This was a prospective observational cohort study conducted from December 2016 to March 2018 at the Department of Adult Intensive Care Medicine, Centre Hospitalier Universitaire Vaudois (CHUV), University Hospital and University of Lausanne, Switzerland. Subjects were medical-surgical critically ill patients requiring sedation and mechanical ventilation for at least 48 hours, at high risk (about 50%) of ICU delirium . Exclusion criteria were mechanical ventilation for ≤ 48 hours, age <18 years, acute brain injury (including traumatic brain injury, ischemic/hemorrhagic stroke, subarachnoid hemorrhage, hypoxic-ischemic brain injury after cardiac arrest, meningo-encephalitis, status epilepticus, hepatic encephalopathy, neurosurgical intervention), previous known cognitive impairment, end-stage renal or hepatic disease (Child-Pugh B and C cirrhosis), pregnancy, ICU readmission or transfer from another ICU. Additional exclusion criteria included a pre-existing ophthalmic condition or disease that may alter pupillary response, including cataract surgery, multiple sclerosis, amyloidosis, sclerodermia and multiple system atrophy. Patients who were expected to die within 72 hours were also excluded. A convenience sample size was used (n=100). The study was approved by the ethical committee of the Lausanne University (project-ID 2016-01923), and a waiver of consent was granted because non-invasive pupillometry is standard care. The study conforms with the STROBE guidelines for the report of observational studies.
Management of analgesia and sedation
Management of analgesia and sedation was based on a written institutional algorithm, in line with current recommendations . Sedation was targeted to Richmond Sedation Agitation Scale (RASS)  with continuous infusions of propofol (2-3 mg/kg/h) and/or midazolam. Propofol was generally first-line agent; midazolam was used in patients with hemodynamic instability (defined as norepinephrine >0.25 μg/kg/min), or when the propofol dose exceeded 4 mg/kg/h or was maintained for more than 48 hours. Analgesia was maintained with fentanyl (1-1.5 μg/kg/h).
Automated infrared pupillometry
An automated infrared pupillometer (AlgiScan®, ID-Med, Marseille, France) was used for repeated measurements of quantitative PLR (q-PLR, expressed as the percentage change of pupillary diameter following the light stimulus) and constriction velocity (CV, measuring the speed of pupil constriction following light stimulation, expressed in mm/sec). Normative values for the q-PLR range between 30% and 40%, and for the CV between 1.5 and 2.2 mm/sec; low values are defined as a q-PLR <20% and a CV <1 mm/sec, respectively .
Pupillary measurements were conducted on both eyes by an experienced clinician or nurse, who was not involved in patient care, and were performed during the day in stable conditions of ambient light. Measurements were performed twice daily (in the morning and the afternoon), starting at day 3 from mechanical ventilation, and were repeated at day 4 and 5, up to a maximum of day 7. At each time-point, the average values of q-PLR and CV from both eyes were retained for the analysis. All pupillometry variables were blinded to clinicians and nurses involved in patient care.
As soon as the Richmond Agitation-Sedation Score was equal or greater than -2, delirium was assessed twice daily until discharge from the ICU using the CAM-ICU. Patients were considered as having delirium when they had at least one positive CAM-ICU during their ICU stay. The duration of coma was calculated as the number of days spent with a motor Glasgow Coma Scale < 6 from ICU admission.
Data processing and statistical analysis
Demographic and clinical variables included age, gender, medical versus surgical ICU admission, admission APACHE II score, daily SOFA score, sepsis diagnosis (according to the Sepsis-3 definition ), cumulative dose of analgesia (fentanyl) and sedatives (midazolam and propofol) during the first 7 days of ICU, duration of mechanical ventilation, length of ICU stay, and 90-day mortality.
Data are presented as median and interquartile range (IQR), except when otherwise stated. Univariate comparisons between the two main outcome groups (delirium vs. no-delirium) were analyzed using the non-parametric Wilcoxon test. A multivariable stepwise logistic regression model was performed by entering the day 3 q-PLR as the variable of interest, and the day 3 SOFA score, the cumulative fentanyl dose and the cumulative dose of sedatives as pre-specified co-variates. Statistical significance was set at p < 0.05. Statistical analysis was conducted with R 3.5.1 and JMP-14®. The statistical analysis was performed by an independent statistician (JP).