Authors: Peter J. Dunbar (1Division of Pulmonary Sciences and Critical Care Medicine, University of Colorado, Aurora, Colorado), Ryan A. Peterson (3Department of Biostatistics and Informatics, Colorado School of Public Health, University of Colorado, Aurora, Colorado), Max McGrath (3Department of Biostatistics and Informatics, Colorado School of Public Health, University of Colorado, Aurora, Colorado), Tyree H. Kiser (4Department of Clinical Pharmacy, Skaggs School of Pharmacy and Pharmaceutical Sciences, University of Colorado, Aurora, Colorado), P. Michael Ho (2Division of Cardiology, Department of Medicine, School of Medicine, University of Colorado, Aurora, Colorado), R. William Vandivier (1Division of Pulmonary Sciences and Critical Care Medicine, University of Colorado, Aurora, Colorado), Ellen L. Burnham (1Division of Pulmonary Sciences and Critical Care Medicine, University of Colorado, Aurora, Colorado), Marc Moss (1Division of Pulmonary Sciences and Critical Care Medicine, University of Colorado, Aurora, Colorado), Peter D. Sottile (1Division of Pulmonary Sciences and Critical Care Medicine, University of Colorado, Aurora, Colorado)
Categories: Article, sedatives, NMB, ARDS, mechanical ventilation
Source: Annals of the American Thoracic Society
Authors: Peter J. Dunbar, Ryan A. Peterson, Max McGrath, Tyree H. Kiser, P. Michael Ho, R. William Vandivier, Ellen L. Burnham, Marc Moss, Peter D. Sottile
Neuromuscular blockade (NMB) is frequently used during acute respiratory distress syndrome (ARDS) to improve ventilator synchrony. Which sedating medications are used concomitantly during NMB and whether sedation choice influences patient outcomes are unclear.
To determine national sedation practice patterns during NMB in patients with and at risk for ARDS and to establish whether the use of propofol and opioids compared with benzodiazepines and opioids is associated with improved outcomes.
Using a U.S. national database from 2010 to 2021, intubated and mechanically ventilated patients receiving NMB for a diagnosis of ARDS or an ARDS risk factor over at least two hospital days after admission were included. Charges for sedation and analgesia during the first two hospital days were recorded for each patient. The relationships between propofol and opioids and between benzodiazepines and opioids, with a primary outcome of ventilator-free days, as well as secondary outcomes of 28-day survival and discharge home were examined in multivariable analyses.
We determined that the use of propofol has increased compared with that of benzodiazepines as the primary sedative used during NMB for ARDS. Compared with benzodiazepine and opioid use, propofol and opioid use during NMB for ARDS was associated with increased ventilator-free days (adjusted odds ratio, 1.38 [95% confidence interval, 1.24–1.54]), greater odds for survival at 28 days (adjusted odds ratio, 1.15 [95% confidence interval, 1.01–1.31]), and greater odds for discharge home (adjusted odds ratio, 1.26 [95% confidence interval, 1.09–1.46]), adjusting for patient-level and hospital-level characteristics.
From 2010 to 2021, sedation practice during NMB for ARDS shifted from predominately benzodiazepine use to predominately propofol use. The use of propofol and opioids is associated with an increase in ventilator-free days compared with the use of benzodiazepines and opioids. These results suggest that sedation choice during NMB for ARDS may affect clinical outcomes; further investigation is needed to validate these findings.
The use of neuromuscular blockade (NMB) for acute respiratory distress syndrome (ARDS) is common, with 37.8% of patients with severe ARDS receiving NMB in a recent, international prospective cohort study (1). The mechanism of potential benefit is believed to involve a decrease in ventilator-induced lung injury via improved patient–ventilator synchrony as well as a reduction in oxygen consumption (2, 3). To ensure patient comfort and prevent postparalysis recall, deep sedation during NMB is considered standard practice (4, 5). However, receipt of deep sedation has been linked with poor patient outcomes, including increased duration of mechanical ventilation and delirium (6, 7). It remains unclear whether medication choices during deep sedation for NMB can affect these outcomes.
