Two independent authors (Zhou X and Fang H) in our team conducted a comprehensive database search for RCTs, where critically ill adults were randomized to receive either hypocaloric enteral feeding (defined as daily provision of <70% of estimated or measured energy requirements4) or standard enteral feeding (defined as daily provision of ≥70% of estimated or measured energy requirements), using medical subject heading terms combined with liberal terms in PubMed, EMBASE, Web of Science and the Cochrane database of clinical trials from database inception through 16 March 2019. We also manually searched the reference lists of previous review articles to further identify the relevant literature. There were no restrictions on language in this meta-analysis. The detailed search strategy is available in the supplementary material.

All records from the database search were filtered to exclude duplicates. Three authors (Zhou X, Xu J, and Wang H) independently screened the title and abstract of the remaining records to determine whether the studies met the inclusion criteria. The full-text of the records deemed eligible during preliminary screening were retrieved and reviewed in accordance with the inclusion and exclusion criteria. The inclusion criteria includes:

• 1)

  Study design: prospective RCT;

• 2)

  Population: critically ill adults (age≥16 years) admitted to the ICU who received EN as the main nutrition supply (>50% energy delivery) within 48hours after ICU admission;

• 3)

  Intervention: hypocaloric enteral feeding with a mean daily caloric intake of <70% of estimated or measured energy requirements4;

• 4)

  Control: standard enteral feeding with a mean daily caloric intake of ≥70% of estimated or measured energy requirements.4

There were no restrictions on protein intake, delivery method, energy dense or formulas of enteral nutrition, or measurement method of energy requirement. Those studies that did not directly compare two dose of caloric intake but led to different dose of caloric intake were also considered. The exclusion criteria includes:

• 1)

  Not RCT;

• 2)

  Not critically ill adults or not admitted to the ICU;

• 3)

  Received parenteral nutrition (PN) as the main source of nutrition;

• 4)

  Both groups received a mean daily caloric intake of ≥ or <70% of estimated or measured energy requirements;

• 5)

  Did not report the proportion of daily caloric intake to goal caloric requirements or the interested outcomes.

We also excluded those abstract without full-text. Any disagreements between the three authors were resolved through discussion until a consensus was reached.

Two authors (Pan J and Sha Y) independently extracted the associated data from each included trials using a standardized data extraction form, including characteristics of the included studies, details of the population enrolled, duration of intervention, and detailed information on EN, including the nutrition regimen, caloric intake, and protein intake. We contacted the authors for the complete data set via email if some interested data were not reported. Discrepancies between the reviewers were resolved through discussion with a third reviewer and by consensus.

The primary outcomes include all-cause short-term mortality (defined as death within 30 days of randomization, including ICU, in-hospital, 28-day, or 30-day mortality) and incident of nosocomial infection. Secondary outcomes include all-cause long-term mortality (defined as death occurring beyond 30 days of randomization, including 60-day, 90-day, or 180-day mortality), incident of bloodstream infection, pneumonia, hypoglycemia and GI intolerance. Tertiary outcomes include duration of ICU stay, in-hospital stay, and duration of mechanical ventilation (MV). Studies that assessed at least one of the above outcomes were included. In studies in which various short-term or long-term mortalities were reported, the longest follow-up mortality was included in analysis. The infection complications (nosocomial infection, bloodstream infection and pneumonia) and hypoglycemia were defined by the authors in the included trials, the GI intolerance was predefined as GI dysfunction with symptoms of vomiting, noninfectious diarrhea, abdominal distension, regurgitation, or large gastric residual volumes, or defined as by the authors in the included trials.

The quality of each included trials were independently evaluated by two authors (Hu C and Xu Z) for the primary outcomes according to the Cochrane Collaboration methods.27 Disagreements were resolved via discussion with a third reviewer to reach a consensus on the quality evaluation of included studies. We assigned a value of high, unclear or low to the following seven domains: adequate sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other bias. Studies that exhibited a low risk of bias in all domains were adjudicated as have low risk of bias, or were adjudicated as have high risk of bias.

