Mechanical ventilation (MV) is a mainstay of supportive therapy for patients with acute respiratory failure, both to maintain adequate systemic oxygenation and to rest the respiratory muscles. After its introduction into clinical practice (1), physiological studies advanced the basic understanding of MV to the point where randomized controlled trials could be undertaken to evaluate the impact of different ventilator strategies on clinical outcome variables. Such trials usually compare a new treatment with conventional MV therapy. However, the difficulty of defining “conventional” MV has emerged because of the extreme variability in its use. Esteban et al. found that, although indications for ventilation were similar in different countries, profound differences were observed in the percentage of ventilated patients, selection of ventilatory modes, and weaning strategies (2). This marked heterogeneity has been associated with the lack of defined standards of care (2-4) and with the observed gap between the results of clinical trials and clinical practice (25). This may be of concern, as ventilatory management has been shown to have an important impact on prognosis (2, 4, 68).

A pre-planned secondary analysis of the SAPS 3 study (9) was performed aiming to describe the characteristics of patients receiving MV, the modes of mechanical ventilation and ventilator settings used and the effects of these variables on clinical outcome;


The study represents a prospective, multinational pre-planned cohort substudy of the SAPS 3 project (9, 10). Data were collected at ICU admission (± 1 h), and on days 1, 2 and 3 after ICU admission, and in the last 24 h of the ICU stay. Data included socio-demographics, chronic conditions, diagnostic information, physiological derangement at ICU admission, number and severity of organ dysfunctions as evaluated by the Sequential Organ Failure Assessment (SOFA) score [11], and vital status at ICU and hospital discharge. SOFA scores were computed as proposed by Moreno et al. [12]. A complete description of the study protocol and patient and ICU characteristics can be found elsewhere (9).


From the SAPS 3 Hospital Outcome Cohort (comprising 16,784 patients from 303 ICUs), 2,946 postoperative patients with planned ICU admission, scheduled surgery, an length of stay (LOS) in the ICU of <=2 days who were discharged alive from the ICU were excluded. Further patients with missing ventilatory support on either all days (n=219) or at admission (n=297) were excluded. Thus, the study cohort comprises of 13,322 patients from 299 ICUs. A list of countries, ICUs and patient numbers can be found in the Appendix of the ESM.

Patients were divided into three mutually exclusive subgroups according to their ventilatory status at ICU admission: patients not receiving any type of MV (noMV; n = 6,261); patients receiving positive-pressure MV via a noninvasive interface (facial, oro-nasal or nasal masks or helmet) (NIV; n = 554), and patients receiving MV via an invasive interface (endotracheal tube or tracheostomy) (invMV; n = 6,507).

Documentation of mechanical ventilation

Ventilatory settings were registered at ICU admission. Ventilatory modes registered during NIV were Continuous Positive Airway Pressure (CPAP), and Pressure Support Ventilation (PSV). Ventilatory modes registered during invMV were CPAP, PSV, Synchronized Intermittent Mandatory Ventilation (SIMV), SIMV+PSV, Pressure Control Ventilation (PCV), Assisted / Controlled Mechanical Ventilation (CMV and A/CMV). Further modalities used during NIV and invMV were captured as “others”. The exact definitions can be found in the Appendix of the ESM.

The following ventilatory settings were registered: tidal volume (Vt) (mL/kg of actual body weight), positive end-expiratory pressure (PEEP) (cm H2O) and fraction of inspired oxygen (FiO2). Unless otherwise indicated, settings from the day of ICU admission were used for analysis.

Data quality

Quality of the recorded data was evaluated for completeness of the documentation and reliability. Inter-rater quality control was performed through rescoring of the data and calculation of kappa coefficients and intra-class correlation coefficients, as appropriate (13). Results of this analysis are presented in detail in the Electronic Supplementary Material (ESM) file of the SAPS 3 primary report [9].

Statistical analysis

Statistical analysis was performed using the SAS system, version 9.1 (SAS Institute Inc., Cary, NC, USA). A P value of < 0.05 was considered significant. Unless otherwise specified, results are expressed as median and quartile ranges. The chi-square test was used for categorical data, and the Wilcoxon rank sum test or the Kruskal-Wallis test was used to compare two or more groups. Missing values (4%) were replaced by imputed values (using SAS procedure MI). For this purpose, imputations were done five times (n=13,322 each) for missing values, thus providing five datasets.