Among nonparalyzed, intubated, and mechanically ventilated patients, sedation practices have changed over time (6). In particular, nonbenzodiazepine sedatives have been prioritized because of the association of benzodiazepines with patient harm (8–13). However, it is uncertain whether these trends in sedation practices and the impact of sedation choice on patient outcomes extend to patients undergoing deep sedation for NMB.
To address these questions, we conducted an observational study among patients with ARDS and ARDS risk factors (e.g., hypoxemia, pneumonia, sepsis, trauma) who received NMB to determine 1) patterns of sedation exposure in the United States between 2010 and 2021 and 2) the association between the two most common sedation strategies and clinical outcomes including ventilator-free days (VFDs), 28-day survival, and discharge home. We hypothesized that the use of benzodiazepines was common and associated with fewer VFD.
Some of these data have been reported in abstract form (14).
We performed an observational cohort study using the Premier Healthcare Database, an all-payer database including approximately 20% of U.S. discharges (15). Data collected on encounters occurring between January 2010 and December 2021 were used. The Colorado Multiple Institution Review Board approved this study (21–4056).
Mechanically ventilated patients with ARDS or ARDS risk factors who received at least two days of NMB after admission were included (see Table E1 in the data supplement). International Classification of Diseases, Ninth Revision, and International Classification of Diseases, Tenth Revision (ICD-10), codes were used to define ARDS and ARDS risk factors (16, 17). Risk factors included acute hypoxemic respiratory failure, pneumonia, sepsis, and trauma. In addition, patients included must have received NMB on admission (minimum, on Hospital Days 1 and 2) to minimize confounding by worsening illness worsening during hospitalization. We chose to evaluate patients who had received cisatracurium or vecuronium as NMB because they are most commonly used (3, 4, 16).
Patients with specific indications for benzodiazepines (e.g., alcohol withdrawal, seizure), patients transferred from outside hospitals, patients <18 years of age, and patients without charges for exposure medications on consecutive Hospital Days 1 and 2 were excluded (see Table E1).
We defined exposure to medications by the presence of at least one charge code for a given medication on each of the first two days of hospitalization, concurrent with NMB charges, as previously performed (18, 19). Medications were benzodiazepines (midazolam and lorazepam), propofol, opioids (fentanyl, morphine, and hydromorphone), dexmedetomidine, and ketamine (see Table E2).
The descriptive outcomes included 1) the percentage of patients exposed to each study medication by year of data collected and 2) the percentage of patients exposed to distinct combinations of medications.
Patients exposed to propofol and opioids or benzodiazepines and opioids were compared in subsequent analyses because these combinations were most common in our descriptive findings.
The primary comparative outcome was VFDs, calculated as days out of 28 without invasive mechanical ventilation charges. Patients expiring within 28 days were assigned zero VFDs. Secondary outcomes included 28-day survival and discharge home (see Table E3).
Patient-level covariates were age, sex, race, ethnicity, payer, admission time, comorbid conditions, and exposure to antipsychotic medications (see Table E4) (20). In addition, receipt of surgery, dialysis, or vasoactive medications on the first two hospital days were included to further account for severity of illness. Hospital-level covariates were size, region, rural versus urban, and teaching status.
Variables were summarized via descriptive statistics and stratified by exposure, testing for differences using Wilcoxon rank sum tests for numeric variables and chi-square tests for categorical variables. For each year (2010–2021), rates of the use of opioids, propofol, benzodiazepines, ketamine, and dexmedetomidine with exact 95% confidence intervals (CIs) were computed. Change in rates over time was tested using the Fisher exact test.
