Prognosis value of pupillometry in COVID-19 patients admitted in intensive care unit

Abstract

Introduction

ICU patients with SARS-CoV-2-related pneumonia are at risk to develop a central dysautonomia which can contribute to mortality and respiratory failure. The pupillary size and its reactivity to light are controlled by the autonomic nervous system. Pupillometry parameters (PP) allow to predict outcomes in various acute brain injuries. We aim at assessing the most predictive PP of in-hospital mortality and the need for invasive mechanical ventilation (IV).

Material and methods

We led a prospective, two centers, observational study. We recruited adult patients admitted to ICU for a severe SARS-CoV-2 related pneumonia between April and August 2020. The pupillometry was performed at admission including the measurement of baseline pupillary diameter (PD), PD variations (PDV), pupillary constriction velocity (PCV) and latency (PDL).

Results

Fifty patients, 90 % males, aged 66 (60–70) years were included. Seven (14 %) patients died in hospital. The baseline PD (4.1 mm [3.5; 4.8] vs 2.6 mm [2.4; 4.0], P = 0.009), PDV (33 % [27; 39] vs 25 % [15; 36], P = 0.03) and PCV (3.5 mm.s⁻¹ [2.8; 4.4] vs 2.0 mm.s⁻¹ [1.9; 3.8], P = 0.02) were significantly lower in patients who will die. A PD value <2.75 mm was the most predictive parameter of in-hospital mortality, with an AUC = 0.81, CI 95 % [0.63; 0.99]. Twenty-four (48 %) patients required IV. PD and PDV were significantly lower in patients who were intubated (3.5 mm [2.8; 4.4] vs 4.2 mm [3.9; 5.2], P = 0.03; 28 % [25; 36 %] vs 35 % [32; 40], P = 0.049, respectively).

Conclusions

A reduced baseline PD is associated with bad outcomes in COVID-19 patients admitted in ICU. It is likely to reflect a brainstem autonomic dysfunction.

Graphical abstract

1. Introduction

Patients with SARS-CoV-2 related pneumonia are at risk to develop neurological disorders, including stroke, encephalopathy and encephalitis (Chen et al., 2020; Filatov et al., 2020; Mao et al., 2020; Poyiadji et al., 2020; Zhao et al., 2020). These clinical observations support experimental evidence of SARS-CoV-2 neurotropism (Aghagoli et al., 2020; Bryche et al., 2020; Bullen et al., 2020).

The brainstem may be as a main target of viral invasion via the olfactive and trigeminal pathway but also of neuro-inflammatory process triggered by the systemic inflammatory response (Baig et al., 2020; Lechien et al., 2020). While the presence of viral material within the brainstem of autopsied patient is controversial, it has been evidenced in animal models (Ferren et al., 2021; Golden et al., 2022). The brainstem controls sleep-wake cycle via the ascending reticular activating system (ARAS), the vital functions via the autonomic nervous system (ANS) and brainstem reflexes, including the pupillary diameter (PD) and light reflex (PLR). In parallel, more and more authors report dysautonomia in patients with COVID-19 in the acute phase and as long-term sequelae (Dani et al., 2021; Goodman et al., 2021; Pan et al., 2021). A brainstem dysfunction (BSD) could contribute to disorders of consciousness (Patel and Hirsch, 2014) and organ failures and death, irrespectively of the severity lung injury and the systemic inflammatory response. This BSD might therefore be routinely assessed in patients with severe SARS-CoV-2 pneumonia admitted in intensive care unit (ICU).

We reasoned that quantitative measurement of PD and PLR could be a marker of BSD and as such have a prognostic value (Larson and Behrends, 2015). Indeed, it has been shown that light-induced variation in PD (PDV) is predictor of outcomes in various acute brain injuries (Behrends et al., 2012; Chen et al., 2011; Taylor et al., 2003; Yokobori et al., 2018), notably traumatic and post-anoxic coma (Behrends et al., 2012; Heimburger et al., 2016; Oddo et al., 2018; Yokobori et al., 2018).

The objectives of our prospective observational study were to assess whether basis PD and light-induced PD variation (PDV), pupillary constriction velocity (PCV) and latency are associated with in-hospital mortality (primary endpoint) and with the need for invasive mechanical ventilation (secondary endpoint) in patients admitted in ICU for SARS-CoV-2 pneumonia.

