In patients on mechanical ventilation (MV), ventilator-associated pneumonia (VAP) is a common iatrogenic infection [1], which results in longer periods of MV and additional broad-spectrum antibiotics, ultimately increasing healthcare costs [2]. Although, some reports emphasize a decline in VAP rates over the last decades [3, 4], the overall burden of this iatrogenic condition is still considerable [5].

In the gastro-pulmonary route of colonization, during MV, the gastrointestinal tract can become colonized by pathogens [6]. Gastrointestinal stress ulcer prophylaxis and enteral nutrition, which alkalinize gastric fluids, were believed to be the main risk for endogenous colonization; yet, recent evidence debates these previous mechanisms [7]. Also, gastroesophageal reflux is consistent in enterally fed patients [8]; thus, pathogens from the stomach may be aspirated into the airways [9]. Conversely, other investigators have focused on pulmonary aspiration of oropharyngeal pathogens [10, 11], irrespective of gastric colonization.

Several VAP preventive strategies have been developed, which are often applied as a multifaceted bundle [12]. Among those, the semirecumbent position (SRP) has been one of the most implemented measures. The SRP reduces gastroesophageal reflux [8, 13], and a randomized clinical trial demonstrated reduced incidence of VAP, in comparison with the supine fully horizontal position [14]. Later studies reported controversial outcomes [15], but they were limited by small sample sizes, applied methods, and heterogeneity in the degree of head elevation [16].

More recently, laboratory studies [17,18,19,20] have shown a drastic reduction in VAP in animals positioned in the Trendelenburg position with a tracheal and endotracheal tube (ETT) orientation below horizontal. In these preclinical studies, the Trendelenburg position limited pulmonary aspiration and improved mucus clearance. Also, in clinical studies [21, 22], reduced risk of respiratory infections was demonstrated in patients in the lateral position, who were turned intermittently from one side to the other. This evidence raised an argument against the rationale of the SRP in the prevention of VAP. Indeed, when the trachea is oriented above horizontal, aspiration of oropharyngeal secretions across the ETT cuff might be facilitated. In humans, the lateral Trendelenburg position (LTP), which is similar to the recovery position, is a feasible option to reproduce the aforementioned interventions.

Theoretically, the LTP could exceed the preventive benefits of the SRP, because pulmonary aspiration, which is the primary mechanism for the development of VAP, would be hindered. In particular, we hypothesized that the LTP would prevent aspiration of bacteria-laden oropharyngeal secretions, and associated pulmonary infections, specifically in critically ill sedated patients, unable to clear respiratory secretions. Therefore, we undertook this translational clinical trial to determine whether the LTP would decrease the incidence of microbiologically confirmed VAP and improve other significant clinical outcomes, in comparison with the SRP.


We conducted a prospective randomized controlled trial in 18 hospitals (Table E1, electronic supplementary material). Patients 18 years old of age or older, expected to be on MV for at least 48 h were assessed for enrollment, within 6 h from endotracheal intubation. Additional detail on inclusion and exclusion criteria is provided in the electronic supplementary material methods. In accordance with local ethical regulations, informed written consents were obtained from the patients or their next of kin. The trial was overseen by an independent data and safety monitoring (DSMB) committee. At the first feasibility/safety interim analysis, the protocol was amended to allow inclusion of patients within 12 h of tracheal intubation. Patients were randomized to be positioned in the LTP, as reported in Fig. 1, or in SRP, with the head of the bed elevated at least 30° (see also electronic supplementary material). Investigators at each institution were blinded to the randomization block length, and laboratory microbiologists were blinded to group assignments. A comprehensive 1-day educational workshop was held before enrollment. In all patients, we used active humidification of respiratory gases. We encouraged study positions until extubation or 14 days of MV, and to not increase sedation to ameliorate compliance. Patients were assessed daily for potential lightening/interruption of sedation, weaning from MV, and compliance with the LTP (Fig. 2). Any change in position and bed angulation was recorded. VAP preventive protocols, implemented prior to the beginning of the trial, were not modified. VAP was clinically suspected in patients with a new pulmonary infiltrate, at least 48 h of MV, and two of the following clinical signs: body temperature greater than 38 °C or less than 35 °C; leukocyte count above 10 × 103 or below 4 × 103 per cubic millimeter; purulent tracheal secretions [23]. In patients who presented pulmonary infiltrates upon randomization, only new pulmonary infiltrates, developing after 48 h of MV were considered pathognomonic of VAP. Study primary outcome was incidence of VAP (within the first 14 days), confirmed by quantitative bronchoalveolar lavage (BAL) or mini-BAL cultures at least 104 colony-forming units (cfu)/ml. Secondary and tertiary outcomes included intensive care unit (ICU) and hospital stays; duration of MV; all-cause mortality at 28 days, in the ICU and hospital; use of sedatives, analgesics, and antibiotics. We computed daily defined dose of antibiotics (DDD) [24], lorazepam and morphine equivalents [25], and propofol dosage. Any adverse (AE) and serious adverse event (SAE) was recorded. Finally, nurses’ feasibility and workload were assessed. After inclusion of 137 patients, the DSMB recommended additional neurological evaluation (see methods in the electronic supplementary material).

