Omega-3 supplementation in patients with sepsis: a systematic review and meta-analysis of randomized trials

Eligible studies met the following criteria: (1) randomized clinical trial (RCT) study design; (2) the population involved adult patients in the intensive care unit (ICU) with sepsis or septic shock; (3) the intervention group received either enteral or parenteral supplementation with omega-3 fatty acids; (4) the outcomes included any of the following: mortality (using the longest available follow-up time), ICU length of stay (LOS), and duration of mechanical ventilation (DMV).

We searched MEDLINE, EMBASE, and the Cochrane Library from inception until December 2016. Our search strategies are given in Additional file 1: Tables S1–S3 and are limited to RCTs but not by language or publication date. We also screened the references from all included studies and relevant systematic reviews. Independently and in duplicate, two reviewers (CL and SS) screened titles and abstracts for eligibility, and conducted full-text reviews of selected studies. Disagreements over study selection were resolved by discussion and consensus. Studies fulfilling all of the eligibility criteria were included in the systematic review and meta-analysis.

Two reviewers (CL and SS) independently extracted data of interest from included studies, with disagreements resolved by discussion and consensus. Mortality was the primary outcome, and ICU LOS and DMV were secondary outcomes. When data were missing or unclear, we contacted study authors for clarification.

Using the Cochrane Collaboration tool [22], two reviewers (CL and SS) independently assessed each study for risk of bias in seven domains: random sequence generation, allocation concealment, blinding of patients and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other sources of bias. Disagreements were resolved by discussion and consensus, and adjudication by a third reviewer (WA) when necessary. For each study, the overall risk of bias was judged to be high if the risk of bias was high in any domain, unclear if the risk of bias was unclear in any domain (and not high in other domains), and low if the risk of bias was low across all domains.

All analyses were performed using RevMan software (Review Manager, version 5.3. Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014). We used inverse variance weighting and the DerSimonian and Laird [23] random-effects model to pool the weighted effect of estimates across studies. We reported relative risks (RRs) with 95% confidence interval (CI) for dichotomous outcomes and mean differences (MDs) with 95% CI for continuous outcomes. In studies where standard deviation (SD) for ICU LOS and DMV was not reported, we calculated SD from other measures of variability (standard error (SE), interquartile range (IQR), or 95% CI) following the methods suggested by the Cochrane Collaboration [24]. We combined low- and high-dose omega-3 intervention groups from one trial [25] into a single intervention group, using formulae described in the Cochrane Handbook [24] to calculate combined means and SDs for relevant outcomes. We assessed between-studies heterogeneity using Chi-square and I ² statistics, with significant heterogeneity defined as I ² > 50% or P < 0.10 [26].

We assessed publication bias for the mortality outcome by visual inspection of funnel plots. We investigated heterogeneity between studies by performing a post hoc subgroup analysis comparing parenteral with enteral administration of omega-3. We did not perform a subgroup analysis comparing risk of bias levels, as all studies had either a “high” or “unclear” overall risk of bias.

To explore the robustness of the results, we conducted the following post hoc sensitivity analyses: For all outcomes, we excluded trials that used per-protocol analysis, trials published in abstract form, trials that did not explicitly blind patients and healthcare personnel, trials that administered control formulae containing omega-3, and trials that did not administer the control group a placebo feed. For the mortality outcome, we excluded trials in which eligibility for inclusion was unclear (based on ICU admission), and further explored odds ratio (OR) as an alternative to RR analysis. For the ICU LOS outcome, we excluded trials that did not directly report SDs. Lastly for the DMV outcome, we excluded trials that did not directly report SDs and trials that did not stratify randomization by mechanical ventilation.

For each outcome of interest, we used the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach [27] to rate the quality of evidence based on risk of bias, indirectness, inconsistency, imprecision, and other factors. Indirectness was evaluated by reviewing each trial’s study population, intervention, control, and outcomes. Inconsistency was evaluated using between-trial Chi-square and I ² heterogeneity analyses. Imprecision was evaluated based on event rate, optimal sample size, and width of confidence intervals. The quality of evidence for each outcome was downgraded one level for “serious” limitations and two levels for “very serious” limitations.

Our search strategy identified a total of 175 citations, and 90 citations remained after removing duplicates. Screening of titles and abstracts led to the exclusion of 57 articles and the retrieval of 33 articles for full-text assessment, of which 16 were excluded for reasons outlined in Fig. 1. A total of 17 RCTs [25, 28‐43] met our inclusion criteria, representing 1239 critically ill patients with sepsis.

Seventeen studies enrolled patients diagnosed with sepsis or subsets of sepsis (early sepsis [36], abdominal sepsis [32], severe sepsis [34, 35, 38], and septic shock [35, 38, 41]). Three studies [31, 35, 37] required both sepsis and mechanical ventilation for inclusion. One study [32] enrolled ICU and post-anaesthesia care unit (PACU) patients with abdominal sepsis; we abstracted data for the ICU subset only.

