The novel coronavirus disease 2019 (COVID-19) has overwhelmed ICUs across the United States with large volumes of critically ill patients with severe hypoxemia and acute respiratory distress syndrome (ARDS). As of December 2020, the United States has recorded over 14 million cases of COVID-19, with early data suggesting an ICU admission rate as high as 11.5% (1,2). Limited therapies have been identified to improve mortality in patients with COVID-19 related respiratory failure (3). Prone positioning ventilation is a well-studied therapeutic intervention in ARDS, with early implementation leading to improved mortality (4), but it has not been studied extensively in patients with COVID-19.
A joint statement by the American Thoracic Society, European Society of Intensive Care Medicine, and Society of Critical Care Medicine in 2017 recommended prone positioning ventilation in adult patients with severe ARDS (5). Despite these consensus recommendations, proning remains underused in intubated patients with moderate-to-severe hypoxemia (6,7). Potential reasons for this underutilization include concerns regarding hemodynamic instability (8), as well as limitations in hospital resources, staff comfort, and education (9,10). These limitations become more pronounced with the surge of patients seen in the COVID-19 pandemic.
Despite the lack of strong evidence as a COVID-19-specific therapeutic intervention, proning has been proposed to be a key component in the management of COVID-19 ARDS (11). Recent studies examining the role of proning in mechanically ventilated patients with COVID-19 demonstrated improvements in oxygenation (12). However, studies examining the effect of proning on mortality in this population are lacking. Given the suggestion that respiratory mechanics differ in COVID-19–associated respiratory failure, there is a lack of consensus on whether similar physiologic and outcome improvements may be observed with proning in this population (13,14). A perceived lack of equipoise in light of the positive results from the Effect of Prone Positioning on Mortality in Patients With Severe and Persistent Acute Respiratory Distress Syndrome (PROSEVA) trial also reduces the feasibility of performing a randomized controlled trial of proning in mechanically ventilated patients with COVID-19 (4).
When data from randomized trials are unavailable, a target trial emulation approach, which applies principles of trial design to analyze observational data, can be used to guide practice recommendations (15–17). We used data from a multicenter study of critically ill patients with COVID-19 to emulate a target trial of early proning versus no early proning to examine its effect on survival.
METHODS
Overview
We used data from the Study of the Treatment and Outcomes in Critically Ill Patients with COVID-19 (STOP-COVID) to emulate a hypothetical target trial in which patients were assigned to receive or not receive prone positioning ventilation in the first 2 days of ICU admission (18).
Study Population
STOP-COVID is a multicenter cohort study of adults with laboratory-confirmed COVID-19 admitted to ICUs at 68 participating U.S. sites (18). Study teams at each institution performed standardized data abstraction and entered data into a secure online platform (Research Electronic Data Capture), collecting data on demographics, symptoms, comorbidities, home medications, vital signs, laboratory values, medications, nonmedication treatments, and organ support. For the current analysis, we included patients admitted to an ICU between March 4, 2020, and May 15, 2020, at 66 of the 68 sites in which proning was used for least one STOP-COVID patient during this timeframe (Supplemental Table 1, https://links.lww.com/CCM/G204). We followed patients until the first of hospital discharge, death, or June 22, 2020 (the date of last follow-up). Patients were not followed after hospital discharge. Detailed laboratory and physiologic measurements, including daily assessments of oxygenation (ratio of Pao2 over the corresponding Fio2 [Pao2/Fio2]), were collected for the first 14 days of ICU admission. Day of proning initiation, along with other critical care interventions, medications, and organ support, were also recorded for the first 14 days of ICU admission. Duration of proning therapy was not collected. Additional details related to the design of STOP-COVID are reported elsewhere (18). The study was approved with a waiver of informed consent by the Institutional Review Board (IRB) of each of the participating sites (protocol number 2007P000003 for the Mass General Brigham IRB).
