Introduction

Mechanical ventilation is a potentially life-saving intervention, though there is an increasing body of evidence for potential harm from this intervention in critically ill patients [1, 2]. Too high tidal volumes (VT) and airway pressures have been shown to be associated with worse outcomes in patients with acute respiratory distress syndrome (ARDS) [3, 4], and there is increasing evidence for the injurious effects of too high VT in ventilated patients without ARDS [5, 6]. While inadequately too low positive end-expiratory pressures (PEEP) have been demonstrated to worsen outcome of patients with ARDS, especially in moderate or severe cases [7], patients without ARDS likely do not benefit from higher PEEP [8]. More recently, a positive association between driving pressures (ΔP) and mortality was demonstrated in patients with ARDS [9], but it is unclear whether ΔP is associated with a worse outcome also in patients without ARDS.

Results from the ‘Large observational study to UNderstand the Global impact of Severe Acute respiratory FailurE’ (LUNG SAFE) [10], a prospective cohort study undertaken in 459 intensive care units (ICUs) in 50 countries, as well as the more recent ‘PRactice of VENTilation in patients without ARDS study’ (PRoVENT) [11], a prospective cohort study undertaken in 119 ICUs in 16 countries, convincingly showed that the practice of invasive mechanical ventilatory support in ICUs has changed remarkably over the recent years [10,11,12]. First, VT size decreased over time, not only in patients with ARDS [10, 12,13,14], but also in patients at risk of ARDS [11]. Presently, VT above 10 to 12 ml/kg predicted body weight (PBW) is seldom used. The median PEEP level that is set has increased over time in patients without ARDS [11, 13, 14]. In patients with ARDS, higher levels of PEEP usually are restricted to patients with more severe hypoxemia [7, 10, 15]. Both investigations, though, suggested there is still potential for improvement in ventilatory management in critically ill patients [10, 11], and one recently published secondary analysis of LUNG SAFE showed that lower PEEP, higher peak inspiratory (Ppeak), plateau (Pplat), ΔP, and increased respiratory rate represent potentially modifiable factors contributing to worse outcome in patients with ARDS [16].

The aim of the present study was to identify modifiable respiratory variables that could potentially change outcome in critically ill patients under invasive mechanical ventilatory support without ARDS. Specifically, we hypothesized that there are several modifiable respiratory variables associated with all-cause in-hospital mortality.

Methods

Study design

PRoVENT was an investigator-initiated international multicenter study; details of its methods have been published elsewhere [11, 17]. Details on study population and data collection are described in the supplement. PRoVENT was registered at Clinicaltrials.gov (NCT01868321).

Patients

Consecutive patients under invasive mechanical ventilatory support were eligible for participation if admitted in a predefined period lasting one week. Inclusion criteria were: (1) age ≥ 18 years and (2) under invasive mechanical ventilatory support, which could have been initiated outside the hospital, in the emergency room, in the normal ward or in the operating room, or start of invasive mechanical ventilatory support in the ICU, after admission. Patients in whom mechanical ventilatory support was started before the study recruitment week of PRoVENT, patients receiving only noninvasive mechanical ventilatory support or transferred from another hospital under invasive mechanical ventilatory support were excluded. Although data were also collected from patients who fulfilled the Berlin definition for ARDS [18] at start of ventilation, data of those patients were not used in the present analysis.

Definitions and calculations

The risk of death was derived from acute physiology and chronic health evaluation (APACHE) II scores [19] or simplified acute physiology score (SAPS) III [20].

Under the assumption that the maximum airway pressure (Pmax) during pressure-controlled assist modes of invasive mechanical ventilatory support is similar to Pplat during volume-controlled assist modes [21, 22], Pmax was defined as Pmax in pressure-controlled assist modes and plateau pressure in volume-controlled assist modes, when available. Also, ΔP was calculated by subtracting PEEP from Pmax during pressure-controlled and volume-controlled ventilation, respectively. This, however, was only done when set and measured respiratory rates were equal, indicating the absence of spontaneous breathing.

VT size was expressed as a volume normalized for predicted body weight (ml/kg PBW). The PBW of male patients was calculated as equal to 50 + 0.91(centimeters of height—152.4); that of female patients was calculated as equal to 45.5 + 0.91(centimeters of height—152.4) [23]. Dead space fraction was calculated as (partial pressure of carbon dioxide in arterial blood (PaCO2)–end-tidal carbon dioxide (etCO2))/PaCO2, and static compliance of the respiratory system as VT/ΔP. ‘Non-pulmonary’ sequential organ failure assessment (SOFA) was calculated by leaving out the pulmonary component and amending the denominator accordingly. The presence of acidosis was split into respiratory and metabolic acidosis to include separately in the univariate analysis, under the assumption that a respiratory acidosis could be modifiable by adjusting respiratory minute volume as opposed to metabolic acidosis. Immunosuppression was defined as the presence of human immunodeficiency virus or the use of chemotherapy, systemic steroids (> 1 mg/kg of prednisone or equivalent), or other immunosuppressive agents.