Ten studies [29, 32‐34, 38‐43] used the parenteral route to administer omega-3 supplements, while seven studies [25, 28, 30, 31, 35‐37] used the enteral route. Brand-name formulae used included Omegaven [29, 32‐34, 38, 41], Oxepa [31, 36, 37], Impact [28, 30], and Lipoplus [43]. Available data on the contents of all brand-name intervention and control formulae are given in Additional file 1: Tables S4–S5. Six studies calculated daily dose by weight [29, 31‐33, 39, 40], five studies calculated daily dose by basal energy expenditure (BEE) and the Harris Benedict equation [28, 30, 35‐37], and five studies administered a fixed daily dose [25, 34, 38, 41, 42]; in one study [43], the dosing method was not specified. The duration of supplementation ranged between 4 and 14 days.

While most studies administered the control group a placebo solution using standard enteral or parenteral formulae without omega-3, four studies [25, 29, 33, 41] assigned controls to standard sepsis care only; two of these explicitly defined standard care according to 2008 Surviving Sepsis Campaign guidelines [29, 33].

For outcomes, all studies assessed mortality, twelve assessed ICU LOS, and seven assessed DMV. Eleven studies evaluated 28-day mortality [25, 31, 33‐36, 39‐43], and the remainder defined mortality as 60-day [37], in-hospital [33], ICU [32], or left parameters undefined [28, 30, 38].

Six studies [28, 32, 34‐36, 38] were appropriately blinded (patients, healthcare personnel, and research personnel), and seven studies [25, 29, 32, 33, 37, 39, 41] conducted a full intention-to-treat analysis of data. Six studies [28‐31, 35, 36] were industry-funded. Finally, one study [32] was published solely as an abstract, but the authors provided missing data via personal communication. Table 1 presents further details of eligible studies.

Using the Cochrane Collaboration tool for risk of bias [22], twelve studies were judged to be at high risk of bias, many of these due to attrition and performance bias. Risk of bias was unclear for the remaining five studies (Fig. 2).

Table 2 presents our GRADE [27] assessment of the quality of evidence by outcome. Evidence quality was assessed as moderate for mortality and very low for both ICU LOS and DMV outcomes.

Mortality was reported by 17 trials enrolling 1239 patients (Fig. 3). Omega-3 was not associated with a significant reduction in mortality (RR 0.85; 95% CI 0.71, 1.03; P = 0.10; I ² = 0%; moderate quality). ICU length of stay was reported in 12 trials enrolling 925 patients (Fig. 4). There was a significant reduction in ICU LOS (MD −3.79 days; 95% CI −5.49, −2.09; P < 0.0001, I ² = 82%; very low quality) for patients supplemented with omega-3. Duration of mechanical ventilation was reported in seven trials enrolling 495 patients (Fig. 5). There was a significant reduction in DMV (MD −2.27 days; 95% CI −4.27, −0.27; P = 0.03, I ² = 60%; very low quality) in the group of patients supplemented with omega-3.

Ten studies used parenteral administration and seven studies used enteral administration of omega-3. We performed a post hoc subgroup analysis comparing these subgroups for the outcome of mortality (Additional file 1: Figure S1) and found no significant differences in treatment effect (P = 0.97, I ² = 0%) between trials that used parenteral (RR 0.89; 95% CI 0.66, 1.19; P = 0.42; I ² = 0%) compared to enteral (RR 0.88; 95% CI 0.64, 1.21; P = 0.43; I ² = 35%) routes. However, the analysis is confounded by another major difference between subgroups: Most enteral formulations administered omega-3 in combination with other supplements (including arginine, selenium, and mRNA), while all parenteral formulations administered omega-3 as the sole supplement. Since the distinct influences of these two characteristics (additional supplementation and route of administration) on the treatment effect cannot be distinguished, no definitive conclusions can be drawn from this analysis.

Sensitivity analyses excluding trials that used per-protocol analysis [28, 30, 31, 34‐36, 38, 40, 42, 43] and trials that used control formulae containing omega-3 [31, 36] produced congruent results for all three outcomes. For the mortality outcome, exclusion of trials in which ICU admission was unclear [39], trials published in abstract form [32], trials without explicit blinding of patients and healthcare personnel [25, 29‐31, 33, 37, 39‐43], and trials that did not administer the control group a placebo feed [25, 29, 33, 41] yielded similar non-significant results.

However, for the ICU LOS outcome, exclusion of trials that did not explicitly blind patients and healthcare personnel to the intervention [25, 30, 31, 33, 37, 40‐43] rendered the significant reduction in ICU LOS non-significant (MD −3.63; 95% CI −7.85, 0.60; P = 0.09, I ² = 85%). Similarly, the significant reduction in DMV was countered by the exclusion of trials that did not explicitly blind patients and healthcare personnel [25, 30, 31, 37, 43] (MD −4.63; 95% CI −10.00, 0.75; P = 0.09, I ² = 70%), trials that did not stratify mechanically ventilated patients during randomization [25, 30, 36] (MD −1.78 days; 95% CI −4.39, 0.83; P = 0.18, I ² = 61%), trials that did not administer the control group a placebo feed [25] (MD −2.10 days; 95% CI −4.48, 0.29; P = 0.08, I ² = 65%), and trials published in abstract form [32] (MD −2.32 days; 95% CI −4.86, 0.22; P = 0.07, I ² = 67%). Details of these analyses are given in Additional file 1: Tables S6–S8.

Visual inspection of funnel plots for the mortality and ICU LOS outcomes (Additional file 1: Figures S2–S3) did not reveal small-study effects suggestive of publication bias.