The primary outcome of VFDs was modeled as an ordinal variable for the two sedation groups, propofol and opioids and benzodiazepines and opioid, using cumulative-logit link proportional-odds logistic regression models, incorporating listed covariates, producing adjusted odds ratios (aORs) for sedation exposure. We included a natural cubic basis spline for year of patient admission, with three internal knots placed via equidistant quantiles (see Figure E1). For hospitals with 20 or more patients, we included hospital-specific fixed effects (convergence issues precluded the inclusion of effects for hospitals with fewer observations). As a sensitivity analysis, we also fit a mixed-effects model with a hospital-level random effect, which required binning the VFD outcome into coarser ordinal groups for computation. Interaction models were developed to discern whether the effect of medications on VFDs was modified by dates of admission, including 2016–2021 (the ICD-10 era) and years 2020–2021 (the coronavirus disease [COVID-19] pandemic). These were performed with and without time splines given concerns for collinearity. For secondary outcomes, we used logistic generalized estimating equations accounting for within-hospital correlation with an exchangeable working correlation matrix and robust standard errors, with otherwise the same covariates described above. Finally, to further mitigate potential bias by confounding variables, especially indication bias related to receipt of sedation, we used doubly robust estimation of causal effects to estimate an average treatment effect (ATE). To do so, propensity scores for receipt of propofol and opioid were estimated using a logit link logistic-effects model, and the outcome (VFD) was modeled using a linear mixed-effects model to produce an ATE. Both models incorporated listed covariates and a hospital-level random effect. We compared weighted and unweighted standardized mean differences by treatment to ensure an adequate balance was achieved.
Of more than 113 million hospital admissions, 1.9 million admitted patients received mechanical ventilation for at least two days, of whom 1.4 million had an admitting diagnosis of ARDS or ARDS risk factors (Figure 1). Among this group, 18,698 had charges for NMB on Hospital Days 1 and 2. Excluded patients comprised transfers from an outside hospital (n = 4,030), those with inadequate medication information (n = 640), those with admission diagnoses that would favor benzodiazepine use (n = 531), and those age younger than 18 years (n = 873). Ultimately, there were 12,624 patients in the study population, of whom 1,288 had an admitting diagnosis of ARDS. In addition, patients with ARDS risk factors included 5,740 with acute hypoxemic respiratory failure, 4,561 with pneumonia, 2,898 with sepsis, and 256 with trauma.
Between 2010 and 2021, the use of opioids (64.3% [95% CI, 59.8–68.6%] in 2010 vs. 77.7% [95% CI, 75.5–79.9%] in 2021) (P < 0.001) and propofol (42.3% [95% CI, 37.8–46.9%] in 2010 vs. 74.1% [95% CI, 71.7–76.3%] in 2021) (P < 0.001) increased (Figure 2 and see Table E5). However, the proportion of patients receiving benzodiazepines decreased (61.9% [95% CI, 57.4–66.3%] in 2010 vs. 42.2% [95% CI, 39.6–44.9%] in 2021) (P < 0.001). The use of dexmedetomidine and ketamine was comparatively low but increased significantly across the study period.
Across all years of study, the most commonly prescribed drug combinations were propofol and opioids (28.2%) or benzodiazepines and opioids (22.0%) (see Table E6). A three-drug combination of benzodiazepines, propofol, and opioids was recorded in 9.8% of patients. Charges for propofol alone were observed in 18.7% of patients. Fentanyl was the most common opioid (used in 95% of patients receiving opioids), while midazolam was the most common benzodiazepine (92% of patients receiving benzodiazepines). Further medication information can be found in the data supplement (see Table E7).
Further analyses focused on the 6,325 patients receiving propofol and opioids (n = 3,554) and benzodiazepines and opioids (n = 2,771) sedation strategies. One additional patient was excluded from this cohort before analysis given unidentified sex. Patients in these two subgroups were similar in age, sex, race, and payer mix (Table 1). Patients receiving propofol and opioids were more likely to receive diagnoses of COVID-19 (13.8% vs. 7.5%; P < 0.001). Patients receiving benzodiazepines and opioids were more likely to receive care in a teaching hospital (66.8% vs. 55.9%; P < 0.001) and to be admitted to hospitals with >500 beds (50.4% vs. 38.8%; P < 0.001). Patients receiving benzodiazepines and opioids also had higher rates of cardiac organ dysfunction (62.4% vs. 56.7%; P < 0.001), and they were more likely to receive renal replacement therapy (6.1% vs. 2.9%; P < 0.001) or to undergo surgery (24.9% vs. 22.3%; P < 0.013).