Fig. 1
figure 1

The lateral Trendelenburg position. Patients were positioned in a semilateral position, similar to the recovery position, with the head of the bed tilted 5–10° in Trendelenburg position. An imaginary line from the sternal notch to the mouth, passing through the middle of the trachea, was used as a surface landmark. The patient was positioned to maintain this line slightly below horizontal to ensure an orientation of the trachea and endotracheal tube 2–5° below horizontal. Every 6 h, the patient was turned from one side to the other

Fig. 2
figure 2

Daily evaluation for patient eligibility to be awaken and extubated. It was strongly recommended to perform, on a daily basis, a comprehensive evaluation of the patient’s need of sedation, tolerance to the lateral Trendelenburg position (LTP) and potential discontinuation of invasive mechanical ventilation. Awaken patients, randomized to the LTP, were allowed to be kept in semirecumbent position (SRP), whenever they did not tolerate the LTP

Statistical analysis

Full description of statistical analyses is provided in the electronic supplementary material. We calculated that 400 patients for each group should have been enrolled to detect a reduction in primary outcome from 15.0% to 7.5%, for a statistical power of 90%, with a two-sided alpha significance level of 0.05. We conducted an intention-to-treat analysis, on the basis of the preplanned statistical analysis plan and in accordance with the CONSORT statement [26]. Fisher’s exact test and Wilcoxon rank-sum tests were used to analyze categorical and continuous variables, respectively. We estimated the measure of effect of the intervention by calculating unadjusted relative risk (RR) and its 95% confidence interval (CI). Also, we used competing risk analysis [27] to assess the effect of body positions on VAP incidence. We estimated the effect of the intervention using cause-specific hazard ratios (HR), through the Cox proportional hazards model. As for continuous secondary outcomes, risk differences were estimated by the median of all paired differences between groups and their 95% CI (Hodges–Lehmann method) [28]. We used the log-rank test to compare Kaplan–Meier curves of 28-day survival and we estimated the 28-day HR using a Cox proportional hazards model. We used complier average causal effect (CACE) analyses to estimate effect of the interventions, while accounting for protocol adherence and respecting the intention-to-treat principle [29]. Finally, after a comprehensive evaluation of the findings of aforementioned analyses, the steering committee agreed to perform a single post hoc analysis, clustering the patients on the basis of the presence or absence of pulmonary infiltrates at baseline. The rationale for this analysis was based on the well-recognized challenges in the diagnosis of VAP, through standard clinical variables and chest radiographs. Indeed, a new pulmonary infiltrate could be difficult to detect in patients with an abnormal chest radiograph at baseline. Post hoc analyses were conducted through the Mantel–Haenszel test for homogeneity for binary outcomes and by modelling a negative-binomial regression model for duration of MV and ICU or hospital stay. All tests were two sided, and there was no adjustment for multiplicity. A p value of 0.05 was considered as statistically significant. Stata 13.1 was used for all analyses.


Between December 2, 2010 and April 20, 2015, we screened 2156 patients in 18 ICUs (Fig. E1 and Table E1, electronic supplementary material). Of these patients, 401 were enrolled; of whom, five later withdrew from the study. One patient in the semirecumbent group was erroneously included twice; thus, only the first randomization was included. This resulted in an intention-to-treat population of 395 patients (194 in the lateral Trendelenburg group and 201 in the semirecumbent group) (Fig. 3). Baseline characteristics of the patients are reported in Table 1.

