FormalPara Take-home message

All patients that develop ARDS have a high risk for mortality, even when the syndrome resolves quickly. Higher tidal volume is associated with ARDS persistence. Patients with severe ARDS on the second day constitute a group at substantially increased risk for mortality. In contrast, the predictive value of ARDS reclassification at day 2 for mortality was relatively limited.

Introduction

ARDS carries a high mortality, especially in observational studies, but with a relatively wide variability across studies [1,2,3]. The definition of ARDS has been repeatedly considered as one of the reasons for this heterogeneity. Because ARDS severity may change significantly in the first 24 h after fulfillment of the diagnostic criteria as the underlying mechanisms leading to lung injury progress or resolve, it has been proposed to reassess patients at 24 h (i.e., day 2) [4], in order to ‘confirm' fulfillment of the ARDS definition [5]. In addition, considerable optimization of management, particularly in regard to mechanical ventilatory support, may take place in the first 24 h, and impact on ARDS severity as captured by the definition.

The Berlin definition for ARDS [6] was designed primarily to make the diagnosis and has high sensitivity in this regard [7]. However, concerns exist regarding the specificity of the Berlin definition. In the LUNG SAFE study a significant portion of patients who initially fulfilled clinical criteria for ARDS did not do so 24 h later [1]. Some of these “resolved ARDS” patients may constitute false positives, transiently fulfilling clinical criteria for ARDS but not possessing the underlying pathophysiologic processes seen in ARDS, and consequently should have a much better outcome. In contrast, patients with confirmed ARDS, i.e., in whom ARDS criteria remain present 24 h later, may constitute a particularly high-risk patient cohort.

Reapplication of ARDS diagnostic criteria on day 2 following fulfillment of the ARDS definition may enhance our understanding of the impact of resolved versus confirmed ARDS, and predict important patient-centered outcomes from ARDS [4, 8, 9]. It may be possible to gain additional insights regarding the influence of management factors on the risk of ARDS persistence and on outcomes in these patients. Identification of ARDS subpopulations at highest risk for mortality might facilitate early consideration of these patients for adjunctive therapeutic strategies that have the potential for greater side effects (e.g., extracorporeal membrane oxygenation). Should reclassification at day 2 markedly improve outcome prognostication, this would have important implications for clinical trials, as another 24 h of confirmation may be necessary before enrolling patients into therapeutic trials.

We wished to address these issues in this secondary analysis of the LUNG SAFE cohort [1]. Our primary objective was to compare outcome in patients with resolved versus confirmed ARDS assessed on the day following ARDS onset. Secondary objectives included identifying factors associated with resolved and confirmed ARDS, determining the factors on the second day that predict hospital mortality, identifying subpopulations with high mortality risk, and determining the predictive value for patient outcome of reclassification of ARDS severity on the day following ARDS onset.

Methods and materials

The detailed methods and protocol have been published elsewhere [1]. In brief, LUNG SAFE was an international, multicenter, prospective cohort study, with a 4-week enrollment window in the winter season [1]. The study, funded by the European Society of Intensive Care Medicine (ESICM), was endorsed by multiple national societies/networks (Online Appendix 1). All participating ICUs obtained ethics committee approval, and either patient consent or ethics committee waiver of consent, as appropriate. National coordinators (Online Appendix 1) and site investigators (Online Appendix 2) were responsible for obtaining ethics committee approval and for ensuring data integrity and validity.

Patients, study design, and data collection

Inclusion criteria were admission to a study ICU (including ICU transfers) within the 4-week enrollment window; and receipt of invasive or noninvasive ventilation. Exclusion criteria were age less than 16 years or inability to obtain informed consent (where required). Patients were classified as having ARDS on the basis of fulfillment of all of the Berlin criteria rather than by clinician determination, as previously described [1]. This analysis was restricted to patients (93%) that fulfilled ARDS criteria within 48 h of the onset of acute hypoxemic respiratory failure (AHRF) and who received invasive mechanical ventilation (Fig. 1) at ARDS onset. All data were recorded for each patient at the same time each day within participating ICUs, normally as close as possible to 10 a.m. each day. Data on ventilatory settings were recorded simultaneously with arterial blood gas analysis.