Eligibility Criteria
We included adult patients (≥ 18 yr old) who had moderate-to-severe hypoxemia (Berlin criteria: Pao2/Fio2 ratio ≤ 200 mm Hg) (19) within the first 2 days of ICU admission while receiving invasive mechanical ventilation. This clinically relevant, wider range of hypoxemia was selected, as compared to the Pao2/Fio2 less than or equal to 150 mm Hg used in the PROSEVA trial (4), since a sizable proportion of intubated patients in the STOP-COVID cohort were initiated on early proning with Pao2/Fio2 ratios greater than 150 mm Hg. Using a threshold of less than or equal to 200 mm Hg captured 82.9% of patients who were initiated on prone positioning ventilation within the first 2 days of ICU admission.
We excluded patients who did not have at least one Pao2/Fio2 ratio less than or equal to 200 mm Hg in the first 2 days of ICU admission, irrespective of whether they were proned. We also excluded patients if they received extracorporeal membrane oxygenation on ICU day 1, experienced cardiac arrest or severe arrhythmia (sustained ventricular tachycardia or ventricular fibrillation) on ICU day 1, were proned prior to ICU admission, or were pregnant (4).
Treatment Strategies
We compared proning initiation versus no initiation in the 2 days following ICU admission. A 2-day period was chosen to provide greater homogeneity between patients and allow longer follow-up. Although the PROSEVA trial limited patient inclusion to the first 36 hours of ICU admission (4), event capture in STOP-COVID including ICU admission and early proning initiation was limited to days, not hours.
Treatment Classification
We classified patients according to whether proning was initiated within the first 2 days of ICU admission. Patients initiated on proning later in their ICU course were included in the nonproning group to reduce the potential for immortal time bias and to emulate an intention-to-treat strategy of a randomized trial. We assumed comparability conditional on the following covariates which were prespecified based on clinical judgment: age; gender; race (White vs non-White); body mass index; premorbid conditions (lung disease, smoking status, coronary artery disease, congestive heart failure, active malignancy); symptom duration prior to ICU admission; severity of illness on ICU admission (Pao2/Fio2 ratio, the renal, liver, and coagulation components of the Sequential Organ Failure Assessment score (20), shock, WBC count, lymphocyte count, inflammation [elevated levels of C-reactive protein or ferritin], and d-dimer level); and therapies administered on ICU day 1 (corticosteroids, tocilizumab, therapeutic anticoagulation, and neuromuscular blockade). Additional covariates included the number of pre-COVID ICU beds and the regional (county) density of COVID-19. Additional details are provided in the Supplemental Digital Content (https://links.lww.com/CCM/G204).
Outcomes
The primary outcome was time to in-hospital death, censored at hospital discharge or last follow-up.
Statistical Analysis
We compared the survival among patients who initiated or did not initiate proning in the first 2 days of ICU admission by estimating Kaplan-Meier survival curves and hazard ratios (HRs) using a Cox model. We used inverse probability (IP) weighting to adjust for confounding in both the estimation of the survival curves and the HRs, with calculation of standardized differences between pre- and post-IP weights (21,22). To do so, we fit a logistic regression model with early proning initiation as the outcome conditional on the variables described above. We used the model’s predicted probabilities to calculate stabilized IP weights (23), which we then applied to weight each individual’s contribution to the survival curves and to the Cox model (24,25). We used a robust (sandwich) variance estimator to account for potential replications of patients induced by IP weighting, which results in conservative (wider) 95% CIs.
We conducted two prespecified and four post hoc sensitivity analyses. First, we included each of the covariates used for the IP weighted model in an unweighted Cox model. Second, to reduce the potential for immortal time bias (26–28), we classified patients as being initiated on proning or not on ICU day 1, and we repeated the process for the remaining eligible patients on ICU day 2. We obtained our final estimates by pooling the data from the nested target trial emulations on ICU days 1 and 2, using IP weighting as described above. Third, as an alternative approach to reduce the potential for immortal time bias, we excluded patients who died in the first 2 days of ICU admission. Fourth, we repeated the primary analysis while further adjusting for the percentage of patients proned within the first 2 days of ICU admission at each site. Fifth, we repeated the primary analysis but censored patients on day 30. Sixth, we repeated the primary analysis, censored patients on day 30, and assumed that patients discharged alive before day 30 remained alive for the full 30 days.
