Outcomes used in randomised controlled trials of nutrition in the critically ill: a systematic review

Relevant RCTs published between January 2000 and August 2018 were identified by electronically searching PubMed/MEDLINE (1 January 2000–31 August 2018) and the Cochrane Central Register of Controlled Trials (CENTRAL) (2000–2018). The following MeSH and free text terms were used: “nutrition”, “feeding”, “protein-calorie”, “nutritional support”, “critical care”, “critically ill”, “critical illness”, “ICU”, “adult”, “randomized controlled trial”, “controlled clinical trial”, “randomized”, and “trial”. The search was limited to articles written in English or French. The reference lists of retrieved articles were screened for additional relevant articles. Unpublished relevant RCTs were identified by searching the ClinicalTrials.gov registry using the following criteria: adult (18–65) and senior (66+); from 1 January 2000 to 31 August 2018; completed or terminated or recruiting or not yet recruiting; interventional studies (clinical trials); and the key words nutrition, feeding, protein-calorie, nutritional support, critical care, critically ill, intensive care, ICU, critical illness, and randomized. The search strategy is detailed in the online-only supplement (Additional file 1).

Among articles retrieved using the aforementioned search strategy, we selected those meeting the following criteria: RCT, including critically ill adults (age > 18 years), evaluating a nutritional intervention, and published after 1999. We excluded neonatal or paediatric intensive care studies, studies focussing primarily on the pharmaceutical properties of enteral nutrition (EN) or parenteral nutrition (PN) without supplying details on nutritional support, abstracts, case reports, review articles, and ancillary studies. Retrieved articles were first assessed by one of us (GT), who read the titles and abstracts to identify relevant trials. When the title and abstract were considered potentially relevant, the full text was retrieved by two of us (GT and MM), who worked independently of each other to apply the aforementioned criteria for inclusion in our study. Figure 1 is the flow chart of the retrieved and included articles.

For each study, the data were extracted by two of us (GT and MM, EP, or FT), who worked independently of each other. The data were recorded on a standardised form that was pre-tested by one of us (GT) on ten randomly selected articles and modified as needed. The following data were recorded: study characteristics (population, study design, blinding method, source of funding, number of centres, and geographic region); main study topic, recorded as clinical nutrition strategy, composition of nutritional support, or nutritional supplementation (see examples in Table 2); and primary and secondary outcomes. Outcomes were classified using predefined categories including mortality (at any time, ICU mortality, in-hospital mortality, day-28 mortality, day-90 mortality), length of stay (in the ICU and in hospital), duration of organ dysfunction (times on mechanical ventilation, catecholamines, renal replacement therapy; organ failures; and antibiotic therapy), ICU-associated complications (infections, metabolic complications, feeding intolerance), functional outcomes during the study period (muscle strength/walking distance tests, quality of life, physical function), and other (metabolic concentration, feeding measures, tube placement success rate, tube placement time, contamination of packs and glass bottles). Outcomes that were not described as primary or secondary were classified as undetermined. All recorded data were cross-checked after two of us read each study independently of each other. Discrepancies were resolved by discussion between the two readers or by adjudication by another of us (TL) if a consensus could not be reached. Data were entered into the study database using EpiData Software (EpiData Association, Odense, Denmark).

Metabolic concentration was defined as any outcome relevant to the blood level of a substance (e.g., a protein). Short-term and/or long-term metabolic complications were recorded as related to EN or PN. Infections were defined according to the International Sepsis Forum definition of infections in the ICU [11]. Feeding intolerance was defined as a complication due to the administration of nutritional preparations, such as diarrhoea or gastrointestinal symptoms. Feeding measures were all outcomes directly related to the nutritional intervention, such as feed volume and gastric residual volume. Physical function refers to any self-reported capabilities or physical performance. The capabilities assessed included upper limb function (dexterity), lower limb function (walking or mobility), spinal function, and instrumental activities of daily living.