2. Methods

2.1. Study approval

This study was approved by independent institutional review boards (IRBN602020/CHUSTE for Saint-Etienne Academic Hospital and CPP Sud-Méditerranée V n°2020-A01003-36 for hospital Sainte-Anne). Written informed consent was obtained from all patients or a legal surrogate before enrollment in this study. All images are taken from anonymized data from patients who provided their prior consent. The CNIL (Commission Nationale Informatique et Libertés) has also given its agreement for the use of data within the framework of this study.

2.2. Design

This was a prospective, two centers, cohort observational study. We included all consecutive patients, aged 18 years or older, admitted to the ICU of Saint-Etienne University Hospital and Saint-Anne Hospital between April 14th and August 13th of 2020 for a SARS-CoV-2 infection confirmed by a nasopharyngeal Reverse Transcriptase polymerase chain reaction (RT-PCR). All included patients presented a COVID-19 associated pneumonia with severe illness, as defined in the NIH COVID-19 guidelines by any of the following criteria: respiratory rate ≥ 30 breaths per minute, SpO2 < 94 %, PaO₂/FiO₂ ratio < 300 and involvement of ≥50 % of the lung parenchyma as evidenced by a reference chest CT-scan (Cascella et al., 2022). Exclusion criteria were represented by preexisting neurologic evolutive/degenerative disorders, pregnancy/nursing condition, mental illness or impossibility for the subject to have a good comprehension of the study, therapeutic limitation before inclusion and lack of health insurance coverage.

2.3. Pupillary variables assessment

Pupillary variables were measured with using the AlgiScan® video pupillometer (iDMed™, Marseille, France) that delivers a calibrated flashlight. The quantitative pupillometry test was performed on each eye and at time of admission in ICU (Day 1 or D1). The technique consisted of placing the tested pupil in the dark for 10 s, with contralateral eye closed, and then to record the baseline and post-flashlight pupillary variables over 3 s. The pupillary variables included the baseline and post-flashlight pupillary diameter (PD in mm), the PD variation (PDV in %), the pupillary constriction velocity (mm.s⁻¹) and the latency (ms).

2.4. Data collection

At time of inclusion (D1), we collected the demographic characteristics (i.e. age, gender, weight and height) and also main COVID-19 associated features (i.e. history of type 2 diabetes mellitus, hypertension, dyslipidemia, overweight or obesity, immunosuppression, neoplasia, ischemic heart disease, liver failure or chronic kidney disease). The presence of multilobed involvement on CT, pulmonary embolism, septic shock and hemorrhagic stroke was assessed during hospitalization as well as the need for mechanical ventilation. We also evaluated the critical illness severity with help of SAPS-II, SOFA score and the Glasgow coma scale (GCS). The use of epinephrine, sedative and opioids, since they can modify the pupillary reactivity, were collected. Patients were followed-up to hospital discharge. The onset and duration of invasive mechanical ventilation (IV), the time and cause of death and the length of stay in ICU and in hospital were measured.

This manuscript adheres to the applicable CONSORT guidelines.

2.5. Statistical analysis

Continuous data were expressed as mean and standard deviation (mean ± SD). The assumption of normality was assessed with the Shapiro-Wilk test. The comparisons between groups (death yes/no) were performed using the Student t-test or the Mann-Whitney test when assumptions required for the t-test were not met. The homoscedasticity was analyzed using the Fisher-Snedecor test. The relationships between continuous variables were explored using Pearson or Spearman correlation coefficient according to statistical distribution. Then, the multivariable analysis was carried out using generalized linear model (logistic for death as dichotomous endpoint) to take into account possible confounding factors. Hence, no specific statistical strategy approach, such as stepwise, was conducted. The covariates were chosen with caution due to sample size according to the univariate results and to clinical relevance (Mickey and Greenland, 1989). The testing and parameter estimation performed using a statistical model clearly depends on the variables included in the model. It is therefore crucial for confounding adjustment that known clinically significant variables are included in the regression model. A clinically significant variable may well be an important confounder also when it is statistically insignificant (Sun et al., 1996). Multivariable regression analyses were run with each parameter and then added together in multivariable model: age, gender, BMI, opioid drugs, SAPSII and SOFA score (Harrell et al., 1996). Furthermore, particular attention was paid to possible multicollinearity using Farrar-Glauber test. The statistical analyses were performed using Stata software version 15 (StataCorp®, College Station, US). Statistical tests were two-sided with the type-I error set at 5 %. As analyses were exploratory, the individual p-values have been reported without applying systematically mathematical correction but with a specific attention paid on the magnitude of differences and to the clinical relevance (van Rijn et al., 2017).