We completed two subgroup analyses in patients receiving invasive mechanical ventilation who had a Pao2/Fio2 ratio less than or equal to 150 (4) and less than or equal to 100 mm Hg in the first 2 days of ICU admission. Further details are provided in the Supplemental Digital Content (https://links.lww.com/CCM/G204). All analyses were conducted using SAS 9.4 software (SAS Institute, Cary, NC).
RESULTS
Of the 4,853 patients considered for inclusion, 2,338 were eligible and were included in the analysis (Fig. 1). A total of 702 patients (30.0%) were initiated on proning within the first 2 days of ICU admission, with 407 (58.0%) initiated on day 1 and 295 (42.0%) initiated on day 2. An additional 457 patients (19.5%) were initiated on proning later in their ICU course (Supplemental Fig. 1, https://links.lww.com/CCM/G204).
;)
Figure 1.: Study cohort. ABG = arterial blood gas, COVID-19 = coronavirus disease 2019, Pao 2/Fio 2 = ratio of Pao 2 over the corresponding Fio 2.
Selected characteristics of patients before and after applying IP weighting are shown in Table 1 (additional characteristics are provided in Supplemental Table 2, https://links.lww.com/CCM/G204). Prior to applying the weighting, proned patients were younger and had a lower prevalence of comorbidities compared with nonproned patients. Proned patients had similar renal and liver components of the SOFA score, as well as similar WBC count, lymphocyte count, and d-dimer levels on ICU day 1, compared with nonproned patients (Supplemental Table 2, https://links.lww.com/CCM/G204). However, proned patients had a higher occurrence of shock on ICU day 1 compared with nonproned patients (114 [26.2%] vs 208 [12.7%]). Corticosteroids and neuromuscular blockade were used on ICU day 1 more often in proned versus nonproned patients (149 [21.2%] vs 217 [13.3%]; 176 [25.1%] vs 218 [13.3%], respectively). Proned patients were more likely to be admitted to hospitals with fewer ICU beds and those with a higher regional density of COVID-19 than nonproned patients.
TABLE 1. -
Selected Baseline Characteristics Before and After Applying Inverse Probability Weighting
|
Pre-IP Weighting |
Post-IP Weighting |
| Covariates |
Proned Early (n = 702) |
Not Proned Early (n = 1,636) |
Proned Early |
Not Proned Early |
| Demographic characteristics |
|
|
|
|
| Age, median (IQR) |
60 (51–69) |
63 (53–72) |
61.0 (52.0–70.0) |
62.0 (53.0–71.0) |
| Male sex, n (%) |
474 (67.5) |
1,053 (64.4) |
65.6 |
65.6 |
| White race, n (%) |
271 (38.6) |
620 (37.9) |
36.5 |
37.7 |
| Body mass index (kg/m2), median (IQR) |
31.5 (27.4–37.2) |
30.6 (26.7–35.9) |
31.1 (27.3–36.5) |
30.8 (26.9–36.2) |
| Coexisting conditions, n (%) |
|
|
|
| Coronary artery disease |
73 (10.4) |
227 (13.9) |
12.2 |
12.7 |
| Congestive heart failure |
42 (6.0) |
163 (10.0) |
7.8 |
8.7 |
| Any lung disease |
135 (19.2) |
367 (22.4) |
20.7 |
21.3 |
| Symptom onset to ICU admission ≤ 7 d, n (%) |
330 (47.0) |
924 (56.5) |
46.6 |
46.4 |
| Severity of illness on ICU admission, n (%) |
|
|
| Pao
2/Fio