Fig. 3
figure 3

Flow diagram. We assessed 2156 patients for eligibility. Of these patients 1755 were excluded: 795 did not meet the inclusion criteria; of the remaining 1361 who met the inclusion criteria, 823 met exclusion criteria, and 137 refused to participate. We enrolled and randomly assigned the remaining 401 patients to be positioned either in the lateral Trendelenburg position (198 patients) or semirecumbent position (203 patients), including five patients who later withdrew from the study. One patient in the semirecumbent group was erroneously included twice into the trial; thus, only the first randomization was included. Thus, the primary outcome was available for 395 (98%) patients: 194 in the lateral Trendelenburg group and 201 in the semirecumbent group

Table 1 Characteristic of the patients at baseline and ventilator-associated pneumonia preventive measures

Recruitment was stopped in May 2015, at the second interim analysis. The DSMB reviewed the results of 395 patients and recommended to stop the study for low incidence of VAP in the control group, lack of benefits in any major secondary outcome, and adverse events in the LTP group (see electronic supplementary material). The trial steering committee adopted the DSMB advice and stopped the trial.


Overall, 23 (11.9%) and 7 (3.5%) of the patients, in LTP and SRP, respectively, were never managed in the randomized position (p < 0.01). The median time to the first positioning was delayed in the LTP, in comparison with the SRP, 3 (IQR 1–7) and 0.3 (IQR 0–2) h, respectively (p < 0.01). Patients in the LTP were appropriately positioned for a 32.7% (median) of the study time (IQR 15.4–57.8%); whereas, patients in the SRP were correctly positioned for 93.6% (IQR 89.0–96.3%) of the time. In hospital with a primary focus on research, patients in the LTP were appropriately positioned for 30.4% of the study time (IQR 15.4–54.1%); whereas, patients in general hospitals were appropriately positioned for 35.9% (IQR 16.1–63.0%, p = 0.51). Crossover never occurred in the SRP group; conversely, it occurred in 160 LTP patients (93.6%). Reasons for changing from the LTP to SRP are depicted in Fig. E2. In compliance with the protocol depicted in Fig. 2, crossover increased after a few days of MV. Indeed, as reported in Fig. E3-A in the electronic supplementary material, median fraction of time in LTP reached 51.8% (IQR 20.7–79.2) during the first 2 days, then progressively decreased, as sedation was lightened (Fig. E3-B). Overall, bed angulation in the LTP was −5.6° ± 2° and 34.1° ± 6° in the SRP position (p < 0.01).

Primary outcomes

The incidence of microbiologically confirmed VAP was 0.5% (1/194 patients) in patients positioned in LTP, and 4.0% (8/201 patients) in patients in SRP (relative risk 0.13, 95% CI 0.02–1.03; risk difference −3.5%, 95% CI −6.4 to −0.6; p = 0.04) (Table 2). Among all patients with clinical suspicion of VAP, 32 underwent distal pulmonary sampling; in particular, 18/21 (85.7%) of the SRP patients and 14/18 (77.8%) of the LTP patients (p = 0.68). Antibiotics were inappropriately changed, before distal sampling, in three patients of each group (p = 0.77). Overall, 26 BALs and six mini-BALs were performed. The cumulative probability of VAP was lower in the LTP than in the SRP group (Fig. 4a), cause-specific HR was 0.13 (95% CI 0.02–1.00, p = 0.05) (Table E2). The most common VAP causative pathogens in BAL fluids were Staphylococcus aureus and Klebsiella pneumonia (Table E3).

Table 2 Main analyses of primary and secondary outcomes
Fig. 4
figure 4

Incidence of ventilator-associated pneumonia and 28-day survival. Cumulative incidence of ventilator-associated pneumonia from random assignment to day 14 (a) and Kaplan–Meier plot for probability of survival from random assignment to day 28 (b) in patients allocated to the lateral Trendelenburg position group (red line) and the semirecumbent position group (black line)

Secondary outcomes

We did not find any between-group difference in ICU or hospital mortality rates and length of stay on MV (Table 2; Fig. 4b). Multiple organ failure was the most common cause of death in ICU, without difference between groups (Table E4). Table E5 depicts CACE analysis of primary and secondary outcomes. Importantly, a risk difference of −7.6% (95% CI −14 to −1.1%, p = 0.02) in microbiologically confirmed VAP was found in favor of the LTP. In the CACE analysis, we did not find any significant difference in mortality rates.