Outcomes

The primary outcome was all-cause in-hospital mortality, defined as mortality at hospital discharge, or at 90 days after start of invasive mechanical ventilatory support while still in hospital, whichever occurred first. The secondary outcome was ICU mortality, defined as mortality at ICU discharge or at 90 days after start of mechanical ventilatory support while still in ICU, whichever occurred first.

Statistical analysis

Daily-collected variables, including Pmax or Pplat, ΔP, PEEP, VT, oxygen fraction of inspired air (FiO2), respiratory rate, dead space fraction, and compliance, and blood gas analysis parameters such as partial pressure of oxygen in arterial blood (PaO2), PaCO2, pH, and bicarbonate level, were presented as medians with their interquartile ranges. Proportions were compared using Chi-square or Fisher’s exact tests, and continuous variables were compared using the t test or Wilcoxon rank sum test, as appropriate. Since the amount of missing data were low, no assumptions were made for missing data.

In all descriptive analyses, survivors were separated from non-survivors according to all-cause in-hospital mortality. In univariate analyses assessing the impact of ventilatory variables on outcome, relative risk (RR) of in-hospital mortality was estimated for patients dividing the study sample according to the median of Pmax (≤ 18 vs. > 18 cm H2O), ΔP (≤ 12 vs. > 12 cm H2O), PEEP (≤ 5 vs. > 5 cm H2O), and VT (≤ 7.9 vs. > 7.9 ml/kg PBW), as measured at the first day of ventilation. For this specific analysis, two separate groups were included: patients not at risk and patients at risk of ARDS according to the Lung Injury Prediction Score (LIPS), where a LIPS ≥ 4 was considered ‘at risk of ARDS’ and a LIPS < 4 ‘not at risk of ARDS.’

To identify potentially modifiable and non-modifiable factors contributing to hospital mortality, a multivariable model was built using demographic factors, comorbidities, illness severities, and respiratory and laboratorial variables at the first day of ventilation. Since Pmax and ΔP have a high collinearity, we chose to include only Pmax in the main model. We conducted multilevel analyses to adjust for clustering of the data. Therefore, a multilevel logistic regression was used to identify factors contributing to mortality by modeling it as the dependent variable. Variables were selected when the univariate analysis p value was< 0.2. Then, a multilevel multivariable logistic model was built with centers treated as random effect. The cluster effects induced by the structure of the data were taken into account through random effects. In the multivariable model, statistical significance was set at a p < 0.05. Results are shown as odds ratios (ORs) with 95% confidence intervals (CI).

The odds ratio for hospital mortality of Pmax was plotted in curves showing the odds ratios according to increases of one standard deviation of the Pmax. These curves were divided according to the risk of ARDS and adjusted for the variables included the final model and reported in Table 3. A similar curve was made using ICU mortality as outcome.

We performed a secondary analysis in which we replaced Pmax with ΔP in the multivariate model for in-hospital and ICU mortality. Since we lacked reliable values for ΔP for a large group of patients, this analysis had a much smaller sample size, increasing the risk of losing power to show an association between ΔP and in-hospital mortality. To test this, we performed one post hoc analysis in which we used Pmax instead of ΔP, but only for patients for whom we had a reliable ΔP.

Statistical significance was considered to be at two-sided p < 0.05. All analyses were performed with SPSS v.20 (IBM SPSS Statistics for Windows, Version 20·0. Armonk, NY: IBM Corp.), and R v.2·12·0 (http://www.r-project.org).

Results

Participating centers and patients

One hundred and nineteen ICUs from 16 countries in four continents enrolled 1021 patients under invasive mechanical ventilatory support. Excluding 86 patients who were admitted to ICU with ARDS, we analyzed the data from a total of 935 patients (Fig. 1). All-cause in-hospital mortality was 21% in all patients. Patients who survived had lower derived risk scores for mortality, were younger, and had lower SOFA scores; patients who died were more often functionally dependent and more often admitted for a medical condition or for emergency surgery (Table 1).

Fig. 1
figure 1

Flowchart of inclusion

Full size image
Table 1 Demographic characteristics of patients without ARDS receiving mechanical ventilation, comparison of survivors and non-survivors
Full size table

Ventilation characteristics

Patients who survived had a lower Pmax or Pplat, lower ΔP, lower PEEP, and lower FiO2 levels than patients who died, but a similar VT (Table 2). PaO2/FiO2, pulse oximetry, and arterial pH were higher and PaCO2 levels were lower in patients who survived (Table 2). The unadjusted impact of ventilatory parameters in the overall cohort and in each group of risk of ARDS is shown in Fig. 2. Mortality risk was similar in patients stratified according tidal volume and ΔP. In the overall cohort, patients receiving higher PEEP had higher risk of hospital mortality (Fig. 2). Patients ventilated with higher Pmax had a higher risk of hospital mortality in the overall cohort and in patients at risk of ARDS (Fig. 2).