2, mm Hga, n (%) |
|
|
|
| Ventilated and Pao
2/Fio
2 151–200 |
94 (13.4) |
330 (20.2) |
16.6 |
18.0 |
| Ventilated and Pao
2/Fio
2 100–150 |
208 (29.6) |
584 (35.7) |
33.4 |
33.7 |
| Ventilated and Pao
2/Fio
2 ≤ 100 |
400 (57.0) |
722 (44.1) |
50.1 |
48.3 |
| Shock (%), n (%) |
114 (16.2) |
208 (12.7) |
14.5 |
14.0 |
| WBC count (per mm3), n (%) |
|
|
| < 4,000 |
31 (4.4) |
77 (4.7) |
5.2 |
4.8 |
| 4,000–11,900 |
448 (63.8) |
1,087 (66.4) |
65.4 |
65.7 |
| ≥ 12,000 |
193 (27.5) |
416 (25.4) |
25.5 |
25.9 |
| Inflammation presentb, n (%) |
551 (78.5) |
1,186 (72.5) |
74.3 |
74.3 |
| Therapies administered, n (%) |
|
|
|
| Corticosteroids |
149 (21.2) |
217 (13.3) |
15.6 |
15.5 |
| Therapeutic anticoagulation |
107 (15.2) |
265 (16.2) |
15.6 |
15.8 |
| Neuromuscular blockade |
176 (25.0) |
218 (13.3) |
17.4 |
17.0 |
| Tocilizumab |
54 (7.7) |
94 (5.7) |
6.6 |
6.4 |
| Hospital characteristics |
|
|
|
|
| ICU bed size, n (%) |
|
|
|
|
| < 60 |
372 (53.0) |
605 (37.0) |
43.1 |
42.1 |
| 60–119 |
199 (28.3) |
686(41.9) |
36.4 |
37.7 |
| ≥ 120 |
131 (18.7) |
345 (21.1) |
20.5 |
20.2 |
| Regional density of coronavirus disease 2019, quartilec, n (%) |
|
|
| 1 |
41 (5.8) |
147 (9.0) |
8.5 |
8.1 |
| 2 |
141 (20.1) |
308 (18.8) |
20.1 |
19.4 |
| 3 |
150 (21.4) |
447 (27.3) |
24.7 |
25.4 |
| 4 |
370 (52.7) |
734 (44.9) |
46.7 |
47.1 |
IP = inverse probability, IQR = interquartile range, Pao2/Fio2 = ratio of Pao2 over the corresponding Fio2.
aPao2/Fio2 was only assessed in patients receiving invasive mechanical ventilation.
bInflamed was defined as at least one of the following on ICU days 1 or 2: C-reactive protein > 100 mg/L or ferritin > 1,000 ng/mL. Noninflamed was defined as at least one value below the thresholds above. The thresholds above were selected based on prior studies (
29–31).
cRegional density of coronavirus disease 2019 (COVID-19) was assessed by categorizing hospitals into quartiles according to the regional (county) density of COVID-19 cases present on the median date of ICU admission for the patients who were contributed by that hospital.
Data regarding body mass index were missing for 23 proned patients (3.3%) and 73 nonproned patients (4.5%). Data regarding WBC count were missing for 30 proned patients (4.3%) and 56 nonproned patients (3.4%). Data regarding inflammation were missing for 70 proned patients (10.0%) and 232 nonproned patients (14.2%).
The standardized differences between groups for each of the 25 covariates pre and post weighting are shown in Figure 2. The standardized differences in the weighted sample were below 0.1 for all 25 covariates, indicating well-balanced groups post weighting.
Figure 2.: Absolute standardized differences before and after applying inverse probability weighting. This figure shows the absolute standardized differences between groups for each of the 25 prespecified covariates pre and post weighting. COVID-19 = coronavirus disease 2019, Pao 2/Fio 2 = ratio of Pao 2 over the corresponding Fio 2, SOFA = Sequential Organ Failure Assessment.