3. Results

3.1. Characteristics of study subjects and pupillary parameters on the day of admission (D1)

From April to august 2020, 50 patients were enrolled out of 65 patients admitted in ICU for a SARS-CoV-2 related pneumonia. The flow-chart of the study is presented on Fig. 1 .

Fig. 1.

Flow-chart of the study.

As one patient withdrew his consent, 49 patients were analyzed (Table 1 ). Patients were mostly male, hypertensive, overweighted and admitted in ICU after a median one week after COVID-19 onset. At ICU admission, patients had a moderate hypoxemia secondary to a multilobe pneumonia diagnosed on CT scan in 92 % of them. Forty-five (92 %) patients were treated with dexamethasone. Eighteen (37 %) and 28 (57 %) patients received opioids and midazolam, respectively. Thirty-five (71 %) received mechanical ventilation, among which 28 (57 %) were intubated for IV. Pupillary variables were assessed within the 12 h from ICU admission and presented in Table 1.

Table 1.

Demographic and clinical characteristics of included patients and values of pupillary parameters.

Characteristics	Number of patients N = 49 (100 %)
Male sex, n (%)	44 (90 %)
Age years, median (IQR)	66 (60–70)
BMI kg/m2, median (IQR)	27 (24–30)
Delay between the onset of symptoms and ICU admission, median (IQR)	8 (7–10)
Smoking history, n (%)	3 (6 %)
Overweight (25 < BMI < 30), n (%)	21 (43 %)
Obesity (BMI > 30), n (%)	14 (29 %)
Type 2 diabetes mellitus, n (%)	13 (27 %)
Hypertension, n (%)	29 (59 %)
Ischemic heart disease, n (%)	5 (10 %)
Chronic kidney disease, n (%)	2 (4 %)
Liver failure, n (%)	2 (4 %)
Multilobe involvement in CT-scan, n (%)	45 (92 %)
SOFA score	4 (3–6)
SAPSII score	40 (29–47)
Admission Glasgow Coma Scale (GCS)	14 (13–15)
Glasgow Coma Scale at time of pupillometry	14 (12–15)
D-Dimers, median (IGR)	1918 (991–2675)
Corticosteroids (dexamethasone), n (%)	45 (92 %)
Opioid drugs, n (%)	18 (37 %)
Vasopressors, n (%)	32 (65 %)
Epinephrine, n (%)	0 (0 %)
Midazolam, n (%)	28 (57 %)
Pa02/Fi02 ratio calculated the day of admission in ICU, median (IQR)	173 (104–154)
No Need for mechanical ventilation, n (%)	14 (29 %)
Need for ventilation support (invasive and/or non-invasive), n (%)a	35 (71 %)
- Mechanical invasive ventilation, n (%)	28 (57 %)
- Mechanical invasive ventilation before enrollment, n (%)	4 (8 %)
- Non-invasive ventilation, n (%)	12 (24 %)
Nasal High Flow Oxygen therapy, n (%)	9 (18 %)
Non-Invasive ventilation with facial mask, n (%)	4 (8 %)
Days on mechanical ventilation (invasive and/or non-invasive), median (IQR)	22 (13–41)
Pulmonary embolism, n (%)	7 (14 %)
Septic shock, n (%)	3 (6 %)
Death during the stay in ICU, n (%)	7 (14 %)
Pupillary parameters measured the day of ICU admission	Mean ± SD
PD (mm)	4.1 ± 1.2
PDV (%)	33.8 ± 7.6
PCV (mm.s−1)	3.7 ± 1.0
Latency (ms)	273.1 ± 36.6

BMI: Body Mass Index; Multilobe involvement: involvement of >2 pulmonary lobes on the CT-scan realized the day of ICU admission; GCS: Glasgow Coma Scale; ICU: Intensive Care Unit; IV: Invasive Ventilation; NIV: Non-Invasive Ventilation (Nasal High Flow Oxygenotherapy or non-invasive ventilation).