In unadjusted analysis, patients receiving propofol and opioids had increased VFDs compared with patients receiving benzodiazepines and opioids (15 VFDs [interquartile range, 0–13] vs. 21 VFDs [interquartile range, 0–21]; P< 0.001) (see Table E8 and Figure E2). Twenty-eight-day survival was higher in the propofol and opioids group compared with the benzodiazepines and opioids group (65.3% vs. 62.6%; P=0.03) (see Table E8). The percentage of patients discharged home was also higher in patients receiving propofol and opioids (22.3% vs. 20.3%; P=0.057) (see Table E8).
In adjusted analysis, accounting for age, sex, race, ethnicity, insurance status, severity of illness indicators, hospital characteristics, and time, exposure to propofol and opioids remained associated with increased odds of higher VFD (aOR, 1.38 [95% CI, 1.24–1.54]; P < 0.001) compared with benzodiazepine and opioid exposure (Figure 3 and see Table E9). Exposure to propofol and opioids was also associated with differentially increased odds of 28-day survival (aOR, 1.15 [95% CI, 1.01–1.31]; P =0.042) (Figure 3 and see Table E10). Finally, propofol and opioid exposure was associated with increased odds of discharge home compared with receipt of benzodiazepines and opioids (aOR, 1.26 [95% CI, 1.09–1.46]; P =0.002) (see Figure 3 and Table E11).
Time periods restricting the population to use of ICD-10 codes (2016–2021) and to the COVID-19 pandemic (2020–2021) were assessed for effect modification (with respect to sedatives and VFD). There was no evidence for interaction between sedation strategy (with a reference group of benzodiazepines and opioids) and the ICD-10 period (aOR, 1.05 [95% CI, 0.86–1.29]; P = 0.6) or the COVID-19 period (aOR, 0.98 [95% CI, 0.76–1.26]; P = 0.9) (see Tables E12 and E13). Interaction models estimated without time splines as sensitivity analysis also showed no evidence of effect modification (see Tables E14 and E15).
In a sensitivity analysis using mixed effects for hospitals and including hospitals with small numbers of observations, exposure to propofol and opioids was associated with increased VFDs (aOR, 1.41 [95% CI, 1.27–1.57]; P< 0.001) compared with exposure to benzodiazepines and opioids (see Table E16).
All patients were included in doubly robust estimation with propensity score weighting, which was able to sufficiently balance differences in characteristics between treatment groups (see Figure E3). There was a high degree of overlap in propensity score ranges (0.014–0.993 in the propofol and opioids group and 0.009–0.986 in the benzodiazepines and opioids group). Only 1.2% had propensity scores outside of the overlap between these ranges. In this model, exposure to propofol and opioids increased the expected VFD compared with exposure to benzodiazepines and opioids (ATE, 1.35 VFDs [95% CI, 0.88–1.82 VFDs]; P < 0.001).
We performed an observational cohort study in a sample of patients enrolled from across the United States to assess practice patterns and outcomes of sedation strategies among patients with ARDS receiving NMB. Using real-world data from over a decadelong period (2010–2021), we demonstrated that the use of benzodiazepines in this context decreased over time but remained common, while propofol use increased. This is study is the first, to our knowledge, to demonstrate an association between exposure to propofol and opioids with increased VFD compared with exposure to benzodiazepines and opioids among patients with ARDS or ARDS risk factors on NMB. This association was consistent through multivariable modeling, including available confounders as well as various sensitivity analyses.
We are not aware of previous studies describing sedation selection during NMB for ARDS. As such, sedation use in a previously published randomized controlled trial, in which deep sedation was mandated during NMB but sedation choice was not dictated, provides the best context for our descriptive findings. Papazian and colleagues performed a trial of NMB versus deep sedation in ARDS published in 2010 in which 89% of patients in the NMB arm received benzodiazepines and 38% received propofol (4). Medication use in this trial thus differs from our observational data, which showed less benzodiazepine use compared with propofol use. Apart from differences in methodology and geography (e.g., a randomized controlled trial in French hospitals vs. observational data in U.S. hospitals), trends over time may help resolve this apparent contradiction. Specifically, our data derive from the decade after Papazian and colleagues’ trial, and we found that benzodiazepine use was greater than propofol use in 2010 but substantially less by 2021. Comparison between published trial data and our findings thus underscores changing prescribing patterns over time, with increased use of propofol relative to benzodiazepines during NMB for ARDS.