To explain which factors contributed to the use of MV, a stepwise multiple logistic regression was performed in each of the five imputed data sets. Those variables were selected for the final model, which showed a significant contribution at a two-sided level of 0.05 in all five analyses. The final estimates of the coefficients were determined by a weighted average of the coefficients found in the five imputed data sets (using SAS procedure MIANALYZE). In order to adjust for regional differences, regions were entered as fixed effects into the model. To evaluate the impact of imputation on the results, a sensitivity analysis with the complete cases dataset, excluding patients with at least one missing value in the explanatory variables, was performed.

To explain which factors contributed to the use of either NIV or invMV, a stepwise multiple logistic regression was performed in the same way as described above. The list of variables which were entered as independent variables into the models can be found in the ESM (Appendix A).

Moreover, to analyze the association between PEEP level and hospital mortality, a multivariate analysis was done using different PEEP categories as independent and hospital mortality as dependent variables. SAPS 3 was used to adjust for severity of illness.


The SAPS 3 ventilation cohort comprises 13,322 patients admitted to 299 participating ICUs from 35 countries. Median SAPS 3 score was 51 (41-63), and 60.2% of patients were male, with a median age of 63 (48-74) years. More than two thirds of the patients were admitted to the ICU from the same hospital, mostly from emergency departments, operating rooms, and normal wards (Table 1). Since planned surgical admission with a LOS < 2 days were excluded from the study cohort, most admissions were unplanned. Roughly half of the patients (52.7%) were admitted for medical reasons.

Table 1 Demographic and clinical data

Use of Mechanical Ventilation

Of the 13,322 admitted patients, 53% were mechanically ventilated at admission, with the majority being invasively (48.8%), and only a small number of patients (4.2%) noninvasively ventilated. 61.5% (8,196) were mechanically ventilated during the first 3 days of intensive care, with 7.3% being managed by NIV. Of the 5,970 patients still in the ICU by day 3, still 59.1% (3,527) were ventilated. Significant differences between ICUs and regions were observed (ESM, Table E2, Figure E1). Surgical patients had a significantly higher relative risk (RR) of being mechanically ventilated at ICU admission (RR, 1.613; 95% confidence interval [95% CI], 1.546-1.684) compared with medical patients.

The main acute medical diseases responsible for ICU admission were cardiovascular, respiratory and digestive diseases (ESM Table E1). The use of MV (noninvasive or invasive) varied significantly by acute medical disease: Patients with traumatic brain injury had the highest incidence of invMV (72.1%), and pregnant women had the lowest (37.2%) (ESM Table E3). Several organ dysfunctions and failures increased the relative risk of receiving MV (Table 2). Multivariate analysis showed several variables to be associated with the use of MV (Table 3): increasing age, chronic heart failure class NYHA III or IV; basic observational care, neurologic and respiratory reasons for admission; emergency surgery; lower respiratory tract infection; hospital acquired infection; lower GCS score, increasing heart rate, increasing white blood count, decreasing pH;

Table 2 Organ dysfunction / failure and the use of mechanical ventilation
Table 3 Multivariate analysis: Predictors for the use of mechanical ventilation (MV versus no MV)

Noninvasive Mechanical Ventilation

CPAP was the most used mode of NIV (69.7%), followed by PSV (28.3%). 67.3% of non-invasively ventilated patients had a medical reason for admission. Acute exacerbation of COPD was the main indication for NIV (24.7% of NIV patients). However, only a minority of COPD patients received NIV (17.4%). Whether or not hospitals were university affiliated did not seem to influence the use of NIV (4.4% vs. 3.9%; p = 0.136). Multivariate analysis showed several variables to be independently favouring the use of NIV (Table E4, ESM).

Invasive Mechanical Ventilation

In patients receiving invasive MV, A/CMV was the main mode of invasive ventilation (46.4%), followed by PCV (19.7%) and SIMV + PSV in 16.3%. PSV was used in 6.4%, SIMV in 6.3%, “other” types of invMV in 3.3% and CPAP in 1.6%. The use of invMV varied between regions (ESM Table E2, Figure E1), and also between different groups of acute medical diseases (ESM, Table E3).

FiO2, Vt and PEEP used during invasive MV were on average 50% (40-80%), 8 mL/kg body weight (6.9-9.4 mL/kg) and 5 cmH2O (3-6 cmH2O), respectively. More than half (51.6%) of patients were ventilated with a Vt ≥ 8 mL/kg body weight, and still 25% were ventilated with a Vt ≥ 9.4 mL/kg. More than a third (33.4%) with a PEEP of <5 cm H2O, with 17.3% (n=1,128) patients with ZEEP.