Fig. 1
figure 1

Stratification and outcomes of patients classified for ARDS severity on day 1 and 2 of ARDS. Of 2377 patients that developed ARDS on day 1 or 2 following the development of AHRF, and were invasively ventilated, 263 (11.1%) were excluded from the analysis due to death (63 patients), or ICU discharge (16 patients) within 2 days, or because it was not possible to determine whether they still had ARDS (according to Berlin definition) on day 2 due to missing data (184 patients). The number of patients with ARDS evaluable for 2 days from onset was 2114, of which 1611 (76%) had confirmed ARDS and 503 (24%) had resolved ARDS at the second day. Patients are classified into three groups according to ARDS severity (mild, moderate, or severe) assessed at day 1, and then each group is stratified on the basis of their severity evaluated at day 2 (mild, moderate, severe, and resolved ARDS). In each group identified by day 1, the proportion of deaths varies among severity subgroups evaluated at day 2. Outcome is similar for each ARDS severity group on day 2, irrespective of their severity status on day 1

Full size image

Definitions

Our definitions have been previously reported [1, 10, 11]. Day 1 refers to the first day on which the patient fulfilled all of the Berlin criteria for ARDS, while day 2 is the day following ARDS onset. For the purposes of this analysis patients were considered to have “resolved ARDS” when they initially fulfilled the Berlin ARDS criteria but did not fulfill at least one criterion on day 2. Patients were considered to have “confirmed ARDS” when they continued to fulfill the Berlin definition when reassessed on day 2. Where chest radiography was not present at day 2, patients could only be considered to have confirmed ARDS if the other criteria were still present. Where data on positive end-expiratory pressure (PEEP) were missing at day 2, patients were considered to have confirmed ARDS if (a) the other criteria were fulfilled, and (b) there were data on the third day indicating ongoing assisted ventilation with a PEEP of 5 cmH2O or greater (Table e1, Supplementary Material). Patients could not be deemed to have resolved ARDS if any of the day 2 data was missing. If ARDS status could not be determined on the second day, despite the aforementioned assumptions, they were excluded from further analysis, as were patients that were not in the ICU on day 2 (i.e., were discharged or dead).

Duration of invasive mechanical ventilation was calculated as the number of days between the date of intubation and the date of extubation in ICU (or death, if the patient died under invasive mechanical ventilation). Survival was evaluated at hospital discharge, or at day 90, whichever occurred first.

Data management and statistical analyses

Descriptive statistics included proportions for categorical variables and mean (standard deviation) or median (interquartile range) for continuous variables. The study population was stratified according to the fulfillment of the Berlin definition on day 2 (resolved versus confirmed ARDS). Comparisons between groups were performed using Chi-squared test (or Fisher exact test) for discrete variables or Student’s t test (or Wilcoxon-Mann–Whitney test) for continuous variables. The Shapiro–Wilk test was used to assess normality in data distribution. Moreover, differences in discrete or continuous distributions of variables measured at day 1 and 2 were tested using McNemar’s test or Wilcoxon signed-rank test, respectively.

To evaluate factors associated with persistence of ARDS, we applied a multivariable logistic regression model and the independent predictors (demographic characteristics and clinical parameters measured at the first day of ARDS) were identified through stepwise regression approach. This approach combines forward and backward selection methods in an iterative procedure to select predictors in the final multivariable model. This approach was also applied to identify factors associated with hospital mortality in patients with resolved and with confirmed ARDS. Moreover, as a result of missing values in the clinical variables measured at the second day of ARDS (Table e2, Supplementary Material), we assessed the same model for hospital mortality, assuming a random distribution of missing data [12]. We applied a multiple imputation approach to handle missing information and the Markov chain Monte Carlo method was used to create 20 imputed datasets [13]. The estimates of each multivariable logistic model were combined using Rubin’s rules to estimate odds ratio (ORs) and corresponding 95% confidence intervals (CIs) [14].

To compare the predictive value for hospital mortality of day 2 and day 1 ARDS classification (Berlin definition), we estimated the area under the receiver operating curve (AUROC) using two different multivariable logistic regression models performed on the whole study population. The first model considered as independent variables all significant factors (demographic and clinical) measured at day 1 and identified by stepwise regression approach, plus ARDS severity at day 1 (mild, moderate, severe). The second model was similar, except for ARDS severity measured at day 2 (resolved, mild, moderate, and severe). The two AUROCs were estimated fitting the two models on the subset of patients without missing values in the covariates, in order to compare them with the non-parametric approach of DeLong, DeLong, and Clarke-Pearson [15].

We analyzed the survival time in patients with resolved and confirmed ARDS using the Kaplan–Meier approach with patients stratified according to ARDS severity at day 1 and/or day 2. When the survival curve involved ARDS severity at day 2, time and probabilities were calculated starting from the second day of ARDS. We assumed that patients discharged alive from hospital before 90 days were alive on day 90. Differences in survival curves were assessed by the log-rank test.

All P values were two-sided, with P values less than 0.05 considered as statistically significant.