Frequency of bacteremia or other potential infections did not differ between groups (Table E6). There were no statistically significant differences between groups in the duration of treatment, nor total doses of sedatives or opioids (Table E7). Likewise, median consumption of antibiotics was similar, 10.2 (IQR 2–23.1) and 8 DDDs (IQR 2–22) in the LTP and SRP group (p = 0.53) (Table E8). We did not find any difference in the use of enteral or parental nutrition and calorie intake (Table E9). Nurses reported greater difficulties in positioning the patient in LTP and higher workload (Table E10). However, approximately 50% of the patients were easily or very easily positioned in LTP. Finally, these challenges slightly ameliorated through practice, as more patients were enrolled (Fig. E4).

Adverse events

Rate of AEs was comparable among the two groups, except for vomiting that was more frequent in LTP (8.3% vs 2.5% in SRP, p = 0.01) (Table 3). Among patients who were placed in the randomized positions, no SAEs were reported in the SRP; whilst, six SAEs were described in LTP patients (0/194 vs. 6/171, respectively, p = 0.01), of whom four had pre-existing liver diseases (Table 3). Among these events, four were definitely related to the LTP—mainly transient oxygen desaturation and hemodynamic instability—while two events (intracranial hemorrhage and brachial plexus injury) were possibly related. Neurological evaluation, upon discharge from the ICU, was comparable between groups (Table E11).

Table 3 Adverse events

Post hoc analysis

Post hoc interaction exploratory analyses showed that incidence of clinically suspected VAP was lower in LTP patients without pulmonary infiltrates at the time of enrollment, in comparison with SRP, but higher in patients with pulmonary infiltrates (Table E12) (RR with pulmonary infiltrates 1.69, 95% CI 0.71–4.04; RR without pulmonary infiltrates 0.42, CI 0.12–1.23, p = 0.04 for interaction).


In our study, the incidence of microbiologically confirmed VAP was lower in LTP, in comparison with the SRP. Yet, the DSMB suggested early trial termination because of low incidence of VAP, lack of benefits in secondary outcomes, and serious adverse events in LTP patients. Therefore, as a result of the aforementioned limitations and inconclusiveness of the findings, the LTP cannot be recommended as a VAP preventive measure until further investigation is available.

In our analysis, we primarily focused on the first 2 weeks of MV. Previous studies [30,31,32] have consistently corroborated that patients are at higher risk of VAP during this period. Additionally, as reported in Table 2, the median duration of MV was approximately 5 days, highlighting the importance of close monitoring and preventive measures during the first week. We did find, in LTP patients, a decrease in VAP, but as a result of the very low incidence, the results are questionable. Such low incidence is in line with other previous trials [30], and might be related to the consistent decline in VAP rates reported in the last decade [3, 4]. Also, all local infection control measures were not modified, and possibly efficiently implemented. Nonetheless, it is important to emphasize that in our population the median time on MV was 4–5 days. In this context, the incidence density of microbiologically confirmed VAP, in the SRP and LTP groups, was 7.19 and 0.12 per 1000 ventilator days, respectively (p = 0.02). Thus, the incidence density in the SRP was comparable to the rates reported in other studies from developed countries [1]. We found comparable clinically suspected episodes of VAP between groups. This could be related to higher risks of atelectasis in LTP, which potentially encouraged physicians to perform BALs in patients with unspecific clinical signs of respiratory infections.

As reported in Table 2, mortality in the LTP was higher than in the SRP group, although the difference was not statistically significant. Our study was specifically aimed at evaluating feasibility and efficacy of the novel intervention, and it was likely underpowered to detect differences in mortality. Yet, these data call for a comprehensive assessment of the safety of the intervention, particularly considering the apparent risk of mortality in patients with pulmonary infiltrates upon intubation (Table E12). Also, SAEs were reported in LTP patients, mainly brief episodes of respiratory and hemodynamic instability, which often occur during routine patient mobilization, specifically in patients who presented ascites. Yet, these differences in SAEs between groups are troubling and suggest potential safety issues. Finally, a slight trend toward higher risk of other infections was reported in LTP patients (Table E6). Potential reasons for lack of benefits on mortality, higher risk of SAEs, and development of infections are difficult to discern and, at this time, can only be speculative. For instance, LTP patients were mobilized more frequently and contralateral propagation of noxious biofluids could have been promoted [33]. Also, frequent manipulation of artificial airways and indwelling devices could have increased risks of nosocomial infections. Nevertheless, on the basis of the results from animal studies [17,18,19], LTP patients could have also benefited from the increased clearance of pulmonary biofluids. It should also be considered that in LTP patients, with severe lung injury, ventilatory management might have been challenging. Indeed, we could speculate that in LTP abdominal contents were pushed upward against the diaphragm and pulmonary atelectasis could have developed, increasing risks of lung injury. We did not observe augmented neurological risks in LTP; yet, it would also be advisable to further explore short- and long-term neurological effects. Finally, higher rates of vomiting did not increase the risk of VAP in LTP, confirming prevention of gastro-pulmonary aspiration.

