Introduction

Acute respiratory distress syndrome (ARDS) and acute lung injury (ALI) remain associated with unacceptably high mortality despite recent advances in supportive treatment [1]. Several open-label studies showed that ventilation in prone position improves arterial oxygenation with few untoward effects, in 58–100% of paediatric and adult patients [211]. Prone position enhances oxygenation via more even distribution of gravitational gradient in pleural pressure [12, 13], better distribution of ventilation to the dorsal areas of the lungs [1416], and potentially lesser overdistension of airspace, reducing thereby the occurrence of ventilator-induced lung injury [1720]. It also improves lung mechanics and alveolar ventilation [16].

In addition to a reduction in the intensity of ventilator support (lower inspired oxygen concentration, lower PEEP and mean airway pressure decreasing the ventilator-associated lung injury), resulting hence in facilitated patient recovery and earlier weaning from mechanical ventilation, beneficial physiologic effects of prone positioning are expected to ensue in improved overall outcome and especially in reduced mortality [21].

Two large randomised controlled trials (RCTs) and two moderately sized RCTs comparing prone position ventilation with supine ventilation in ARDS/ALI patients failed to show an improved survival rate in such patients [2225]. As a consequence of the accumulation of published negative trials, recent surveys have recorded a statistically significant decrease in the use of prone positioning at any time from 13% of ARDS patients in 1998 to 7% in 2004 [26]. This phenomenon has been recorded in the same intensive care units (ICUs) that were surveyed in 1998. It seems that the lack of evidence of an efficacy of prone positioning on a patient-centred outcome (mortality) has been interpreted as evidence of the absence of clinically relevant effect of prone positioning. This was reinforced by the lack of information on the trade-off between benefits and risks of prone positioning [increase in PaO2/FiO2 and reduced rate of ventilator-associated pneumonia (VAP) on the one hand, and increased workload, accidental tracheal tube displacement, and pressure sores on the other hand].

However, some of these RCTs were stopped before the termination of patients' inclusion, yielding studies substantially underpowered to demonstrate potential benefits of prone ventilation [22, 24, 25]. Additional methodological limitations accrue from inclusion of patients with different ARDS severity, patients in the late stages of ARDS, absence of ventilation guidelines, and the use of high tidal volumes. Moreover, these studies highlighted the lack of general agreement on the duration of prone position, both on a 24-h basis and throughout the course of ARDS.

Of interest, the most recent RCT on prone positioning showed in a multivariate analysis that prone ventilation was an independent factor associated with improved survival in patients with severe ARDS [25]. This study, which was prematurely stopped because of a slow inclusion rate, had the peculiar design of targeting a prolonged prone position (20 h per day).

Since none of the published RCTs was sufficiently highly powered to enable definitive conclusions, we undertook this meta-analysis to systematically review the published randomised trials assessing the effect of prone positioning on mortality of patients with ARDS/ALI.

Methods

Search strategy

Pertinent studies were independently searched in PubMed, EMBASE, CINAHL, and BioMedCentral (updated 30 November 2007), by two trained investigators (F.A., L.O.-B.) using the following MeSH and keyword terms: “acute respiratory distress syndrome”, “acute lung injury”, “acute respiratory failure”, and “prone position ventilation”. The literature search included both adult and paediatric populations. No language restriction was imposed.

Study selection

Titles, abstracts, and citations were independently assessed by both reviewers to assess the potential relevance for full review. From the full text, both reviewers independently assessed studies for inclusion based on the criteria for population, intervention, study design, and outcomes. Included studies met the following criteria: (1) All study participants, whether adults or children, had a clinical diagnosis of acute respiratory failure (ARF), acute respiratory distress syndrome or acute lung injury. ARDS and ALI were respectively defined by the radiographic evidence of bilateral pulmonary infiltrates, the absence of clinical evidence of left atrial hypertension, and a PaO2/FiO2 ratio of 300 or less (characteristic of ALI), or 200 or less (characteristic of ARDS). (2) Intervention: conventional ventilation in supine position compared with ventilation in prone position whatever its duration, on a 24-h basis, and during the ICU stay. (3) Design: prospective RCTs. (4) Outcomes: mortality rate, whether the ICU mortality or 28-day mortality, was the main outcome of this meta-analysis. It had to be clearly reported in the manuscript. Secondary outcomes corresponded to the following: the effect of prone positioning on PaO2/FiO2 ratio. This usually corresponded to the average change of PaO2/FiO2 ratio during the duration of the technique implementation. We also analysed the reported rate of VAP whatever its diagnosis method. Analysis also encompassed the incidence of procedure-related major airway complications. These were defined by the occurrence of accidental extubation, selective intubation, or accidental displacement of tracheal tube. We also compared the length of ICU stay.

