To perform the systematic review on the 1-year period of 2013, we followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analysis) statement guidelines .
Our outcome classification was developed according to previous work on patient-important outcome in various medical domains [3, 13, 15].
A scientific committee [including three intensivists (SG, J-DR and DD) and 1 methodologist (FT) particularly involved in designing and conducting RCTs in critically ill patients] established a classification of outcome categories relevant to ICU trials.
The experts identified six outcome categories:
Patient-important outcomes that included two entities: on the one hand, mortality at any time and on the other, quality of life, functional/cognitive/neurological outcomes assessed after ICU discharge.
Clinical outcomes in ICU and hospital organ failure, complication/adverse outcomes (for instance: drug induced skin reaction or hypotension during renal replacement therapy), healthcare-associated outcomes (nosocomial pneumonia, catheter-related infections), delirium, clinical events (such as venous thromboembolism, myocardial infarction), pain (in ICU), anxiety (in ICU), conscience level, return to spontaneous circulation, muscle strength/circumference, sleep duration, National Institute of Health Stroke Score (NIHSS) (for acute phase of stroke), clinical response to antibiotics, dyspnea (in ICU), noninvasive ventilation tolerance.
Biological/physiological/radiological outcomes such as brain natriuretic peptide (BNP), neutrophil gelatinase-associated lipocalin (NGAL), total lung capacity, chest X-ray severity score.
Care provider decision-related outcomes e.g., mechanical ventilation duration, length of stay, antibiotic exposure, volume of fluid resuscitation, intubation or reintubation, number of gastric tubes for aspiration, sedation exposure (dose/time), renal replacement therapy, ICU readmission, noninvasive ventilation, tracheostomy, transfusion, use of a prokinetic agent, need for surgery, dose of local anesthesia, hospital discharge disposition.
Care performance outcomes care procedure quality and noise/light exposure.
Other outcomes family satisfaction, physician/nurse or other provider satisfaction, cost/charges, withholding/withdrawal of care, patient judgment about his readiness to discharge, workload for staff team, compliance to a care protocol, medicolegal conflict.
Besides, for primary outcomes, we defined a “surrogate outcome” as an outcome measuring a substitute for some other variable (e.g., a biomarker intended to substitute for a clinical endpoint) .
Article eligibility criteria
Articles were eligible if they met the following criteria: article published between January 2013 and December 2013; reporting an RCT involving critically ill adult patients (i.e., adults hospitalized in ICU); written in English.
We considered only the first report of the trial results and trial extension follow-up, i.e., we excluded articles reporting post hoc analyses and sub-analyses of RCTs. Indeed, our aim was to focus on the RCTs’ initial objective. We also excluded phase II studies, pilot studies and physiological crossover studies because studies at this stage of clinical research are expected to explore mainly physiological and feasibility outcomes.
Search strategy and article selection
Main literature search (for the January 2013 to December 2013 period) was conducted on the July 16, 2014, in MEDLINE (via Pubmed®) to identify eligible articles indexed between January 2013 and December 2013. The search strategy relied on two algorithms, one dedicated to articles indexed with Mesh terms and the other dedicated to articles not indexed (at the time of the search), using exclusively free text. Terms related to intensive care were combined with terms related to RCTs. Details regarding the literature search strategy and the terms used are provided in Additional file 1.
Two senior intensivists (SG and JM) independently screened the titles and abstracts for the eligibility criteria, to identify articles to be read in full text. Definite article selection was only achieved after examination of the full text confirmed that inclusion criteria were met.
A standardized extraction form (available from the corresponding author) was established from a literature review and a priori discussion. This extraction form was pretested by two authors (SG and JM) independently, in a set of ten articles. This test enabled to identify items needing rewording to avoid any confusion. Disagreements were discussed with an epidemiologist (last author, FT), to ensure similar understanding. Once all litigious points were settled, two reviewers (SG and JM) independently extracted the following data from the selected articles (using the full text and the Additional file 1): general data (funding source, geographical origin, topic, number of centers), methods (intervention assessed, study design, randomization design), quality assessment (by use of the risk of bias tool ), trial characteristics (inclusion period, length of follow-up, number of randomized patients) and outcomes (time from randomization to assessment for primary outcome, type and characteristic of all outcomes, see paragraph above).
For all articles included in the systematic review, disagreements between the two reviewers (SG and JM) were resolved by consensus. In case of persistent disagreement, arbitration by a third reviewer (FT) settled the discrepancy.
A 1-year time frame was chosen because it yielded a convenient study sample. Because of the significant lag in study indexation, the closest complete year available at the time of the literature search was 2013.
Continuous variables are described with median and interquartile range (IQR). Categorical variables are described with frequencies and percentages. Distribution of outcomes into the six categories (patient-important outcomes, clinical outcomes, bio/physio/radio outcomes, care provider decision-related outcomes, care performance outcomes and others outcomes) is presented as radar plot. For primary and secondary outcomes, distribution is presented for all outcomes and according to three major topics.
Trial characteristics associated with the presence of at least one patient-important outcome (primary or secondary outcome) were identified in univariate analysis, using Chi-square test or Fisher’s test for categorical variables and Student’s test or Wilcoxon’s test for continuous variables.
Inter-reviewer agreement was measured by the kappa statistic for the following categorical variables: funding source, geographical origin, intervention assessed, unit of randomization and primary outcome category.
Statistical analysis was performed with GraphPad Prism 5 (GraphPad Software, San Diego, USA) and SAS version 9.3 (SAS Institute Inc, Cary, NC).