Association of Sedation, Coma, and In-Hospital Mortality in Mechanically Ventilated Patients With Coronavirus Disease 2019–Related Acute Respiratory Distress Syndrome

A Retrospective Cohort Study

Karuna Wongtangman, MD; Peter Santer, MD, DPhil; Luca J. Wachtendorf; Omid Azimaraghi, MD; Elias Baedorf Kassis, MD; Bijan Teja, MD, MBA; Kadhiresan R. Murugappan, MD; Shahla Siddiqui, MD; Matthias Eikermann, MD, PhD


Crit Care Med. 2021;49(9):1524-1534. 

In This Article

Materials and Methods

Study Design

This retrospective hospital registry study was approved by the Committee on Clinical Investigations at Beth Israel Deaconess Medical Center (BIDMC) in Boston, MA (protocol number 2020P000694) and met criteria for exemption from review. BIDMC is an academic teaching hospital of Harvard Medical School and a tertiary referral center covering intensive care of eight hospitals in the Beth Israel Lahey Health system. During the surge of COVID-19, the medical center doubled its ICU capacity to a total of 140 beds.

Data Sources

We combined multiple electronic data sources to obtain comprehensive information for all included patients. Demographic data, as well as a patients' past medical history, were collected from online medical records. Laboratory values and ICU data including level of consciousness, ventilator variables, and drug administration were extracted from Metavision, an electronic medical record interface routinely used in all ICUs. Encounter dates and discharge information were retrieved from Casemix and the Admission Discharge Transfer database. Radiographic imaging reports were obtained from the radiology database (supplementary document, section 1,

Study Population

Adult patients (≥ 18 yr) admitted to the ICUs were included if they were mechanically ventilated and were diagnosed with COVID-19 between March 2020 and May 2020, based on World Health Organization interim guidelines,[8] or were diagnosed with ARDS between January 2008 and June 2019. Patients with missing baseline characteristics for the propensity matching model were excluded.

Primary Outcomes and Analyses

We tested outcomes in an a priori defined, hierarchical order. The primary outcome was in-hospital mortality. The coprimary outcome was the percentage of comatose days.

The level of consciousness was routinely recorded at least every 4 hours using the Richmond Agitation Sedation Scale (RASS). Patients with a mean RASS score of −3 to −5 were classified as comatose based on previously published literature, irrespective of whether the state was induced by disease or sedation.[9] The daily sedation variables during the first 10 days of mechanical ventilation are provided in eTable 2 ( We defined the percentage of days spent in coma as the ratio of days with coma during the first 10 days of mechanical ventilation for each patient. The percentage of comatose days was dichotomized into low and high by using the median as the cut off value.

In the primary analysis, we tested the hypothesis that COVID-19 patients had a higher risk of in-hospital mortality compared with non–COVID-19 ARDS patients. Contingent on this assumption, we tested the coprimary hypothesis that the higher percentage of coma was a mediator of in-hospital mortality in COVID-19 patients.

Secondary Outcomes and Analyses

In secondary analyses, we investigated the causes that might affect coma. Sedative and analgesic medications used during mechanical ventilation, the Sedation Burden Index (SBI) during the first 10 days of mechanical ventilation, and structural brain lesions were compared between COVID-19 and non–COVID-19 patients. We tested the hypothesis that the hypnotic agent dose was associated with coma and mediated coma in COVID-19 patients. In addition, we tested the hypothesis that the association between sedation-related coma and mortality may be different than the association between neurologic injury-related coma and mortality.

We collected the daily cumulative doses of sedatives, analgesics, and neuromuscular blocking agents (NMBAs). Opioid doses were converted to oral morphine equivalents, and benzodiazepines were converted to midazolam equivalents for comparison.[10–12] For simplification and better comparison between an individual's exposure to sedative and analgesic medications, we calculated the SBI which was defined as the cumulative burden from every sedative or analgesic medication that patients received on each day during the 10-day mechanical ventilated period. The average SBI over 10 days was calculated for each patient. To calculate the hypnotic agent dose, we then identified the patient with the highest average SBI over 10 days and expressed individual values as the percent maximum value (supplementary document, section 2,

All CT scans of the brain and neurologic consultation results were reviewed and confirmed by an experienced intensivist.

Propensity Score Matching

One-to-two propensity score matching for patients with COVID-19 and without COVID-19 was performed as described detailed in the supplementary document, section 3 ([13,14] Propensity score estimates, calibrated for both the odd of the exposure (COVID-19–associated ARDS) and the outcome (in-hospital mortality) for each patient, were derived from clinical variables; demographic characteristics (age, sex, and body mass index), comorbidities (Charlson Comorbidity Index), disease severity (Acute Physiology and Chronic Health Evaluation II score), organ failure (renal impairment and severe liver injury), vasopressor support, and baseline ventilator variables at onset of mechanical ventilation (respiratory system compliance, PaO 2:FIO 2 [P/F] ratio, PaCO 2, and alveolar–arterial [A-a] gradient).[15–17] Covariates with residual imbalances following propensity score matching (standardized difference ≥ 0.1) were added as adjustment covariates to a model following matching.

Sensitivity and Exploratory Analyses

We conducted several sensitivity analyses to test the robustness of our findings. In the primary analysis, we matched for possible predictors of the exposure and outcome.[18] In order to address a potential selection bias, we repeated the analysis using the same covariate model in the complete, unselected cohort of all ARDS patients, which are described in detail with other sensitivity analyses in the supplementary document, section 4 (

With an exploratory intent, the hypnotic agent doses were compared in subgroups of patients who received NMBAs and prone positioning. We examined the trend in ARDS treatment by means of NMBA infusion, prone positioning, and sedative medications used during mechanical ventilation, compared between the COVID-19 pandemic period and the period before the COVID-19 pandemic. Delirium-free days were compared between the two groups.[9]

Statistical Analysis

Continuous variables and counts are described using mean ± SD or median (interquartile range [IQR]); categorical variables are reported as percentages.

Analyses were performed using chi-square test, multivariable logistic regression, negative binomial regression, and mediation analysis. For mediation analysis, we tested whether COVID-19 patients had a higher percentage of coma during the first 10 days of mechanical ventilation and whether a higher percentage of coma was associated with in-hospital mortality, indicating a possible effect mediation. Conditional on both associations being significant, we used formal mediation analysis to estimate odds ratios of the indirect (mediated) effect of high percentage of comatose state and the total (unmediated) effect of COVID-19 on in-hospital mortality.[19,20] A Cox proportional hazards regression analysis was used to compare the effects of coma on mortality of patients with sedation-related coma versus neurologic injury-related coma. We considered a two-tailed p value of below 0.05 as statistically significant. All analyses were performed using Stata, Version 15.1 (StataCorp LP, College Station, TX).

Power analysis

Based on previously published mortality in mechanically ventilated COVID-19 patients[21] and assuming a two-tailed alpha of 0.01 and a power greater than 95% to detect a significant effect, a total sample size of 190 patients was needed.