Excluded from the meta-analysis were: (1) non-controlled studies; (2) studies that examined only the physiologic effects of prone positioning. These studies included either ARDS/ALI patients or patients with acute exacerbation of COPD.

Data abstraction and study characteristics

We extracted study design (including patient selection and randomisation), population, prone position duration on a 24-h basis, ventilatory strategy (whether a pulmonary protective strategy was used or not), and duration of follow-up. We also made a post-hoc power analysis corresponding to the power of the analysed study to detect the relative risk of mortality actually observed in that experiment. The primary end-point of our analysis was to determine the effect of prone ventilation on the incidence of ICU or 28-day mortality. Secondary end-points included effects on oxygenation during the acute phase of illness assessed by the PaO2/FiO2 change, the incidence of VAP and that of major airway adverse effects in relation with prone position. We also analysed the duration of ICU stay. We attempted to contact authors of included trials to request additional data if necessary.

Internal validity assessment

The methodological quality of each trial was evaluated using the 5-point scale (0 = worst and 5 = best) as described by Jadad et al. [27]. This instrument assesses the adequacy of randomisation, blinding, and the handling of withdrawals and dropouts.

Data analysis and synthesis

Our primary outcome was mortality in the ICU or at 28 days. Secondary end-points included respective effects of prone and supine position on oxygenation during the acute phase of illness (ranging from 4 to 10 days), the incidence of VAP, and that of major airway adverse complications (extubation, selective intubation) and minor complications (pressure sores) in relation to prone position. We also compared the duration of ICU stay.

Binary outcomes from individual studies were analysed according to the Mantel–Haenszel model to compute individual odds ratios (ORs) with pertinent 95% confidence intervals (CIs), and a pooled summary effect estimate was calculated by means of a fixed-effects model. Weighted mean differences (WMDs) and 95% CI were computed for continuous variables. Statistical heterogeneity and inconsistency were measured by using Cochran Q tests and I 2, respectively [28]. Since the majority of published studies were negative, the risk of publication bias was assessed by using visual inspection of funnel plot. Statistical significance is set at the two-tailed 0.05 level for hypothesis testing and 0.10 for heterogeneity testing. I 2 values around 25%, 50%, and 75% were considered to represent low, moderate, and severe statistical inconsistency, respectively. Unadjusted P values are reported throughout. The relationship between baseline PaO2/FiO2 ratio (a surrogate for patient's disease severity) or the duration of prone positioning (a surrogate for treatment intensity) on one hand, and the odds of mortality on the other hand, were evaluated using Spearman correlation. To investigate the hypothesis that higher effect might be disclosed in more severely ill patients, we also sought for a relation between the relative risk reduction recorded in each study and the mortality rate in the corresponding control (supine) group. The meta-analysis was conducted using RevMan 4.2.10. This study was performed in compliance with The Cochrane Collaboration and the Quality of Reporting of Meta-Analyses (QUOROM) guidelines [29].

Results

Study characteristics

Figure 1 shows a flow chart of studies assessed and excluded at various stages of the review. Finally, a total of six prospective RCTs were selected. One of these studies examined the effect of prone positioning on the prevention of lung worsening in comatose patients [30]. It contributed only to the analysis of the effect of prone positioning on the occurrence of VAP. The remaining five studies included hypoxaemic patients with ARF with ARDS or ALI. One multicentre study included paediatric patients [24]. Four studies included only adult patients (one including only severe ARDS patients [25], two including patients suffering either from ALI or ARDS [22, 31], and one including all patients suffering from ARF [23]). Although inclusion criteria corresponded not only to ALI/ARDS patients in the latter study, analysis concerned all included patients without stratification according to ARF aetiology. Following personal contact, the authors reported that mortality rates were fairly similar across ARF aetiologies. One of the studies included patients with post-traumatic ALI/ARDS [31]. Table 1 gives details of the studies' characteristics. There was total agreement between the two independent reviewers on inclusion of studies and the Jadad study quality grading.