Important outliers in benzodiazepine use in our trial are teaching hospitals and large hospitals (>500 beds), which were identified to use benzodiazepines more frequently than nonteaching and smaller hospitals. Although the drivers of this practice pattern are unclear, it may be reflective of sicker patient populations in these hospitals, as patients with indicators of more critical illness—those with cardiac or hepatic dysfunction and those requiring renal replacement therapy or surgery—were also more likely to receive benzodiazepines, perhaps because of concerns about the vasodilatory effects of propofol (21). Nevertheless, our analyses control for these indicators of critical illness and reveal evidence of an association with increased duration of mechanical ventilation and decreased survival through exposure to benzodiazepines and opioids compared with exposure to propofol and opioids.
Our study has numerous strengths. First, the Premier Healthcare Database permitted the interrogation of medication data and outcomes for a nationwide sample of thousands of patients across 12 years, a generalizable cross-section of patients. Second, we used database information to control for myriad confounders. Diagnosis codes and charge data permitted estimation of severity of illness, while hospital identifiers also permitted us to account for local variation in practice; furthermore, we accounted for nonlinear trends which may have otherwise confounded our results. Third, we used interaction models to provide reassurance that important historical events (the implementation of ICD-10 coding, the COVID-19 pandemic) did not modify the relationship between sedation exposure and our primary outcome. Fourth, we applied doubly robust estimation to further account for indication bias in treatment selection in addition to confounders of the primary outcome.
Our study also has limitations. First, the observational design results in nonrandom allocation of treatments and potential confounding by indication, particularly given that benzodiazepine and opioid exposure was more common in patients with certain indicators of critical illness. We attempted to minimize this bias by incorporating patient- and hospital-level covariates to account for residual sources of indication bias in multivariable modeling and sensitivity analyses. Assuming residual confounding is stable over time, the observed temporal trends in sedation patterns are less likely to be affected by these concerns.
Second, the use of a database limited available information for study. For instance, we did not have vital sign data or laboratory information to control conventionally for severity of illness, which raises the possibility of unmeasured confounders. Instead, we used admission diagnosis codes to identify organ failure predictive of in-hospital mortality in a manner which has been validated against commonly used scoring systems (20). We were unable to reliably identify further secondary outcomes of interest, such as delirium or paralysis recall rates. Furthermore, estimations of medication doses and the identification of continuous infusions were also limited by available charge data. We defined exposure to sedation by receipt in consecutive days as used in prior pharmacologic studies in this database (18, 19). However, it is possible that these charges reflect boluses rather than infusions of medications. This definition may alternatively be overly restrictive; we were surprised that some patients received propofol monotherapy. Given limited medication dosing data, we were unable to explore whether dose or depth of sedation mediated the associations we found.
Third, the use of diagnosis codes and billing charges to identify patients, exposures, and outcomes may have led to classification errors. In general, such random misclassification leads to underestimation of effect size, but there is evidence that using discharge status to determine mortality can lead to underclassification with unclear direction of bias (22).
Fourth, we lack information to map the patterns of medication use on to corresponding national drug shortages that may have affected prescribing trends, particularly during the COVID-19 pandemic.
Despite these limitations, this study offers important observations regarding use of sedation and analgesia during NMB for patients with ARDS risk factors. We demonstrate that rates of benzodiazepine use decreased over time, but benzodiazepine use remains common. This is despite evidence generated in our large, real-world dataset, which demonstrates that exposure to propofol and opioids is associated with increased VFDs relative to exposure to benzodiazepines and opioids. Further study is needed to determine optimal sedation strategies during NMB to achieve patient comfort while minimizing prolonged ventilation secondary to oversedation. In the meantime, given the addition of our evidence to a large body of research that demonstrates poor outcomes with benzodiazepines among the larger population of nonparalyzed, intubated patients, these agents should be used judiciously.
This article has a data supplement, which is accessible at the Supplements tab.