Among the 2,338 patients included in this analysis, the median follow-up for the proned and nonproned patients was 34 days (IQR, 25–46 d) and 30 days (IQR, 20–42 d), respectively, and was 31 days (IQR, 22–43 d) overall. A total of 1,017 patients (43.5%) were discharged alive, 1,101 (47.1%) died, and 220 (9.4%) remained hospitalized at last follow-up. The 1,101 patients who died included 327 of the 702 patients (46.6%) treated with early proning and 774 of the 1,636 patients (47.3%) not treated with early proning (unadjusted HR, 0.89 [95% CI, 0.79–1.02]).
In the primary analysis, the mortality HR was 0.84 (95% CI, 0.73–0.97) for early proning initiation versus no early proning initiation (Fig. 3). In the first two sensitivity analyses, the HR was 0.81 (95% CI, 0.70–0.93) when using an unweighted Cox model with all covariates (Supplemental Table 3, https://links.lww.com/CCM/G204) and 0.85 (95% CI, 0.74–0.98) when using the nested target trial approach with IP weighting. The additional sensitivity analyses also showed comparable results (Fig. 3B). The prespecified secondary analyses using IP weighted models and eligibility according to Pao2/Fio2 ratio less than or equal to 150 and less than or equal to 100 mm Hg produced similar results to the primary analysis, with HR 0.86 (95% CI, 0.74–0.995) and 0.80 (95% CI, 0.67–0.96), respectively (Fig. 3B).
Figure 3.: Survival of patients initiated versus not initiated on early prone positioning ventilation within the first 2 d of ICU admission. A, Shows Kaplan-Meier survival curves for those who underwent prone positioning ventilation within first 2 ICU days versus those who did not. B, Shows the hazard ratios (HRs) for survival for proned versus nonproned patients. Sensitivity analysis # 1 used an unweighted multivariable Cox model rather than inverse probability weighting to adjust for confounding. Sensitivity analysis # 2 used a nested target trial approach to reduce the potential for immortal time bias. Sensitivity analysis # 3 excluded patients who died in the first 2 d of ICU admission as an alternative approach to reduce the potential for immortal time bias. Sensitivity analysis # 4 is similar to the primary analysis but is further adjusted for the percentage of patients at each hospital who were proned in the first 2 d of ICU admission. Sensitivity analysis #5 is similar to the primary analysis but censored patients on day 30. Sensitivity analysis #6 is similar to the sensitivity analysis #5 but assumes that individuals discharged alive before day 30 remained alive for the full 30 days. Also shown are the analyses of prespecified subgroups with the ratio of Pao 2 over the corresponding Fio 2 (Pao 2/Fio 2) less than or equal to 150 and less than or equal to 100 mm Hg.
DISCUSSION
Using a nationally representative cohort of critically ill mechanically ventilated patients with COVID-19 and moderate-to-severe hypoxemia, we found that the risk of death was lower in patients initiated on early prone positioning ventilation within the first 2 days of ICU admission compared with patients who were not. Our findings are consistent with and expand upon the findings from smaller observational studies in mechanically ventilated patients with COVID-19, which found improved oxygenation in proned patients (12). Furthermore, our findings are consistent with the survival benefit of early proning in non-COVID-19 ARDS patients with severe hypoxemia that was first demonstrated in the PROSEVA trial (4), with our results also being inclusive of those with more moderate hypoxemia.
Patients in our study initiated on proning in the first 2 days of ICU admission were similar to those in previous studies examining proning effectiveness. However, they displayed higher rates of shock and corticosteroid use (4), expected with the clinical severity seen with COVID-19 infection and the recommended treatment guidelines (32). In our study, proned patients were also younger and had fewer comorbidities compared with those not initiated on proning in the first 2 days, suggesting that these clinical factors may have influenced the decision to use proning. Hospitals with fewer ICU beds and those with higher regional density of COVID-19 used proning more often, raising the possibility that resource limitations may play a more nuanced role in the decision to prone than that previously considered (6). Hospitals with smaller ICUs may prioritize proning initiation, especially if access to more costly or intensive interventions (e.g., extracorporeal membrane oxygenation), are unavailable. Hospitals with higher local case volume may have established proning teams as part of surge response. Additional study is needed to better characterize the reasons for interhospital variation in proning.