Fig. 1
figure 1

Flow chart of the meta-analysis

Table 1 Characteristics of trials included in the meta-analysis

Data for 1,372 patients were available for analysis of the primary outcome, namely mortality; 713 patients were ventilated in prone position, and 659 were ventilated in supine position. All eligible reports were described as RCTs. They were published between 2001 and 2006. In the most recently published studies, patients were ventilated using a lung-protective ventilation strategy corresponding to a limited tidal volume (6–8 ml/kg of body weight) and a plateau pressure not exceeding 35 cm H2O [24, 25, 31]. The remaining two studies did not use such a ventilator strategy [22, 23]. In these five studies, duration of ventilation in prone position varied from 7 h to 18 h a day (mean duration per day 12 ± 5 h). It lasted between 4 days and the whole duration of hospitalisation. Three studies allowed crossover from one study arm to another [22, 23, 25] . This usually corresponded to patients in the supine position who were subsequently ventilated in prone position because of the persistence of severe hypoxaemia. In these studies, analysis was made on intention-to-treat basis. The main outcome evaluated in these studies was mortality reduction (either ICU mortality, or 28-day mortality) by ventilation in prone position in comparison with supine ventilation. Three studies were prematurely stopped because of a slow inclusion rate [22, 24, 25]. Calculation of studies' power showed that all positive studies were underpowered (< 10%–40%) to detect a statistical significance of the observed mortality reduction by prone positioning (Table 1).

Quantitative data synthesis

Effects on mortality

When we pooled all studies, ventilation in prone position was associated with a non-significant 3% reduction of the odds of mortality [249 of 713 patients (34.9%) in the prone ventilation group versus 234 of 659 patients (35.5%) in the supine position: OR 0.97, 95% CI 0.77–1.22, p for effect = 0.79, p for heterogeneity = 0.35; I 2 = 9.3%; Fig. 2]. The funnel plot of standard error versus OR for mortality did not suggest publication bias, with the effects of the largest studies closer to the non-effect line (Fig. 3). No correlation was found between the odds of mortality and either the PaO2/FiO2 ratio or the length of daily prone position in each of the included studies. Also, no correlation was found between the relative risk reduction of mortality and the corresponding mortality rate in the supine group.

Fig. 2
figure 2

Effect of ventilation in prone position on mortality. Weight is the relative contribution of each study to the overall estimate of treatment effect on a log scale assuming a fixed effects model

Fig. 3
figure 3

Funnel plot for outcome of mortality in trials of prone ventilation in ALI/ARDS. Each point represents one trial

Assessment of heterogeneity

Regarding the meta-analysis results, using the Q statistic test, we did not find statistical heterogeneity (p = 0.35). I 2, an alternative test for heterogeneity, was 9.3%, indicating low heterogeneity. However, there was substantial clinical heterogeneity in the included studies, with different patient populations (adult and paediatric populations; trauma and non-trauma patients; severe ARDS patients and mixed ALI/ARDS patients; ARF from other than ARDS/ALI origin), prone duration on 24-h basis, standardised lung protective ventilation or non-protective ventilation.

Secondary outcome measures

Effects on oxygenation: During the acute phase of the illness (which was assessed between the 4th and the 10th day of procedure implementation), both ventilation in prone position and ventilation in supine position were associated with an increase of the PaO2/FiO2 ratio. However, ventilation in prone position improved the PaO2/FiO2 significantly more (WMD by fixed effects model 25 mmHg, 95% CI 15–35, p for effect < 0.00001, p for heterogeneity = 0.06, I 2 = 56% corresponding to a moderate heterogeneity; Fig. 4).

Fig. 4
figure 4

Effects of prone positioning on the PaO2/FiO2 ratio

Effects on incidence of VAP: Data on the effects of prone positioning on VAP rate were available in only three of the former analysed studies for mortality rate and PaO2/FiO2 change. The results of the study by Beuret et al. [30] were added and contributed only to the analysis of this outcome. Accordingly, pooling data corresponded to those of 535 patients allocated to prone positioning and 482 patients ventilated in supine position. Pooled results show a reduction of the incidence of VAP by prone positioning (OR 0.77, 95% CI 0.57–1.04). This difference did not achieve statistical significance (p = 0.09) and there was a moderate heterogeneity among included studies (I 2 = 48.2%; Fig. 5).

Fig. 5
figure 5

Effects on the incidence of ventilator-associated pneumonia

Adverse effects of prone positioning: Major adverse airways events (defined as displacement of tracheal tube, selective intubation, or accidental extubation) occurred with similar frequency in prone ventilation and in supine ventilation: 75 of 713 patients (10.5%) in the prone ventilation group versus 69 of 659 patients (10.4%) in the supine position group: OR 1.01, 95% CI 0.71–1.43, p for effect = 0.95, p for heterogeneity = 0.61, I 2 = 0%; Fig. 6).