Statistical analyses were performed with R, version 3.3.3. (R Project for Statistical Computing, http://www.R-project.org) and SAS software, version 9.4 (SAS Institute, Cary, NC, USA).

Results

Of 2377 patients that developed ARDS within 48 h of the occurrence of AHRF, and were invasively ventilated, we excluded from the main analysis 263 patients (11.1%) because of death (63 patients) or ICU discharge (16 patients) by day 2, or because it was not possible to determine whether they still had ARDS on day 2 according to the Berlin definition, as a result of missing data (184 patients) (Fig. 1 and Tables e1–2, Supplementary Material).

Evolution of ARDS over first 24 h

ARDS severity profile varied significantly from day 1 to day 2. Of 2114 patients fulfilling ARDS criteria on day 1 and included in this study, 503 (24%) patients had resolved ARDS, no longer fulfilling ARDS criteria on day 2, while 1611 (76%) patients had confirmed ARDS (Table 1 and Fig. 1). Among confirmed ARDS patients, 31% improved to a less severe ARDS category, 55% were unchanged, while 14% progressed to a more severe ARDS category (Table e3, Supplementary Material).

Table 1 Patients with resolved versus confirmed ARDS
Full size table

Resolved ARDS

Patients with resolved ARDS were less severely ill, with higher PaO2/FiO2 (P/F) ratio and lower SOFA score, compared to those with confirmed ARDS (Table 1); 52% of the resolved ARDS patients had moderate or severe ARDS. Resolved ARDS patients had a lower prevalence of pneumonia and a higher prevalence of non-cardiogenic shock (Table e4). Clinicians were less likely to recognize ARDS on day 1 in patients with resolved ARDS compared to patients with confirmed ARDS (24% vs 35%, P < 0.0001). Patients with resolved ARDS received significantly higher tidal volumes and lower PEEP on day 2 compared to day 1 (Fig. e1, Supplementary Material).

Mortality was 31% in these patients (Table 1). Factors associated with mortality in patients with resolved ARDS were higher age, higher respiratory SOFA score, higher respiratory rate, lower P/F ratio, lower PEEP, and higher PIP (Table 2).

Table 2 Factors associated with hospital mortality in study population
Full size table

Confirmed ARDS

Confirmed ARDS patients had higher day 1 driving pressures (mean value 14.9 vs. 13.3 cmH2O, P < 0.0001),and received higher PEEP (mean value 8.6 vs. 7.8 cmH2O, P = 0.0016) compared to patients with resolved ARDS (Table 1). Mechanical ventilation-related changes from day 1 to day 2 were modest, with no change in tidal volume (Fig. e1A), a small increase in PEEP (in moderate or severe ARDS) (Fig. e1B), and a decrease in peak inspiratory pressure (PIP) (mild or moderate ARDS) (Fig. e1C, Supplementary Material). Sixty-nine percent of patients with confirmed ARDS received protective mechanical ventilation on day 2 compared to 65% on day 1 (P = 0.41), defined as a tidal volume of 8 mL/kg of predicted body weight or less and a plateau pressure of 30 cmH2O or less (Fig. e2, Supplementary Material).

In the multivariable model, pneumonia was a risk factor for confirmed ARDS (Table 3). Other factors measured on day 1 significantly associated with confirmed ARDS were lower P/F ratio, higher PIP, higher non-respiratory SOFA score, and higher tidal volume.

Table 3 Factors associated with persistence of ARDS
Full size table

On day 2, tidal volume decreased modestly, while PEEP increased (mean values from 8.1 cmH2O in mild to 10.5 cmH2O in severe ARDS) with greater ARDS severity category (Table e3). Median inspired fractional oxygen use increased substantially, from 0.4 in mild, to 0.5 in moderate and 0.9 in severe ARDS (Table e3, Supplementary Material).

Hospital mortality of patients with confirmed ARDS was 41%: this increased from 36% with mild, to 39% with moderate and 57% with severe ARDS at day 2 (P < 0.0001) (Table 1 and Table e3, Supplementary Material). In multivariable analyses, age, the presence of immunosuppression and/or neoplasm and chronic liver failure were each associated with increased hospital mortality (Table 2). Illness severity factors on day 2 significantly associated with increased mortality were lower P/F ratio, lower pH, and higher non-respiratory SOFA score. In regard to ventilation-related factors, the use of higher respiratory rate and lower PEEP, and the presence of higher PIP were each associated with higher hospital mortality.

