Protocol for a systematic review and individual patient data meta-analysis of benefit of so-called lung-protective ventilation settings in patients under general anesthesia for surgery

Search strategy

Published and unpublished observational studies and randomized controlled trials were identified by two previous systematic searches of the literature by our group [4, 5]. The sensitive search strategies followed Medical Subject Headings and Keywords ([protective ventilation OR lower tidal volume OR low tidal volume]). We will perform a new search strategy to assess studies dealing only with the effects of PEEP. Thus, we will include the following Medical Subject Heading and Keywords ([protective ventilation OR lower tidal volume OR low tidal volume OR positive end-expiratory pressure OR positive end expiratory pressure OR PEEP]). Two authors will perform a computerized blinded search of MEDLINE, Cumulative Index to Nursing and Allied Health Literature (CINAHL), Web of Science, and Cochrane Central Register of Controlled Trials (CENTRAL). The list of studies and trials will be updated to identify studies and trial published after the original searches, and thus updated to August 2013.

Selection of studies

All observational studies and randomized controlled trials of protective ventilation that used outcomes relevant for this meta-analysis will be screened for inclusion. Key inclusion criteria are: 1) lower versus higher tidal volume in each arm, or higher versus lower PEEP in each arm, or both, 2) age > 18 years; 3) patients undergoing any surgical procedures under general anesthesia and mechanical ventilation; and 4) patients without acute respiratory distress syndrome (ARDS) at the onset of mechanical ventilation. Studies overlapping with other studies as indicated by the corresponding author will be excluded. There is no restriction in study design or language. Studies in patients in the ICU are also excluded.

Methodological quality assessment

Two investigators will carry out data extraction and quality assessment from all the retrieved published studies based on the full text articles. Discrepancies will be resolved by consensus. In randomized controlled trials, we will assess allocation concealment, baseline similarity of groups (with regard to age, and severity of illness), early stopping of treatment, and loss to follow-up. Also, Jadad score will be used to assess the quality of the randomized controlled trials.

Bias

We expect to obtain at least 90% or more of individuals analyzed in the studies identified in the search strategy. With this number of patients it will be possible to avoid the bias due to selective availability of study data. Sensitivity analysis combining the results of any unavailable studies (as extracted from publications or obtained in tabular form) and comparing these with the main individual patient data results will be used to interpret the data.

Collection of individual patient data

Corresponding authors of the identified eligible published studies and trials will be contacted via email with a cover letter, or in a personal conversation, detailing the objectives of the collaborative meta-analysis, background information, and a datasheet for input of individual patient results for the project. The cover letter and the datasheet are shown in Additional file 1. The filled-out data templates will be sent back to the corresponding author and further communication will be mainly by email. Corresponding authors will also be contacted about unpublished data to enlarge the clinical data pool.

Data management, security and validation

The same two investigators who perform the electronic database search will also collect and assemble individual patient data provided by the investigators. Data will be accepted in any kind of electronic format (SPSS, STATA, Word document, Excel document, and Access document) and only the coordinators of the collaboration will have direct access to the data. Both investigators will perform the data validation, checking the received data set for data entry mistakes and inconsistency. Differences will be discussed and settled in consensus.

Mechanical ventilator parameters

The corresponding authors of studies and trials will be asked to fill the datasheet with mechanical ventilation parameters (plateau pressure, peak pressure, PEEP, respiratory rate, inspired fraction of oxygen (FiO₂), and minute ventilation) and oxygenation parameters (partial pressure of oxygen (PaO₂), partial pressure of carbon dioxide (PaCO₂), pH, and the PaO₂/FiO₂ ratio) obtained hourly during the procedure.

Analysis plan

The primary outcomes will be the occurrence of ARDS, according to the definition used by the authors, and the composite of ARDS development and in-hospital all-cause mortality. Secondary clinical outcomes include: 1) in-hospital mortality, defined as any death during hospital stay; 2) duration of mechanical ventilation, defined as the time since initiation of mechanical ventilation and successful discontinuation; 3) length of stay in the ICU, defined as the time from ICU admission to ICU discharge or death; 4) length of stay in hospital, defined as the time from hospital admission to hospital discharge or death; 5) occurrence of any intra-operative complication; and 6) occurrence of any postoperative pulmonary and extra-pulmonary complication, including pulmonary infections, according to the definition as used by the authors, need for red blood cell transfusions, defined as the amount of red blood cells in ml used during the follow-up, and need for fresh frozen plasma transfusion, defined as the amount of fresh frozen plasma in ml used during the follow-up. Secondary laboratorial outcomes will include: 1) levels of plasma IL-6, IL-8, IL-10, and TNF-α in pg/ml; and 2) levels of bronchoalveolar IL-6, IL-8, and TNF-α in pg/ml. Finally, safety outcomes will include: 1) lowest PaO₂ in mmHg; 2) lowest and highest PaCO₂ in mmHg; 3) worst PaO₂/FiO₂ ratio; 4) lowest and highest pH; and 5) incidence of acidosis, defined as pH <7.35.