Fig. 6
figure 6

Incidence of major airway complications

Pressure sores and facial oedema were reported more frequently with prone positioning: in 296 (41%) out of 719 patients in the prone group and in 225 (34.1%) of 659 patients in the supine group (OR 1.35, 95% CI 1.08–1.69, p = 0.007).

Data on the effects of prone positioning on the length of ICU stay were reported in two studies only [23, 24]. Following personal contact we obtained data concerning the study by Mancebo et al. [25]. Overall, data concerned 540 patients in the prone group and 488 patients in the supine group. Ventilation in prone position was associated with a non-significant prolongation of the ICU length of stay (WMD = 0.96 days, 95% CI −1.11 to 3.02, p for effect = 0.36, p for heterogeneity = 0.96, I 2 = 0%; Fig. 7).

Fig. 7
figure 7

Pooled estimates of ICU stay (days)

Discussion

The current meta-analysis pooled data from 1,372 patients included in RCTs evaluating the effect of prone positioning in ALI/ARDS patients. It showed a substantial additional increase in the PaO2/FiO2 ratio in comparison to that evoked during ventilation in supine position, and a non-significant 3% reduction in mortality. Prone positioning was associated with a statistically non-significant 23% reduction of the risk of VAP, and did not increase the rate of major respiratory adverse events although specific untoward effects of prone positioning (pressure sores) were significantly more frequent. A marginal and non-significant increase of 0.96 days in the ICU length of stay was associated with prone positioning.

The optimal indication for a meta-analysis is pooling the results of trials with variable size but with similar design, in particular assessing the same outcomes related to a particular therapeutic tool in a well-delineated disease. Accordingly, one may wonder whether a meta-analysis is actually suitable for the evaluation of a technique such as prone positioning in a spectrum of disease like ALI/ARDS or ARF. Neither the intervention nor the disease spectrum is standardised. The included studies exhibited large variations in the intensity of the intervention (prone positioning was performed for between 7 h and 17 h per day, for 4 days or up to the whole length of stay), and there were substantial differences among studies in the stage of illness. Nevertheless, among other objectives, meta-analyses allow identification of heterogeneity effects among multiple studies. They also help identify data gaps in the knowledge base and may suggest direction for future research. Regarding prone positioning in ALI/ARDS patients, there is a gap between what published research actually says and the knowledge base of potential users. This is illustrated by the significant decline in the use of prone positioning, which was underscored in the recent international survey on the use of mechanical ventilation [26]. In this report, two consecutive surveys were conducted 5 years apart in the same ICUs belonging to 23 countries (and presumably among the same physicians), and were separated only by landmark publications on the effects of prone positioning in ALI/ARDS patients. Although knowledge translation to clinical practice may be delayed [32], in this particular setting, negative studies have rapidly been interpreted as definitively negative. These changes in patterns suggest that a scientifically sound approach like prone positioning, with a solid physiologic rationale, may be abandoned if nothing is done to rectify the impact of these publications and critically appraise their conclusions [33]. Careful analysis of the lower and higher boundaries of the CI around the point estimate of the odds of mortality, shows that obviously our meta-analysis does not rule out the possibility of an increase by 22% of mortality by prone positioning. Nonetheless, this meta-analysis does not also exclude a mortality reduction by 23%. In addition, this analysis shows that this strategy does not cause harm and might even have beneficial effects on the occurrence of VAP. Hence, an adequately powered RCT standardising the procedure of prone ventilation (prone positioning duration and use of lung-protective ventilation) is warranted in order to enable definitive conclusions on the benefits of this technique.

Ventilation in prone position is primarily used to increase oxygenation in patients with ARF through an improvement in ventilation perfusion matching, even distribution of gravitational gradient in pleural pressure, and a favourable modification of transpleural pressure gradients with a decrease of ventilator-induced lung injury [1215, 1720]. Compression of lung segments by the heart is also reduced [34]. Small non-controlled clinical studies evaluating prone position ventilation with or without additional therapeutic modalities such as nitric oxide or high-frequency oscillatory ventilation showed beneficial effects of prone positioning on oxygenation [29]. However, prospective randomised trials focusing on patient-centred outcomes did not find evidence of beneficial effect of prone positioning on mortality [2225, 31]. That improvement in physiologic parameters or surrogate end-points do not translate to clinically relevant benefit is not infrequent in critically ill patients [35, 36]. In addition, there are several possible reasons why improved oxygenation does not translate to improved mortality rate. First, improvement in oxygenation seems to have no impact on survival in ALI/ARDS patients [1]. Second, multiple organ failure rather than refractory oxygenation accounts for the majority of deaths related to ARDS [37]. Third, the protective experimental effect of the prone position could only be temporary [8, 17, 18]. Accordingly, if prone positioning merely postpones lung injury, it is crucial to take advantage of the window of opportunity associated with prone positioning.