More strikingly, in our cohort of moderate-to-severely hypoxemic mechanically ventilated patients, 30% underwent proning within the first 2 days of ICU admission. Compared with prior studies examining proning utilization, this frequency was relatively high. In the Large observational study to UNderstand the Global impact of Severe Acute respiratory FailurE (LUNG SAFE) study, only 16.3% and 5.5% of severe and moderate ARDS patients, respectively, underwent proning (7). This suggests that during the COVID-19 pandemic, ICUs may have been more open to proning, perhaps reflecting a lack of alternative evidence-based therapeutic options to treat patients with COVID-19.
Despite the relatively high rates of proning compared with past studies, absolute proning utilization rates remained low, even with prepandemic recommendations supporting its use in acute respiratory failure and general consensus statements endorsing prone positioning ventilation as a recommended practice in COVID-19 (4,11). Reasonable motives for deferring proning such as hemodynamic compromise and spinal instability exist but are unlikely to fully account for such low utilization rates. During the COVID-19 surge, many hospitals experienced space and staffing constraints which may have contributed to these treatment decisions. Proning is perceived as a labor-intensive intervention; with limited personal protective equipment supply and desire to minimize staff exposure, this intervention may have been less used (10). Increased ICU strain has been associated with decreased adherence to other protocolized interventions such as venous thromboembolism prophylaxis (33,34). Similarly, high patient volume may have contributed to low proning utilization during the COVID-19 pandemic. As more institutions have adopted proning as standard of care, many protocols and checklists have been published to assist ICUs with its implementation (35). A better understanding of the true risks of prone positioning may alleviate fears of complications (36). Fortunately, later studies have shown increased utilization of proning (37); examination of the operational differences between hospital cohorts may shed light on how to better implement standard therapies in the pandemic.
Our study has several strengths. First, we used observational data to explicitly emulate a target trial. Eligibility criteria were selected to be comparable with randomized trials, and treatment initiation was limited to only the first 2 days of ICU admission for invasively mechanically ventilated patients with moderate-to-severe hypoxemia, in line with evidence from non-COVID-19 settings suggesting proning benefit if initiated in the first 36 hours of mechanical ventilation (4). We adjusted for multiple potential confounders, including hospital-level covariates, and used methods that reduce the potential of immortal time bias. Additionally, the STOP-COVID registry used comprehensive and standardized manual chart abstraction for a large number of geographically varied sites across the United States. Finally, our target trial emulation results were consistent across multiple sensitivity analyses that used alternative methodologic approaches, along with alternative thresholds of hypoxemia to define eligibility. As randomized controlled trials of proning in COVID-19 are unlikely to be feasible given a perceived lack of equipoise, target trial emulation may offer the best available evidence on which to base current practice.
We also acknowledge several limitations, including the potential for unmeasured confounding such as that related to missing data on Pao2/Fio2 ratio. Although we did not have mortality data post hospital discharge, we did have complete in-patient follow-up for a minimum of 28 days for the entire cohort. We did not collect data on the duration of proning, the number of proning sessions, changes in goals of care, or other contraindications to proning, which may have played a role in the decision not to initiate early proning for certain patients. We also did not collect data on tidal volume settings or plateau pressures, limiting our ability to draw conclusion about the benefit of proning in patients with differing lung compliance. Although unavailable for this study, we recognize that practice differences, expertise, and resources across hospitals contribute to hospital-level variation in proning delivery and patient outcomes. Data on individual sites’ selection criteria for proning initiation or the presence/absence of standard proning protocols were also not available, and we did not have access to site-level strain metrics, although we did account for regional density of COVID-19 cases as an imperfect surrogate for patient volume.