Table 2 Characteristics of critically ill patients without ARDS receiving mechanical ventilation, comparison of survivors and non-survivors
Full size table
Fig. 2
figure 2

Unadjusted relative risks of hospital mortality in the overall cohort and in patients at risk and not at risk of ARDS and according to the median of the: a Pmax; b PEEP; c ∆P; and d tidal volume. Abbreviations: Pmax: maximum airway pressure; PEEP: positive end-expiratory pressure; VT: tidal volume; ∆P: driving pressure; RR: relative risk; CI: confidence interval

Full size image

Factors associated with in-hospital mortality

The results of the univariable analysis of factors associated with in-hospital mortality are provided in Additional file 1: Table S1. In multivariable analysis, Pmax was the only ventilatory variable associated with higher in-hospital mortality; in this analysis, the ΔP was excluded due to the collinearity with Pmax (Table 3). Non-modifiable factors associated with worse outcome were older age, presence of immunosuppression, higher non-pulmonary SOFA, lower pulse oximetry readings, higher heart rates, and functional dependency (Table 3).

Table 3 Factors associated with in-hospital mortality in patients without ARDS receiving mechanical ventilation
Full size table

Figure 3 shows the odds ratio for hospital mortality per increase in one standard deviation in Pmax for patients not at risk of ARDS and patients at risk of ARDS and adjusted for the variables indicated in Table 3.

Fig. 3
figure 3

Odds ratio of hospital mortality according to increases in one standard deviation of Pmax and in the patients at risk and not at risk of ARDS. All curves are adjusted by the same set of variables described in Table 3

Full size image

Factors associated with ICU mortality

Results of the univariable analysis of factors associated with ICU mortality are provided in Additional file 1: Table S2. After multivariable adjustments, Pmax was the only ventilatory variable associated with worse outcome (Additional file 1: Table S2); non-modifiable factors associated with worse outcome were history of COPD, presence of immunosuppression, higher non-pulmonary SOFA scores, and functional dependency.

Additional file 1: Figure S1 shows the odds ratio for ICU mortality per increase in one standard deviation in Pmax for patients not at risk of ARDS and patients at risk of ARDS and adjusted for the variables indicated in Additional file 1: Table S2.

Driving pressure

The analysis including ΔP was only possible in 343 patients for whom ΔP could be calculated in a reliable way. When considering ΔP instead of Pmax in the model, there was an association between ΔP and ICU (Additional file 1: Table S3), but not between ΔP and in-hospital mortality (Additional file 1: Table S4). The lack of an association between ΔP and in-hospital mortality could very well have been caused by the smaller sample size, since the post hoc analysis in which we used Pmax in the model, now using the same number of patients as done for the analysis including ΔP, also showed no association between Pmax and in-hospital mortality (Additional file 1: Table S5), while the association between Pmax and ICU mortality remained present (Additional file 1: Table S6).

Discussion

In the present study, older age, presence of immunosuppression, a more dependent premorbid condition, and severity of illness markers such as the pulse oximetry, the non-pulmonary SOFA score, and a higher heart rate were all independently associated with increased in-hospital mortality. In the present analysis, Pmax was the single ventilator factor associated with in-hospital mortality, suggesting this is the only potentially modifiable factor in these patients. Parts of our findings are in line with prior studies in this field. Older age is independently associated with worse outcome in patients with ARDS [16, 24, 25] and patients without ARDS [13], and also immunosuppression is a risk factor for mortality in our study and in trials that included patients with ARDS [16, 25]. Severity of illness factors associated with outcome was a higher heart rate and higher non-pulmonary SOFA score, consistent with previous studies in patients with [16, 25], as well as in patients without ARDS [13]. In addition to the results of the LUNG SAFE [16], we here show that, irrespective of the presence of ARDS, older patients, patients with immunosuppression, patients with high non-pulmonary SOFA score, and higher heart rate are at increased risk of worse outcomes. Ventilatory support with a higher Pmax was independently associated with both increased hospital mortality and ICU mortality. This finding is in accordance with previous studies where higher Pmax was associated with worse outcomes, for example increased risk of ventilator-induced lung injury (VILI) [26, 27], and increased mortality in patients without ARDS [13] and those with ARDS [16, 21, 28, 29].