Reclassification of ARDS severity on day 2

Hospital mortality in patients with ARDS severity classified as mild on day 1 was 34%, compared to 36% in mild ARDS patients on day 2 (Fig. 1 and Table e3, Supplementary Material). Mortality rates were unchanged at 39% in patients classified as moderate ARDS severity on day 1 or 2. In contrast, in patients classified as severe ARDS, mortality increased from 43% in patients classified on day 1 to 57% in patients classified on day 2. Kaplan–Meier curves for outcome based on ARDS severity on day 1 versus day 2 confirm these findings, with mortality significantly higher for severe confirmed ARDS compared to all other severity categories on day 2 (Fig. 2a, b).

Fig. 2
figure 2

Hospital survival probability in patients with ARDS classified on day 1 and day 2 (Kaplan–Meier approach). Kaplan–Meier curves for survival during hospital stay in patients with ARDS stratified by ARDS severity on day 1 (a) or day 2 (b). In b, time and probabilities were calculated starting from the second day of ARDS. *P < 0.05 versus mild ARDS. †P < 0.05 versus resolved ARDS. ‡P < 0.05 versus moderate ARDS

Full size image

Kaplan–Meier curves for each category of ARDS severity on day 1 demonstrated a significantly different survival probability when stratified by ARDS severity on day 2 (Fig. e3A–C). In contrast, survival probability observed in patients with ARDS severity assessed at day 2 demonstrated similar probability when stratified by ARDS severity on day 1 (Fig. e3D–F, Supplementary Material).

Reclassification of ARDS severity on day 2 demonstrated a marginal (though statistically significant) increase in predictive value for mortality compared to day 1 (estimated difference between AUROCs was 0.0089, P = 0.045) (Fig. e4). Of interest, day 2 non-pulmonary SOFA demonstrated greater accuracy to predict mortality (AUROC 0.67, 95% CI 0.65–0.70) than P/F ratio (AUROC 0.57, 95% CI 0.55–0.60), while the combination of both variables did not improve AUROC (0.68, 95% CI 0.65–0.70) over SOFA score alone (Table e5, Supplementary Material).

Discussion

Almost one quarter of patients that fulfill ARDS criteria demonstrate resolution of one or more criteria 24 h later, while ARDS is confirmed or persists for at least 24 h in the other three quarters of patients. All patients with ARDS, whether resolved or confirmed at day 2, had a high mortality. Of concern, higher tidal volume on day 1 was independently associated with persistence of ARDS at day 2, while higher respiratory rate was associated with increased mortality. Reclassification of ARDS severity on day 2 identified patients with severe confirmed ARDS as a group with substantially increased mortality. In contrast, the predictive value of ARDS reclassification at day 2 for mortality was relatively limited.

Evolution of ARDS severity over first 24 h

Substantial changes were seen in the ARDS severity profile over the first 24 h, with ARDS resolving or decreasing in severity in over half of patients, while ARDS severity worsened in only 14%. ARDS resolved in 17% of patients classified as severe ARDS on day 1. Perhaps not surprisingly, greater day 1 severity of ARDS and higher non-respiratory SOFA score both increased the likelihood of ARDS persistence. Of interest, ARDS duration and the severity of hypoxemia have been identified in post-mortem studies as being predictive of the presence of diffuse alveolar damage [16]. Higher tidal volume use on day 1 was associated with ARDS persistence, supporting concerns raised regarding higher tidal volume use [17, 18], especially in early ARDS [19]. Given the extent of evolution of the clinical condition over the first 24 h, specifically the clinical improvement of most patients, selection of patients for clinical trials based on day 1 data—particularly where the intervention may have significant side effects (e.g., higher PEEP or extracorporeal membrane oxygenation)—may not be optimal.

Impact of ventilator management

The role of ventilator-induced lung injury in worsening outcomes from ARDS is clear from multiple studies [17, 18]. Our findings emphasize the need for continued focus on ventilator management to minimize iatrogenic injury and improve outcome in patients with ARDS. The use of higher tidal volume on day 1 of ARDS was associated with ARDS persistence. In patients with resolved and confirmed ARDS, the use of higher respiratory rate and lower PEEP, and the presence of higher PIP were each associated with higher hospital mortality.

Overall, we found relatively limited changes in ventilator management from day 1 to 2. There was no overall change in tidal volume in patients with confirmed (whether mild, moderate, or severe) ARDS from day 1 to 2. Patients with resolved ARDS actually received higher tidal volumes and lower PEEP on day 2 than on day 1. While potentially a reflection of an improvement in the patients’ condition, with increased spontaneous ventilation, this remains a concern [20].