^aRegarding the use of ventilatory support, some patients had, throughout the hospital-stay, both mechanical ventilation techniques successively (non-invasive then invasive).

3.2. Pupillary variables and in-hospital mortality

Seven (14 %) patients died in hospital, all during their stay in ICU. Univariate analysis revealed that, in comparison to patients who will survive (Table 2 ), the SAPS-II score (51 [44; 59] vs 38 [29; 45], p = 0.03), SOFA score (11 [6; 11] vs 4 [2; 6], p = 0.002) and GCS score at time of pupillometry (14 [9; 15] vs 4 [3; 11], p = 0.012) were significantly greater in patients who died. The PD (2.6 mm [2.4; 4.0] vs 4.1 mm [3.5; 4.8], p = 0.009), PDV (33 % [27; 39] vs 25 % [15; 36], P = 0.03) and PCV (3.5 mm.s⁻¹ [2.8; 4.4] vs 2.0 mm.s⁻¹ [1.9; 3.8], P = 0.02) were significantly lower in dead patients in univariate analysis. The PD latency (298 ms [237; 322] vs 274 ms [250; 301], p = 0.68) was not significantly associated with in-hospital mortality.

Table 2.

Relationships between severity scores, pupillary variables and in-hospital mortality and between need for intubation for mechanical ventilation and pupillary parameters measured at the day of admission (D1) (univariate analysis).

Parameters	Alive n = 42 (100 %)	Death n = 7 (100 %)	P
Gender (male)	37 (88 %)	7 (100 %)	1.00
Age	66 [58; 70]	70 [65; 78]	0.05
BMI	27.5 [24.2; 30.5]	26.4 [23.9; 28.0]	0.61
Diabetes mellitus	12 (29 %)	1 (14 %)	0.66
SOFA	4 [2; 6]	11 [6; 11]	0.002*
SAPS-II	38 [29; 45]	51 [44; 59]	0.03*
GCS at ICU admission	14 [14; 15]	14 [6; 15]	0.26
GCS at time of pupillometry	14 [9; 15]	4 [3; 11]	0.012*
Opioids (y/n)	13 (31 %)	5 (71 %)	0.08
PD (mm)	4.1 [3.5; 4.8]	2.6 [2.4; 4.0]	0.009*
PDV (%)	33 [27; 39]	25 [15; 36]	0.03*
PCV (mm.s−1)	3.5 [2.8; 4.4]	2.0 [1.9; 3.8]	0.02*
Latency (ms)	274 [250; 301]	298 [237; 322]	0.68

Pupillary parameters	Need for intubation n = 24	No need for intubation n = 24	P
PD (mm)	3.5 [2.8; 4.4]	4.2 [3.9; 5.2]	0.03*
PDV (%)	27.5 [24.5; 36]	35.0 [31.5; 39.5]	0.049*
PCV (mm.s−1)	3.1 [2.4; 4.0]	3.9 [3.0; 4.3]	0.17

The asterisk represents the values of p < 0.05 and therefore a statistically significant difference between the 2 groups (p-values < 0.05).

Data are expressed in mean ± SD of the different pupillary parameters unless otherwise indicated. P-value for multivariate analysis. GCS: Glasgow Coma Scale; PD: Pupillary diameter, PDV: Pupillary Diameter Variation, PCV: Pupillary Constriction Velocity.

We then constructed the ROC curves corresponding to each parameter (Fig. 2 panel A, B, C and D and Table 3 ). The baseline PD has the highest AUC, sensitivity, and specificity (Table 3 and Fig. 2). A PDV <26.5 % and PCV <2.5 mm.s⁻¹ were statistically associated with in-hospital mortality but a PD threshold value of 2.75 mm was found to be the most discriminative and was reported in 5/7 (71 %) who will die and 3/42 (5 %) who will survive.

Fig. 2.

Receiver operating characteristic (ROC) curve for pupillary parameters recorded on the day of ICU admission and in-hospital mortality: panel A for PD, panel B for PDV, panel C for PCV and panel D for latency, respectively. PD: Pupillary Diameter; PDV: Pupillary Diameter Variations; PCV: Pupillary constriction Velocity; AUC: air under the curve.