As for the compliance to the novel intervention, the study was designed to promote LTP, specifically in patients requiring higher level of sedation—i.e., during first hours of MV and with higher severity of critical illness—and at higher risk of pulmonary aspiration. Therefore, the highest compliance was achieved in deeply sedated patients, during the first days of MV (Fig. E3). Then, as expected, compliance decreased when the patient progressed through the weaning period. As a result, crossover from the LTP to SRP occurred in more than 90% of the patients; whereas, SRP patients were maintained in the same position, throughout the study time. Of note, hemodynamic and respiratory instability was the reason for changing from LTP to SRP in more than 30% of the cases (Fig. E2). This indicates potential safety risks that should be further explored to identify populations at risk. At times, the nurses encountered difficulties executing the LTP and a few factors should be considered when interpreting these findings. First, patients in the SRP are mobilized less, causing better acceptance of the intervention. Indeed, recommended lateral rotation every 2–4 h is seldom implemented [34], and patients are repositioned only when they slide down the bed [35]. Second, in the SRP surgical drains, tubes, and vascular lines can be easily accessed. Finally, in ICU settings concomitant procedures and understaffed nursing shifts limit applicability of rigid schedules for patient positioning.

In comparison with previous randomized trials on body position [14], this work presents several strengths. First, this is the largest clinical trial that evaluated the effects of body positions in the prevention of VAP. Second, we used invasive methods to increase specificity of VAP diagnosis. Indeed, we expected that LTP amplified gravity-driven clearance of mucus and risks of tracheal colonization. Therefore, culture of endotracheal aspirates would have overestimated VAP incidence. Third, according to the latest statistical standards in this field [36], we used competing risk analysis to adjust for duration of MV and death. Fourth, data were analyzed according to the intention-to-treat principle and CACE analysis, and the robustness of this latter analysis should be noted [37]. Indeed, in randomized trials testing new interventions, non-adherence could occur in a nonrandom fashion and differ among groups. In this case, the traditional “per-protocol” approach would generate biased results. Conversely, CACE analysis account for adherence, while preserving the intention-to-treat principle, and generates more accurate estimations of efficacy and safety.

Yet, our study has various limitations. Only microbiologists, assessing the primary outcome, were blinded to group assignments, because blinding the attending physician was impossible. Second, as reported in Fig. 3, approximately 60% of the screened patients met exclusion criteria. This could have caused selection bias and reduced generalizability of our findings. Nevertheless, several exclusion criteria were specifically listed to ensure patient safety. Third, continuous computerized monitoring of body position was not employed [16]; hence, changes in position might have been underreported. Fourth, VAP cases were not subject to central, independent adjudication, by a committee blinded to the study group. This rigorous adjudication would have likely strengthened the reliability of the diagnostic process. Yet, we used strict microbiological methods to ultimately diagnose VAP. Fifth, as it could be inferred by Table 2, we enrolled a population specifically at risk of early-onset VAP. Thus, the LTP might have promoted clearance of pathogens aspirated at intubation, or early thereafter. However, the effects of the LTP on VAP prevention cannot be straightforwardly extrapolated to the population at risk of late-onset VAP. Sixth, assessment of neurological complications upon discharge was suboptimal and only initiated after DSMB recommendation. Also, we did not closely monitor agitation/delirium; therefore, a comprehensive evaluation of neurological effects of LTP should be undertaken in future studies. Finally, premature termination of our trial has limited inferences regarding VAP preventive efficacy.


In a population at low risk of VAP, LTP caused a reduction in microbiologically confirmed VAP, but the results are inconclusive because of the lack of other clinical benefits, increased safety risks, and challenges in nursing compliance. Given the limitations of our study, potential future clinical studies should carefully re-examine the targeted population, VAP diagnostic methods, outcomes, and incidence of nosocomial infections among collaborating centers. Finally, considering that in our study the LTP was commonly applied during the first days of MV, prolonged use of such a position seems futile and at risk of adverse events.