CONCLUSIONS
Our findings suggest that early prone positioning ventilation initiated within the first 2 days of ICU admission may increase survival in mechanically ventilated patients with moderate-to-severe hypoxemia due to COVID-19–associated respiratory failure. Further study is needed to identify subgroups of patients with COVID-19 associated respiratory failure who might benefit from more protocolized proning, as well as general study on the operationalization and implementation of proning during periods of resource limitation and capacity strain.
REFERENCES
1. Centers for Disease Control and Prevention. Coronavirus Disease 2019 (COVID-19), Cases, Data & Surveillance, 2020. Available at:
https://www.cdc.gov/coronavirus/2019-ncov/cases-updates/cases-in-us.html. Accessed December 6, 2020
2. CDC COVID-19 Response TeamSevere outcomes among patients with coronavirus disease 2019 (COVID-19) — United States, February 12–March 16, 2020. Morb Mortal Wkly Rep. 2020; 69:343–346
3. Lim WS, Emberson JR, Mafham M, et al.; The RECOVERY Collaborative GroupDexamethasone in hospitalized patients with COVID-19 — preliminary report. N Engl J Med. 2020 Jul 17. [online ahead of print]
4. Guérin C, Reignier J, Richard JC, et al.; PROSEVA Study GroupProne positioning in severe acute respiratory distress syndrome. N Engl J Med. 2013; 368:2159–2168
5. Fan E, Del Sorbo L, Goligher EC, et al.; American Thoracic Society, European Society of Intensive Care Medicine, and Society of Critical Care MedicineAn official American thoracic society/European society of intensive care medicine/society of critical care medicine clinical practice guideline: Mechanical ventilation in adult patients with acute respiratory distress syndrome. Am J Respir Crit Care Med. 2017; 195:1253–1263
6. Guérin C, Beuret P, Constantin JM, et al.; investigators of the APRONET Study Group, the REVA Network, the Réseau recherche de la Société Française d’Anesthésie-Réanimation (SFAR-recherche) and the ESICM Trials GroupA prospective international observational prevalence study on prone positioning of ARDS patients: The APRONET (ARDS prone position network) study. Intensive Care Med. 2018; 44:22–37
7. Bellani G, Laffey JG, Pham T, et al.; LUNG SAFE Investigators; ESICM Trials GroupEpidemiology, patterns of care, and mortality for patients with acute respiratory distress syndrome in intensive care units in 50 countries. JAMA. 2016; 315:788–800
8. Hepokoski ML, Odish M, Malhotra A. Prone positioning in acute respiratory distress syndrome: Why aren’t we using it more? J Thorac Dis. 2018; 10:S1020–S1024
9. Chertoff J. Why is prone positioning so unpopular? J Intensive Care. 2016; 4:70
10. Poor AD, Acquah SO, Wells CM, et al. Implementing automated prone ventilation for acute respiratory distress syndrome via simulation-based training. Am J Crit Care. 2020; 29:e52–e59
11. Poston JT, Patel BK, Davis AM. Management of critically ill adults with COVID-19. JAMA. 2020; 323:1839–1841
12. Ziehr DR, Alladina J, Petri CR, et al. Respiratory pathophysiology of mechanically ventilated patients with COVID-19: A cohort study. Am J Respir Crit Care Med. 2020; 201:1560–1564
13. Gattinoni L, Coppola S, Cressoni M, et al. COVID-19 does not lead to a “typical” acute respiratory distress syndrome. Am J Respir Crit Care Med. 2020; 201:1299–1300
14. Gattinoni L, Chiumello D, Rossi S. COVID-19 pneumonia: ARDS or not? Crit Care. 2020; 24:154
15. Dickerman BA, García-Albéniz X, Logan RW, et al. Avoidable flaws in observational analyses: An application to statins and cancer. Nat Med. 2019; 25:1601–1606