Table 3.

Thresholds of the tested pupillary variables and their predictive P-values for in-hospital mortality.

Parameters	AUC [IC 95 %]	TS	Sensitivity	Specificity	PPV	NPV
PD (mm)	0.81 [0.63; 0.99]	2.75	72 [29; 96]	93 [80; 99]	63 [25; 92]	95 [83; 99]
PDV (%)	0.76 [0.55; 0.98]	26.5	72 [29; 96]	78 [62; 89]	36 [13; 65]	94 [80; 99]
PCV (mm.s−1)	0.78 [0.57; 1.00]	2.5	71 [29; 96]	85 [71; 94]	46 [17; 77]	95 [82; 99]
Latency (ms)	0.55 [0.26; 0.83]	300	43 [10; 82]	71 [55; 84]	20 [4; 48]	88 [72; 97]

Data are expressed in number (N) and percentages (%) unless otherwise indicated. P-value <0.05. TS: Threshold Value, PPV: Positive Predictive Value, NPV: Negative Predictive Value, AUC: Area Under Curve, PD: Pupillary diameter, PDV: Pupillary Diameter Variation, PCV: Pupillary Constriction Velocity.

We performed a multivariable analysis presented on the Fig. 3 (Supplemental Digit Content: Forest-plot representation). We showed that a PD < 2.75 mm remained associated with in-hospital mortality after adjustment to the SAPS-II and SOFA scores, use of opioids or vasopressors, BMI, age and diabetes.

Fig. 3.

Forest-plot representation of the association between pupillary diameter (PD), treated as a continuous variable, and in-hospital mortality, adjusted on demographic characteristics, opioids, SAPS-II score, SOFA score and GCS score (univariate analysis). Subtype-specific odds-ratios (95 % CI) are denoted by black boxes (black lines). GCS: Glasgow Coma Scale; GCSa: GCS at ICU admission; GCSp: GCS at time of pupillometry.

3.3. Pupillary parameters and prediction of need for mechanical ventilation

Twenty-four (48 %) patients required invasive mechanical ventilation after their ICU-admission. Baseline PD (3.5 mm [2.8; 4.4] vs 4.2 mm [3.9; 5.2], P = 0.03) and PDV (27.5 % [24.5; 36] vs 35.0 % [31.5; 39.5], P = 0.049) were significantly lower in patients who will be subsequently intubated than in those who will not during their ICU stay (Table 3).

4. Discussion

Our study shows that assessment of pupillary diameter was useful for predicting mortality and need for mechanical ventilation in COVID-19 patients admitted in ICU, independently of critical illness severity score (i.e. SAPS-II and SOFA), use of opioids or sedation. We found that a decrease in baseline PD was the most associated with bad outcome, indicating an impairment of sympathetic/parasympathetic control balance. A summary of these observations is represented on the Supplemental Fig. 1.

The discrimination value and reliability of the clinical signs and the scores for assessing dysautonomia are limited in critically ill patient are limited. In contrast, pupillometry is simple and validated methods for investigating the ANS dysfunction of the ANS in ICU patients (Erdal et al., 2022; Lai et al., 2020; Weise et al., 2022). The iris muscle is indeed innervated by the sympathetic and parasympathetic systems, which induce pupillary dilatation and contraction, respectively. The decreases in baseline PD and in pupillary reactivity to light (i.e. in PDV and Pupillary constriction velocity PCV) rather support a sympathetic impairment. The latter could be then a component of a broader central autonomic dysfunction (Loewenfeld and Thompson, 1967; Lowenstein and Loewenfeld, 1950). Indeed, severe COVID-19 can be associated, in the acute phase, with cardiovascular dysautonomia, impaired respiratory response (UR and Verma, 2020) and thermoregulation but also gastrointestinal transit and sexual disorders (Baig et al., 2020). These disorders, including pupillary abnormalities, have more recently been described as sequelae manifestations in patients with “long-COVID” (Buoite Stella et al., 2022; Dani et al., 2021). The main mechanisms would be a viral invasion or a non-viral neuro-inflammation of the brainstem autonomic centers. In addition to dysautonomia, there are neurological and neurophysiological arguments for a brainstem dysfunction (Bocci et al., 2021). Thus, abolition of brainstem reflexes was associated with mortality in patients requiring deep sedation for a severe SARS-CoV-2 pneumonia (Bocci et al., 2021). More recently, Karahan et al. showed the interest of pupillometry as a non-invasive tool to assess the ANS and for the long-term follow-up of patients after SARS-CoV-2 infection (Karahan et al., 2021). An impaired blink reflex, involving ponto-medullary circuits and therefore the brainstem, seems to be associated with respiratory failure in COVID-19 patients (Bocci et al., 2021). Several pharmacological agents can modify the PD or PMR (Larson and Behrends, 2015; Larson and Talke, 2001; Larson and Berry, 2000; Larson et al., 1993). In our study, neither the opioids nor sedatives interact on the relationships between baseline PD and outcomes.