For reasons of analysis, patients were grouped into four quartiles of the PaO2/FiO2 ratio: 106.5 (79.3-131.4), 196.9 (175.0-219.6), 294.5 (267.5-320.0) and 430 (386.7-494.0) respectively. With decreasing PaO2/FiO2 ratio, patients were more often ventilated with a PEEP ≥ 5 cm H2O and tidal volumes <8 mL/kg (Figure 1, ESM Table E5). In patients with evidence for severe respiratory failure (lowest PaO2/FiO2 ratio quartile) this difference was statistically significant (both p<0.001). Moreover, in the subgroup of patients with ALI/ARDS (n=998), 15.9% of patients were ventilated with a PEEP < 5cm H2O and 30.5% with a Vt ≥ 8 mL/kg body weight.

Fig. 1
figure 1

Tidal Volume and PEEP according to PaO2/FiO2 ratios. The graph shows the percentage of patients ventilated with a PEEP ≥5cmH2O (grey bars) or tidal volumes <8mL/kg body weight (white bars) for different groups of the PaO2/FiO2 ratio


Mechanically ventilated patients exhibited a significantly increased raw mortality rate compared to noMV patients (36.4 vs. 17.5%, p<0.001). However, after risk adjustment, mortality rates were not different between the two groups (1.01 [0.99-1.04] and 1.00 [0.95-1.05], respectively). Patients assisted with NIV exhibited a significantly reduced risk-adjusted mortality compared to patients assisted with invMV (0.79 [0.69-0.90] vs. 1.03 [1.00-1.06, respectively]. The excluded postsurgical patients (n=2,946) exhibited a hospital mortality of 2.5%.

Hospital mortality was not different whether Vt was higher or lower than 8mL/kg in the whole cohort (35.7 vs. 37.4%; n.s., O/E ratio 1.01 (0.97-1.05) and 1.02 (0.97-1.06), respectively). However, in the least hypoxemic patients a Vt ≥ 8mL/kg was associated with a trend for an increased risk of death, (relative risk 1.08, (95% CI: 0.98 - 1.19).

There was no difference in mortality rates whether PEEP was higher or lower than 5 cm H2O in the whole cohort (36.3 vs. 35.4%, O/E ratio 0.99 (0.96-1.02) and 1.03 (0.98-1.08), respectively) or in the different PaO2/FiO2 quartiles. However, in invMV patients, ZEEP was associated with significantly increased odds to die in the hospital (OR 1.29 [1.08-1.53], ESM Table E7). This resulted also in a significantly increased risk-adjusted mortality for these patients (O/E ratio 1.12 [1.05-1.18], ESM Figure E2].

Sensitivity analysis revealed that that some influence factors which were significant in the imputed data set did not reach statistical significance within the complete cases data set (data not shown).


This study is, to the best of our knowledge, one of the largest ever performed on the epidemiology of mechanical ventilation in the ICU. Most of the previous studies involved specific patient typologies (e.g., acute lung injury, acute respiratory distress syndrome or COPD) or were done in fewer centres. On the other hand, this study used a high-quality multinational database that, although not primarily designed for the assessment of the epidemiology of mechanical ventilation, may in fact reflect the heterogeneity of current ICU case mix and typology from 299 ICUs in 35 countries all over the world.

We found that mechanical ventilation is one of the most common interventions employed in intensive care: More than half of the patients were ventilated in the first 24 h after ICU admission, a value greater than that described in some studies (2, 3, 14, 15) but similar to that in others (16, 17). Regarding this figure, it should be noted that 2946 post-surgical patients were excluded, since they represent a cohort that needs rather intermediate then intensive care. The fact that these patients exhibited an extremely low hospital mortality of 2.5% supports this.

Despite the widespread recommendation for the use of NIV in patients with acute exacerbation of COPD and with cardiogenic pulmonary edema, immuno-suppression and specific postoperative conditions (18), our study shows that NIV was used only in a minority of ventilated patients (4.2% at admissi\\\on, 7.3% during the first three days). These values are lower than the 11% reported by Esteban et al. in a worldwide cohort (15) or the 23% of the French survey (17). Consistently we found that use of NIV varied significantly with ICU type and region. The fact, that patients assisted with NIV exhibited a better outcome than invasively ventilated patients confirms the need for a careful selection of possible candidates for NIV (18, 19, 20).