Resolved versus confirmed ARDS

Patients with resolved ARDS had similar mortality to patients with mild confirmed ARDS, underlining the fact that this constitutes a patient group with ongoing critical illness. Older age, non-respiratory SOFA score, and day 1 ventilator parameters including respiratory rate, PEEP, and PIP were each associated with outcome in these patients. Our data provide novel insights into this poorly characterized patient group, and emphasize the requirement for careful attention to ventilator parameters and ongoing critical care to optimize clinical outcomes in these patients.

Consistent with previous studies, persistence of ARDS was associated with longer ICU and hospital stay, fewer ventilator-free days, and lower hospital survival [4, 8, 9]. The high mortality (57%) in patients with severe ARDS on day 2 suggests that these patients may warrant early consideration for alternative approaches, such as extracorporeal membrane oxygenation, to improve outcomes.

Predictive value of reapplication of ARDS criteria on day 2

Accurately predicting patient outcome, while not the primary role of diagnostic criteria such as the Berlin definition, is an important aspect of validity in the absence of a “gold standard” [6]. While the Berlin definition demonstrates somewhat better predictive value for mortality compared to the older American-European Consensus Conference definition [6], it remains low which is a potential limitation [7].

Reapplication of ARDS criteria on day 2 of ARDS provided greater discriminatory capacity for patient outcome, with greater separation in mortality rates by ARDS severity compared to classification on day 1. Of interest, patients with confirmed ARDS did not have a uniformly worse outcome, such as might be explained by the removal of patients with resolved ARDS from the analysis. Hospital mortality was substantially increased in patients with severe ARDS on day 2, suggesting that this approach identifies a higher-risk patient cohort.

In contrast, the overall increase in the predictive value for outcome of reclassification of ARDS severity on day 2, while statistically significant, was only marginal. This is perhaps not surprising, given all the other factors known to be important (including age, comorbid disease, other organ dysfunction, vasopressor requirement, etc.) in determining outcome from critical illnesses, including ARDS [11, 21, 22]. In fact, it is well recognized that many patients die with, rather than of, ARDS, i.e., they die from other complications of their critical illness such as vasopressor-resistant shock due to sepsis [23]. This is supported by our finding that the predictive value of the non-pulmonary SOFA score was slightly higher than the P/F ratio in the patient population with confirmed ARDS. Including all variables identified in the model further increased predictive value. This may have important implications for clinical trials since it has been claimed that waiting for “confirmed ARDS” to enroll patients results in a patient population with higher severity and/or greater homogeneity.

Limitations

This study has several limitations. Our patient cohort, while large and geographically diverse, may not be representative of actual clinical practice in ICUs across the globe. We did not have access to the source data for the patients in the enrolling ICUs, and it is possible that not all patients with ARDS in participating centers were enrolled. However, enrollment of patients with ARDS from participating ICUs met expectations based on their recorded 2013 admission rates, while data from lower-recruiting ICUs was not different from that from higher-enrolling ICUs, suggesting the absence of reporting biases.

We instituted a robust data quality control program in which all centers were requested to verify data that appeared inconsistent or erroneous. Despite our data quality control, missing data meant that we had to apply two assumptions regarding chest radiography and PEEP settings to classify one-fifth of patients. Missing radiologic data for day 2 likely reflects the fact that routine daily chest radiographs are no longer justified [24, 25]. We were unable to classify 8% of patients because of missing day 2 oxygenation data. Analysis of the clinical characteristics and presence of other Berlin criteria of this latter group demonstrate a similar profile to included patients, suggesting their exclusion does not bias this analysis. Use of a multiple imputation approach to assess the possible influence of missing data demonstrated similar results for predictors associated with mortality to that seen with the study population. Taken together, these findings support the robustness of our analytic approach.

While we have adjusted our analyses to account for known measured confounders, the possibility remains that some of our findings may arise from unmeasured or residual confounding. Moreover, we cannot make causal inferences for the associations seen, given the observational nature of our study. In this is an observational dataset, we are unable to test promising approaches to determining persistence of ARDS that involve standardization of ventilator settings, such as those proposed by Villar et al. [26]. Given the size of our database, small differences between groups may achieve statistical significance but be of uncertain clinical significance. Lastly, our assumption that inpatients at day 90 survived to hospital discharge is a further limitation.

Conclusions

All patients with ARDS, whether resolved or confirmed on the second day after onset, had a high mortality. Reclassification of ARDS severity on day 2 identified patients with severe confirmed ARDS as constituting a group at substantially increased risk for mortality. Clinical trials of complex interventions, especially those with significant side effects, might best be focused on this population. In contrast, the predictive value of ARDS reclassification at day 2 for mortality was relatively limited.