The methodological quality of each RCT was assessed using the Cochrane risk of bias tool as indicated in chapter 8 of the Cochrane Handbook [12]. Methodological quality domains were those in the risk-of-bias table, i.e., randomisation and method of randomisation; allocation concealment; blinding of participants, healthcare providers, or outcome assessors; proportion of participants who did not complete follow up; partial reporting of outcomes; and any other areas of potential bias. The risk of bias in each RCT was classified as low, high, or unclear. RCTs were classified as having an overall high risk of bias if the risk of bias was high for at least one domain and low if the risk of bias was low for all domains.

The statistical analysis was performed using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA). Qualitative variables were described as numbers and percentages and quantitative variables as mean and standard deviation if normally distributed and median (25th–75th percentiles) otherwise. Ranges are also reported.

The initial electronic search identified 1135 articles. No additional articles were identified from other sources. Among the articles found on ClinicalTrials.gov, 250 duplicates were removed. Of the 885 remaining articles, 634 were excluded based on the title and abstract and 81 based on the full text (Fig. 1), leaving 170 RCTs for the study analysis (see Additional file 2).

Among the 170 selected RCTs, 142 (84%) were published in 36 different journals and 28 (16%) were registered on ClinicalTrials.gov. The median date of publication was 2010 (2004–2013), and most RCTs (n = 89, 52%) were conducted in Europe. Table 1 reports the main characteristics of the selected RCTs. Published trials collectively randomised 30,003 patients (median, 78 (37–157)). Only 54 (32%) RCTs involved multiple centres, with a median of 11 (4–18). The duration of the intervention was reported in 107/170 (63%) RCTs; median duration was 7 (5–14) days.

Table 2 depicts the distribution of the various topics of the selected RCTs. The most common topic category was clinical nutrition strategy (69/170, 40%), with strategies to optimise delivery and minimise the risk associated with EN being the most common example (n = 36). Clinical nutrition composition was the second most common category (66/170, 39%), followed by clinical nutrition supplementation (35/170, 21%).

The 170 RCTs used 24 different outcomes, which we classified into six categories: mortality, length of stay, duration of organ dysfunction, complications, functional outcomes, and other. A single primary outcome was described in 136/170 (80%) RCTs and at least one secondary outcome in 114/170 (67%) RCTs (two per study on average), whereas 34/170 (20%) RCTs did not specify whether the outcome was primary or secondary; we designated these unspecified outcomes “undetermined outcomes”.

Of the 136/170 RCTs with one primary outcome, complications represented the most common primary outcome (65/136, 48%). Mortality was the primary outcome in 17/136 (13%) RCTs. Six different mortality time points were used: in-ICU mortality, in-hospital mortality, day-28 mortality, mortality between day 29 and day 89, day-90 mortality, and mortality after day 90. The most common secondary outcomes were length of stay (90/114, 79%), mortality (82/114, 72%), duration of organ dysfunction (75/114, 66%) and complications (74/114, 65%). The primary outcome was functional in 2 (2/136, 1%) of the published trials and 3 (3/136, 2%) of the ongoing or future trials. Figure 2a illustrates the number of published studies by year from 2000 to 2018 according to primary outcome category, number of included patients, and result of the primary outcome. Complications and mortality were the most widely used defined primary outcome. Most of the trials with complications as the primary outcome included fewer than 500 patients, whereas most of the large trials used mortality as the primary outcome and found no significant effect of the tested nutritional strategy. Figure 2b depicts the distribution of topic categories by primary outcome according to the number of included patients. The studies including the largest numbers of patients mainly used length of stay and mortality, with clinical nutrition strategy as the topic. The use of complications as the outcome was fairly well balanced across topic categories. Figure 3 shows the primary outcomes in 28 protocols of future or ongoing (i.e., unpublished) RCTs registered on ClinicalTrials.gov, according to population size and nature of the nutritional intervention, with the goal of predicting the forthcoming results. Mortality is the primary outcome in the largest studies.

Figure 4 shows the risk-of-bias assessments for the individual domains of the Cochrane risk of bias tool. Absence of blinding of participants and personnel (staff) was the most common reason for a high risk of bias.