The different parameters of the PMR had already shown their interest as prognostic tool in patients with severe brain injury (Braakman et al., 1980; Choi et al., 1988; Tamura et al., 2018) and in comatose patients after cardiorespiratory arrest (Heimburger et al., 2016; Oddo et al., 2018; Solari et al., 2017; Tamura et al., 2018). If most of the studies reports an improvement of prognostic performance using quantitative PLR or Neurological Pupil index (NPi) in comparison with qualitative assessment of PLR, the characteristics of each pupillary parameter or the relationship between parameters and in-hospital outcomes were never studied in those patients. Our study highlights the interest of simple measurements performed in a few seconds at the bedside to predict mortality of patients underwent a SARS-CoV-2 infection.

Our prospective observational study has several limitations. First, its small sample size has limited the identification of potential confounders. Secondly, the general ICU care and specific treatment of COVID-19 patients since the study period have dramatically changed. In addition, our study does not compare the parameters studied in COVID-19 patients with a control group of uninfected patients. Currently, only the team of Vrettou et al. studied ANS dysfunction by comparing pupillary parameters of patients with SARS-CoV-2 infection with a control group (Vrettou et al., 2020). They did not find any significant difference between the 2 groups. However, these parameters were collected from patients already intubated for >48 h and their association with morbidity and mortality was not part of the objectives. Finally, we are not able to confirm a structural insult of the brainstem, as neither a brain MRI nor an autopsy have never been performed in our patients.

For instance, criteria for invasive mechanical ventilation are established nowadays while they were debated at time of our study. These limits prompt us not to draw a definitive conclusion on prognosis value of pupillometry in COVID-19 patients. Furthermore, the low values of PD and PDV and the presence of overlapping confidence intervals for these 2 parameters in our study also limit, for the time being, the use of such measures to help in the decision of goals of care in these patients.

Finally, our study was not designed with long-term follow-up, making it impossible to collect pupillary parameters far from the acute phase. However, our results would be useful for designing a validation study that would require a standardization of care and well-defined outcomes. Second, ICU physicians were not blinded to the results of quantitative pupillary responses. However, the pupillometry has never been considered in care of COVID-19 patients, minimizing then the risk of fulfilling self-prophecy. Therefore, only a large cohort study would enable to confirm their predictive value but also to develop a new algorithm that could be tested against NPi.

5. Conclusion

In conclusion, our observational study suggests that a reduced baseline PD is associated with in-hospital mortality and with the need for invasive mechanical ventilation in COVID-19 patients admitted in ICU. A reduced PD is likely to result from a central autonomic impairment and support the existence of a brainstem dysfunction in severe COVID-19.

The following is the supplementary data related to this article.

Supplemental Fig. 1

Graphic representation summarizing the various pupillary parameters analyzed and their values found significantly associated with in hospital mortality (multivariate analysis). PD: Pupillary Diameter; PDV: Pupillary Diameter Variation; SOFA: Sepsis-related Organ Failure Assessment.

CRediT authorship contribution statement

DC and MD: study design, data acquisition, quality assessment, data interpretation, statistical analysis and manuscript drafting. BP: statistical analysis, manuscript drafting and quality assessment. MP: data acquisition and quality assessment. SM, TS: study design, data interpretation and manuscript drafting. All authors provided critical reviews of the manuscript and approved the final version. MD is the guarantor of the content of this manuscript.

Funding sources

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Declaration of competing interest

None.

Data availability

Data will be made available on request.