In addition to these physiologic explanations, methodological weaknesses in many of published studies on prone positioning deserve consideration. Three studies were prematurely stopped because of a slow inclusion rate and lacked therefore adequate power to detect a statistical significance of the observed reduction in the mortality rate [22, 24, 25].

Along with a type II error, it is very important to consider to what extent the issue of case-mix might have contributed to the lack of evidence of prone position efficacy. ARDS is indeed at the severe end of the ALI spectrum, and published RCTs included patients suffering from various degrees of severity of ALI. This was consistently observed in the two largest studies evaluating prone positioning in a controlled and randomised design [22, 23]. Guerin et al. even included patients suffering ARF from other origins than ARD or ALI [23]. Of note, the study by Mancebo et al., which demonstrated a positive effect of prone positioning in the multivariate analysis, included only patients with severe ARDS [25]. These findings are consistent with those of Gattinoni et al., who showed in a post-hoc analysis that prone ventilation was more effective in the most severely ill patients (SAPS II > 49). Case-mix might also pertain to the “age” of ARDS. In Mancebo et al.'s study patients were included within 48 h of meeting inclusion criteria, while in Gattinoni's study inclusion of a substantial sample of the population occurred even late in the course of ARDS [22, 25].

Another explanation stems from the lack of standardisation of the prone procedure and that of ventilation strategy. Indeed, duration of prone positioning varied from 7 h to 17 h on a 24-h basis in the five studies included in the meta-analysis. The study by Mancebo et al. [25], which applied the longest prone duration (17 h/24 h) and found potential benefit from this procedure in the multivariate analysis, suggests that prone positioning should be applied for the longest time (more than 20 h) during the 24-h period. Along with this inference, it has been shown that the time course of alveolar recruitment during prone position is not consistent and in fact differs markedly from one patient to another [38]. In some patients, the plateau of complete alveolar recruitment may not be reached even after 8 h of prone positioning [38]. Accordingly, it has been suggested that the duration of prone positioning should be tailored on an individual basis by repeated measurements of alveolar recruitment. Future studies evaluating prone position ventilation in ARDS should probably apply a prone position duration closer to that of Mancebo et al. [25] than to that of Gattinoni's study [22].

Standardisation of the ventilation strategy is also to consider. Three of the five studies included in this meta-analysis were conducted prior to the generalisation of protective lung ventilation in ARDS [22, 23, 31]. Protective lung strategy could indeed be synergistic with the effects of prone position.

Compliance with assigned treatment is another concern raised by open-label studies assessing ventilatory techniques such as prone positioning. Indeed, Gattinoni et al. planned to use prone position for at least 6 h per day [22]. However, 27% of the patients allocated to prone positioning were ventilated prone for fewer hours than expected. A similar proportion of patients (25%) randomised to prone ventilation in the study by Guerin et al. were actually ventilated in prone position for a shorter time than the scheduled 8 h [23]. In addition, a large number of patients assigned in both studies to supine ventilation actually crossed over to prone ventilation because of a worsening in arterial oxygenation. Therefore, ascertainment bias, inherent to every unblinded trial, and limited compliance with the scheduled prone position sessions might have seriously compromised the accuracy of reported results of RCTs on prone position ventilation in ARDS/ALI patients.

Obviously, the involvement of the aforementioned limitations in published studies would make the drawing of definitive conclusions a matter of great difficulty. Given the difficulties in properly applying the results of RCTs in the real world, individualised care should definitively not be abandoned based on negative RCTs [39]. Despite the lack of strong evidence for the efficacy of prone position ventilation in reducing mortality in ARDS patients, and the fact that prone positioning has relatively lost ground, this technique remains in more than only exceptional use in most ICUs, and our study shows no sign of any harmful effect. From a pragmatic standpoint, physicians prefer to deal with a patient who is well oxygenated and requiring the lowest FiO2. Meanwhile, the optimal randomised controlled study devised to definitively answer the question of whether prone position ventilation should be used in ARDS is still to be conducted. Several studies are currently at different stages (ClinicalTrials.gov identifier: NCT00159939 and NCT00527813) [40, 41].