As reported by others, A/CMV was the main invasive ventilatory mode, (3, 15, 19). Our data confirm the widespread use of SIMV as an isolated ventilatory mode. SIMV was used in 6.3% of invasively ventilated patients, which is similar to the 5.8% reported by Esteban et al. in 2000 (3) (see also ESM, Figure E1 + E2).

Ventilation with high Vt has been repeatedly linked to lung injury in ALI/ARDS (4, 24-26) and in normal lungs (6, 7, 27). In our cohort, setting of Vt varied significantly: more than half of invMV patients were ventilated with a Vt ≥ 8 mL/kg actual body weight. Moreover, a third of patients admitted due to ALI/ARDS were ventilated with such a setting. This finding is consistent with data from previous studies (21-23) and might be cause for concern. Especially if we consider that actual body weight usually exceeds predicted body weight by 20% [4], but ventilator settings should rather use the latter one. Unfortunately, there were too many missing data on patients’ height to calculate predicted body weight in our cohort. This means that our median Vt of 8.0 mL/kg of actual body weight might even underestimate the really used Vt – which could be as high as 9.6 mL/kg of predicted body weight, a value similar to that described in previous series [28].

Patients with the lowest PaO2/FiO2 ratios were ventilated more often with tidal volumes <8mL/kg body weight, than were patients with higher PaO2/FiO2 ratios (Figure 1, white bars). This is consistent with results from previous studies showing that lower values of Vt were used in patients with a higher severity of lung injury (21, 23). However, it should be noticed that in our cohort still 45.8% of these patients were ventilated with tidal volumes ≥8 mL/kg. Since a low PaO2/FiO2 might be indicative for ALI, this seems to be at least a reason for concern. Correspondingly, higher PEEP values were used in these patients (Figure 1, grey bars). Again it should be noticed, that still approximately a third of patients with the lowest PaO2/FiO2 ratio (27.8%) and 15.9% of patients with the admission diagnosis ALI/ARDS were ventilated with PEEP values <5 cm H2O (Figure 1, ESM Table E5).

A substantial portion of invasively ventilated patients was managed without any PEEP at all. Invasive mechanical ventilation with ZEEP was independently associated with mortality (ESM Table E6, Figure 2). It is well known that PEEP may prevent alveolar de-recruitment associated with lower Vt, thus reducing shear stress and also reduce the impact of denitrogenation atelectasis (observed in patients ventilated with high oxygen concentrations) (29-32). Moreover, Ferguson et al. (23) have shown an association between low values of PEEP and mortality in ARDS patients. Although an independent association between ZEEP and higher mortality has not been consistently found in all patient typologies, a large amount of data exists which supports this finding (4, 22, 24, 25, 33-37).

Fig. 2
figure 2

PEEP-values and hospital mortality in invasively mechanically ventilated patients.

The graph shows the odds ratios for hospital mortality for different groups of PEEP-levels in invasively ventilated patients. Median and interquartile ranges are given for the respective deciles

Our study has obviously some limitations: First, data on the use of MV were limited to the first 3 days of the ICU stay. Second, due to the epidemiologic nature of our study, no data concerning ventilatory mechanics or specific morbidity potentially associated with MV, such as barotrauma or infection are available. Moreover, data about the eventual use of MV before study enrolment were not available. Also, data about indications for MV were not available. For this reason we cannot fully exclude a potential bias which might confound the association between MV settings and outcome.

With respect to the use of NIV we cannot present data on the use of NIV before ICU admission, nor about those patients who were failing on NIV and therefore later during the ICU stay switched to invMV. Thus, the improved outcome (although risk-adjusted) must be viewed with caution and needs further prospective studies.

Multiple imputations to compensate for missing values were performed in cases of missing values (which made up 4%). However, differences were found between the complete cases data set and the imputed data sets in a sense that some influence factors in the imputed data set did not reach statistical significance within the complete cases data set. This may be explained by the much smaller sample size of the complete cases dataset (n= 7,.050) and is furthersupported by the observation, that most of the non-significant odds ratios in the complete cases data set were further away from one than the corresponding significant odds ratios in the imputed data set. For this reason we decided to report only the imputed values here.

In conclusion, MV is one of the most common interventions employed in the intensive care unit, with more than half of the patients being mechanically ventilated. However, the use of MV is still variable and possibly also a reason for concern: Several patients with evidence for acute lung injury (low PaO2/FiO2 ratios) were ventilated with tidal volumes >8 mL/kg actual body weight. Moreover, ZEEP was found to be used in a high percentage of invasively ventilated patients and this was associated with a worse outcome, even after controlling for severity of disease.