshold for the index. We repeated all analyses with thresholds of 2 and 4 instead, and the associations were preserved in direction and magnitude across the three thresholds, which supports the robustness of our findings while confirming that the value of 3 has no particular status for the QPi. That the median proportion of measurements below 4 reached 0.514 in our cohort, whereas healthy volunteers had almost exclusively values of 5, illustrates how strongly the choice of threshold governs the apparent frequency of abnormality. Nonetheless, Bland–Altman analysis confirmed acceptable agreement across the measurement range, without systematic bias [19].

In healthy volunteers, NPi was marginally but significantly higher in dark than in bright environments, while QPi was unaffected by lighting conditions [34]. Although this effect was small (β = 0.13; 95% CI 0.09–0.17), the differential sensitivity of the two indices to ambient light represents another dimension along which they diverge in their measurement properties. These observations are consistent with a recent prospective crossover study of the influence of ambient light on quantitative pupillometry [35].

With respect to mortality, NPi showed a significant association in adjusted survival analysis (HR per 10% increase in abnormal values 2.78; 95% CI 1.15–6.70), whereas the estimate for QPi did not reach significance (HR 3.56; 95% CI 0.82–15.40; p = 0.089), likely reflecting reduced statistical power in a relatively small cohort. The wide confidence intervals preclude firm conclusions regarding the prognostic value of QPi for mortality, and the limited number of outcome events relative to the number of predictors raises the risk of overfitting and effect-size inflation. These mortality analyses should therefore be considered strictly exploratory and hypothesis-generating, pending replication in adequately powered cohorts.

This study has several limitations. First, the sample size of 108 brain-injured patients limits the precision of estimates and statistical power for secondary endpoints, particularly mortality. Second, as an observational study, residual confounding cannot be excluded, and the absence of randomisation precludes causal inference. Third, the study was conducted in three European tertiary centres, which may limit generalisability to other healthcare systems, patient populations or devices from different manufacturers. Fourth, and most importantly, sedation levels were not systematically adjusted for in the primary outcome models, which represents a major methodological limitation given the demonstrated sensitivity of QPi to anaesthetic agents. Future studies comparing QPi and NPi should incorporate sedation depth as a formal covariate or perform pre-specified sensitivity analyses stratified by sedation status. The present dataset did not record sedation depth or cumulative drug doses with sufficient precision for either to be used as a covariate, and this question warrants a dedicated study. Fifth, resting pupil diameter was available only in the sedated non–brain-injured cohort and not in brain-injured patients or in healthy volunteers; because the QPi is computed as a percent constriction of the resting diameter, its values in the main cohort cannot be interpreted independently of pupil size, and the contribution of miosis to the abnormal QPi values observed there cannot be quantified. Sixth, only 108 of the 666 patients assessed for eligibility were enrolled. Most non-inclusions followed from the eligibility criteria themselves rather than from logistical constraints: 285 patients did not meet the requirement for mechanical ventilation lasting at least 24 h and 76 had treatment limitations or died within the first 24 h. A further 122 patients were missed for operational reasons, having been screened more than 24 h after admission or admitted when no pupillometer was available, which is likely to skew the cohort towards weekday admissions and towards periods of full device availability. The enrolled population is thus restricted to intubated patients surviving beyond the first day, which limits the generalisability of the absolute frequencies we report, although it is less likely to bias the comparison between two indices measured concurrently in the same patient.

Conclusions

In patients with non-anoxic acute brain injury, QPi and NPi values were moderately to strongly correlated and both showed significant prognostic value. However, QPi was substantially more sensitive to the effects of anaesthetic agents than NPi. These results suggest that QPi should not be interpreted as a straightforward equivalent of NPi in sedated critically ill patients, and that the two indices are not interchangeable. Pupil diameter should be reported alongside the QPi, and device-specific thresholds are required before the index can be used for prognostication in sedated patients.

Electronic Supplementary Material

Below is the link to the electronic supplementary material.

Abbreviations

ABI

Acute brain injury

ASA

American Society of Anesthesiologists

CI

Confidence interval

eCRF

electronic case report form

GCS

Glasgow Coma Scale

GOSE

Glasgow Outcome Scale–Extended

HR

Hazard ratio

ICH

Intracerebral haemorrhage

ICP

Intracranial pressure

ICU

Intensive care unit

IQR

Interquartile range

NPi

Neurological pupil index

OR

Odds ratio

PLR

Pupillary light reactivity

QPi

Quantitative pupil index

SAH

Subarachnoid haemorrhage

STROBE

Strengthening the Reporting of Observational Studies in Epidemiology

TBI

Traumatic brain injury

Author contributions

PB, TG and FST conceived and designed the study, interpreted the data and wrote the manuscript. AV performed the statistical analysis and was responsible for data management. ABO, AP, SZ, SV and MP participated in study management, data collection and data interpretation. All authors read and approved the final manuscript.

Funding

The study received no specific funding.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

The study was approved at the sponsor site by the Ethics Committee of Grenoble University on 6 September 2022 and was performed in accordance with the Declaration of Helsinki and the International Conference on Harmonisation Good Clinical Practice guidelines. Because comatose patients could not provide informed consent at the time of study recruitment, each centre applied the local or national law on lack of capacity. Patients who regained capacity at the follow-up visit were required to provide informed consent for the use of their sensitive data and for follow-up, or could decline research participation. Local investigators at the international study sites obtained approvals according to local regulations. The study was registered at ClinicalTrials.gov (NCT05567978).

Consent for publication

Not applicable.

Competing interests

FST has received lecture fees from BD, Medtronic, Integra and ZOLL, and personal fees as an advisory board member from Nihon Kohden, Werfen, Eurosets and NeurOptics, all outside the submitted work. PB has received personal fees as a consultant from NeurOptics. The other authors (ABO, AV, AP, SZ, SV, MP, TG) declare that they have no competing interests.