ΔP was only associated with ICU mortality and not with in-hospital mortality. It should be recognized, though, that that analysis was only possible for 343 patients, and this smaller sample size may have reduced the power so that there was no association between ΔP and in-hospital mortality. This could also be concluded from the results of the post hoc analysis of Pmax, using the same smaller cohort of patients. Nevertheless, the finding that ΔP was not associated with in-hospital mortality is in line with a recently published investigation in a cohort of patients without ARDS [30]. In addition, the small range of tidal volumes used in this cohort also led to a small range of ΔP, which could blunt the effect of ΔP on mortality, which may be much subtler than is found in patients with ARDS [9]. Similar findings came from a recently published study that failed to find an association between ΔP and mortality, even though their results show a trend for higher mortality rates with each cm H2O increase of ΔP [30]. Yet the influence of ΔP on outcome is consistent with previous reports exposing the importance of ΔP on development of pulmonary complications also in patients without ARDS undergoing general anesthesia for surgery [31], and on ventilator-induced diaphragmatic injury in critically patients receiving mechanical ventilation [32]. Similarly, experimental studies suggested an association between higher ΔP and development of VILI. In studies considering patients with ARDS, ΔP was the ventilation variable that best stratified mortality risk, even in those undergoing ECMO for refractory hypoxemia [9, 16, 28, 33, 34].

While higher VT was related to worse outcomes in critically ill patients without ARDS [5, 6, 35], and with pulmonary complications in patients undergoing general anesthesia for surgery [36,37,38], in this analysis as well as the earlier reported primary analysis of PRoVENT [11], such an association was not found. The lack of a relationship between VT and outcome in the present study likely reflects the widespread adoption of lower VT ventilation, as VT in our cohort concentrated in a narrow range around a median of 7.9 ml/kg PBW. With less patients receiving ventilation with high VT, the association between VT and outcome was no longer present. This finding is in line with the abovementioned recently published investigation in a cohort of patients without ARDS [30]. We are awaiting the results from two randomized controlled trials (RCT) testing different VT in patients without ARDS [39, 40].

A higher PEEP level was not associated with outcome in our study, and this is similar to previous findings [8, 11, 41]. However, one small randomized controlled trial found that application of ‘prophylactic’ PEEP in non-hypoxemic ICU patients not only reduced the number of hypoxemic episodes, but also the incidence of ventilator-associated pneumonia [42]. Nevertheless, most trials performed so far that addressed the effects of PEEP on outcomes in ICU patients without ARDS were relatively small and mainly assessed other outcomes than mortality, for example development of pulmonary complications [8]. Well-designed RCT are needed to address the true impact of PEEP in ICU patients without ARDS.

We suggest that the risk of ARDS can act as an additive to ‘injurious’ ventilation, which can be explained by a smaller inspiratory capacity in these patients. When the inspiratory capacity is exceeded, stress failure occurs [43, 44]; thus, the level of a certain ventilation parameter could be well within the inspiratory capacity of a patient not at risk, while exceeding the smaller capacity of a patient at risk. These findings are particularly important since PRoVENT found differences between the ventilatory management of patients at risk and not at risk of ARDS [11]. While within the inspiratory capacity, the only independent variable for VILI is dynamic strain, i.e., VT, above the inspiratory capacity, the combination of all ventilation parameters can lead to VILI and worse outcome [43, 44].

The present analysis has several limitations. It is important to note that we classified pulse oximetry as non-modifiable; however, one could argue that this is modifiable through adjustment of FiO2. Also, although respiratory variables are potentially modifiable, adjustment of the ventilator can be influenced by certain non-modifiable factors that are present at the time of adjustment. For example, PEEP is affected by hypoxemia; some protocols allow higher plateau pressures in the presence of severe acidemia, and ΔP is directly influenced by changes in the respiratory system compliance. These interactions are complex, and ventilator settings may not always turn out to be modifiable when treating a patient. Another limitation is the use of maximal airway pressure in pressure-controlled mode as a surrogate for the plateau pressure to calculate ΔP, although this was only done when there was no proof of spontaneous breathing efforts to minimize erroneous measurements. Prospective trials are needed investigating specifically the directly measured pressures in the lung, including the transpulmonary driving pressure, to explore their effect on outcome in patients without ARDS.

By identifying potentially modifiable factors in care of ICU patients, we indicate what future implementation studies should focus on to actually prove benefit of the suggested strategies on outcome. The identification of non- or less-modifiable factors points out which patients are more vulnerable and potentially may benefit most from an early start of protective treatment strategies.

Conclusion

The present analysis of a large prospective observational study suggests that higher Pmax was a potentially modifiable factor associated with increased in-hospital mortality in critically ill patients without ARDS. Whether ΔP is also a potentially modifiable factor associated with increased in-hospital mortality needs further testing in larger patient cohorts.