Completeness of data

Careful evaluation will be performed to ensure completeness of data and to check consistency. Since some authors may not have recorded ventilatory parameters hourly, we will divide the measurements into three periods: 1) beginning of the surgery, defined as the parameters measured in the first hour of the procedure; 2) middle of the surgery, defined as the parameters measured closest to the middle of the procedure (total time of procedure divided by two); and 3) end of the surgery, defined as the parameters measured in the last hour of the procedure. Since parameters at the end of the study can suffer influence of any lung injury developed during surgery and parameters at the beginning of surgery do not have sufficient time to induce changes in the lung, we will use the parameters in the middle of the surgery in the outcome analyses.

Model selection

A multivariate model will be constructed for discrimination of the effects of lower tidal volume from those of higher levels of PEEP. The initial model will include age, gender, type of study, body mass index, type of surgery, (ASA) (American Society of Anesthesiology score), type of ventilation, highest PEEP used during surgery, highest plateau pressure achieved during surgery, highest compliance achieved during surgery, transfusion of red blood cells in the perioperative period, and risk factors for ARDS. Variables with P < 0.2 in the univariate analysis are included in the multivariate regression. The final model will be developed by dropping each variable in turn from the model and conducting a likelihood-ratio test to compare the full and the nested models. We will use a significance level of 0.05 as the cutoff to exclude a variable from the model. Finally, the variables of tidal volume (protective versus conventional in an intention-to-treat (ITT) analysis and ≤7 ml/kg predicted body weight (PBW) versus 7 to 10 ml/kg PBW versus >10 ml/kg PBW in a per protocol analysis) and PEEP (protective versus conventional in an ITT analysis and ≤5 cmH₂O versus >5 cmH₂O in a per protocol analysis) will be added to the model in order to test the resultant model against that without the variable. We will construct Kaplan-Meier curves and use log-rank test to determine the univariate significance of the study variables [9].

Statistical analysis

Baseline characteristics of patients will be presented separately for each trial and overall. Continuous variables will be presented as mean ± SD or median and interquartile if not normally distributed. Binary and categorical variables will be presented as frequencies and percentages.

All analyses will be conducted in two ways: 1) an ITT analysis, where patients are analyzed according to the group of tidal volume they had in the original trial, taking into account only the randomized controlled trial; and 2) a per protocol analysis, where patients are analyzed according to the tidal volume they really received, taking into account the randomized controlled trial and observational studies. In ITT analyses the patients will be divided into two groups of tidal volume (protective versus conventional), and in per protocol analyses the patients will be divided into three groups of tidal volume (≤7 ml/kg PBW versus 7 to 10 ml/kg PBW versus >10 ml/kg PBW).

Time-to-event is defined as time from the day of surgery to the event. We will use a Cox proportional hazards regression model to examine simultaneously effects of multiple covariates on outcomes, censoring a patient’s data at the time of death, hospital discharge, or after 30 days [10]. In all models, the categorical variables will be tested for trend with the large tidal volume as reference and the proportional hazards assumption will be assessed. A test for interaction between pairs of variables in the final model will be performed. The effect of each variable in these models is assessed with the use of the Wald test and described by the hazard ratio with a 95% confidence interval. Binary outcomes will be analyzed by the chi-square test and by logistic regression including the same set of covariates as for the Cox proportional hazards regression model.

Restricted cubic spline analysis will be used to characterize the dose–response relationship between median tidal volume in ml/kg PBW and ARDS development, while adjusting for the same set of covariates as used in the final Cox model. A cubic or quadratic term will be used in the final model. Time-course variables (for example, repeated measures of ventilatory parameters, vital signs, oxygenation parameters and so forth) will also be analyzed by a linear mixed model. The linear mixed models procedure expands the generalized linear model so that the data are permitted to exhibit correlated and non-constant variability. The model includes two factors: 1) study group (fixed factor, defined as protective or conventional in the ITT analysis and tidal volume ≤7 ml/kg PBW versus 7 to 10 ml/kg PBW versus >10 ml/kg PBW in the per protocol analysis); each level of the study group factor can have a different linear effect on the value of the dependent variable; and 2) time as covariate; time is considered to be a random sample from a larger population of values, the effect is not limited to the chosen times.

Subgroup analyses will be used to assess the effects of tidal volume and PEEP in the following pre-specified subgroups: 1) type of surgery (cardiac, abdominal, thoracic, and orthopedic); 2) study design (randomized versus non-randomized controlled trial); 3) ASA score (<3 versus ≥3 and for each level); 4) presence of risk factor for ARDS; 5) mode of ventilation (volume versus pressure controlled); 6) age (<65 versus ≥65 years); 7) gender (male versus female); 8) PEEP during surgery (<5 versus ≥5 cmH₂O); and 9) body mass index (<17, 18 to 25, 26 to 30, 31 to 35, and >35 kg/m²). For data not normally distributed, analyses will be performed after log₁₀ transformation to permit the use of parametric statistics. If the data still differ significantly from normal even after log₁₀ transformation, these data will be analyzed by non-parametric tests.

All analyses will be conducted with Review Manager v.5.1.1 (The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark), SPSS v.20 (IBM Corporation, New York, USA) and R v.2.12.0 (R Foundation for Statistical Computing, Vienna, Austria). For all analyses two-sided P values <0.05 will be considered significant.

Publication policy

This protocol is not registered in the PROSPERO. The results of this meta-analysis will be sent for publication in a peer-reviewed journal and all collaborators will be included as co-authors of the paper.