16. Petito LC, García-Albéniz X, Logan RW, et al. Estimates of overall survival in patients with cancer receiving different treatment regimens: Emulating hypothetical target trials in the surveillance, epidemiology, and end results (SEER)-medicare linked database. JAMA Netw Open. 2020; 3:e200452
17. Admon AJ, Donnelly JP, Casey JD, et al. Emulating a novel clinical trial using existing observational data. Predicting results of the PreVent study. Ann Am Thorac Soc. 2019; 16:998–1007
18. Gupta S, Hayek SS, Wang W, et al. Factors associated with death in critically ill patients with coronavirus disease 2019 in the US. JAMA Intern Med. 2020; 180:1–12
19. Ferguson ND, Fan E, Camporota L, et al. The Berlin definition of ARDS: An expanded rationale, justification, and supplementary material. Intensive Care Med. 2012; 38:1573–1582
20. Vincent JL, Moreno R, Takala J, et al. The SOFA (sepsis-related organ failure assessment) score to describe organ dysfunction/failure. On behalf of the working group on sepsis-related problems of the European society of intensive care medicine. Intensive Care Med. 1996; 22:707–710
21. Austin PC. Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples. Stat Med. 2009; 28:3083–3107
22. Yang D, Dalton JE. A unified approach to measuring the effect size between two groups using SAS®. SAS Glob Forum. 2012335–2012
23. Xu S, Ross C, Raebel MA, et al. Use of stabilized inverse propensity scores as weights to directly estimate relative risk and its confidence intervals. Value Health. 2010; 13:273–277
24. Robins JM, Hernán MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology. 2000; 11:550–560
25. Hernán MA, Brumback B, Robins JM. Marginal structural models to estimate the joint causal effect of nonrandomized treatments. J Am Stat Assoc. 2001; 96:440–448
26. Hernán MA, Robins JM. Using big data to emulate a target trial when a randomized trial is not available. Am J Epidemiol. 2016; 183:758–764
27. Danaei G, García Rodríguez LA, Cantero OF, et al. Electronic medical records can be used to emulate target trials of sustained treatment strategies. J Clin Epidemiol. 2018; 96:12–22
28. Gupta S, Wang W, Hayek SS, et al. Association between early treatment with tocilizumab and mortality among critically ill patients with COVID-19. JAMA Intern Med. 2020; 181:41–51
29. Herold T, Jurinovic V, Arnreich C, et al. Elevated levels of IL-6 and CRP predict the need for mechanical ventilation in COVID-19. J Allergy Clin Immunol. 2020; 146:128–136.e4
30. Landry A, Docherty P, Ouellette S, et al. Causes and outcomes of markedly elevated C-reactive protein levels. Can Fam Physician. 2017; 63:e316–e323
31. Moore C Jr, Ormseth M, Fuchs H. Causes and significance of markedly elevated serum ferritin levels in an academic medical center. J Clin Rheumatol. 2013; 19:324–328
32. Sanders JM, Monogue ML, Jodlowski TZ, et al. Pharmacologic treatments for coronavirus disease 2019 (COVID-19): A review. JAMA. 2020; 323:1824–1836
33. Weissman GE, Gabler NB, Brown SE, et al. Intensive care unit capacity strain and adherence to prophylaxis guidelines. J Crit Care. 2015; 30:1303–1309
34. Tarnow-Mordi WO, Hau C, Warden A, et al. Hospital mortality in relation to staff workload: A 4-year study in an adult intensive-care unit. Lancet. 2000; 356:185–189
35. Messerole E, Peine P, Wittkopp S, et al. The pragmatics of prone positioning. Am J Respir Crit Care Med. 2002; 165:1359–1363
36. Munshi L, Del Sorbo L, Adhikari NKJ, et al. Prone position for acute respiratory distress syndrome. A systematic review and meta-analysis. Ann Am Thorac Soc. 2017; 14:S280–S288
37. COVID-ICU Group on behalf of the REVA Network and the COVID-ICU InvestigatorsClinical characteristics and day-90 outcomes of 4244 critically ill adults with COVID-19: A prospective cohort study. Intensive Care Med. 2020; 49:1–14