All statistical analyses were performed using Stata/SE 11.0 (StataCorp, College Station, TX, USA). The pooled relative risk (RR) for dichotomous data and the pooled weight mean difference (WMD) for continuous data, with corresponding 95% confidence interval (CI), were calculated. However, data on duration of MV were pooled using the standardized mean difference (SMD) due to the different time unit. We assessed the statistical heterogeneity among trials by inspecting the forest plots and quantitatively by using the diversity (D2) and inconsistency factor (I2) statistics. Both fixed-effects and random-effects meta-analyses were conducted for all outcomes (Table 1). Given the substantial clinical or statistical heterogeneity (in primary diagnosis, enteral nutrition formulas, energy measure method, etc.) between included trials, we reported the pooled data from random-effects model as the main result, otherwise we reported the results from a fixed-effects model if one or two trials accounts for approximately 80% or more of the total weight in a fixed-effect meta-analysis.28 Funnel plots combined with Begg's and Egger's tests were performed to assess publication bias for the primary outcomes if 10 or more studies were included. Considering the multiple outcomes reported, we assessed the primary outcomes with statistical significance set at p<0.033, secondary outcomes at p<0.017, and tertiary outcomes at p<0.025.28

Subgroup analyses were also conducted for all outcomes based on the following factors: (1) the overall risk of bias of included studies (low or high risk of bias); (2) whether the dose of protein intake between the two groups was similar or different (statistically significant or if the difference was ≥10%); (3) the design type of study (single-center or multi-center).

We conducted trial sequential analysis (TSA) using TSA program version 0.9.5.10 beta (available from www.ctu.dk/tsa) for the primary and secondary outcomes (Table 1) to assess the increased risk of random errors due to sparsity data and repeated significance testing in cumulative meta-analyses.29 Both random-effects and fixed-effects models were used to calculate the TSA-adjusted 95% CI for heterogeneity [Diversity (D2) adjustment], the random-effects result was reported as the main result.

Because the overall risk of falsely rejecting the null hypothesis (the family-wise error rate) will increase when performing multiple hypothesis tests,28 we calculated the TSA-adjusted 95% CI with a family-wise error rate of 5% with a statistical significance level of 3.3% for the anticipated two co-primary outcomes and 1.7% for each of the five co-secondary outcomes, a beta (power) of 80%, and a D230 suggested by the included trials. If the actual measured D2 was zero, a D2 of 20% was used because the heterogeneity may be expected to increase when further studies are included.31 The required information size was calculated based on a relative risk reduction of 15% in the event proportion of control group calculated from the conventional meta-analysis.

According to the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) approach,32 we assessed the overall quality of evidence for each outcome measure using GRADE profiler version 3.6 and presented the summary of findings in Table 2. The quality of evidence and our confidence in the effect-estimates were evaluated on the following six elements: study design, risk of bias, inconsistency, indirectness, inprecision, and risk of publication bias. Accordingly, we assigned a value of “high”, “moderate”, “low” or “very low” to each outcome to classify the quality of evidence.

Table 2.

Summary of findings and the quality of evidence of all outcomes.

The database search was completed on March 16, 2019. We searched a total of 5918 records from the abovementioned database, of which 1256 duplicate were filtered, and 4425 records deemed ineligible were also excluded after screening the title and abstract according to our inclusion and exclusion criteria. The remaining 237 records and additional 32 records retrieved from other review articles were included in the review of the full text. Finally, a total of 11 RCTs8–13,23–25,33,34 enrolled 6986 subjects met the eligibility criteria and were included in analysis. The flowchart of study selection is shown in Fig. 1 and the reasons for exclusion of studies are presented in the supplementary material (Table S1).

Figure 1.

The flowchart of the study selection process.

(0.46MB).

Among the included RCTs,8–13,23–25,33,34 all were published from 2011 to 2018, two9,25 were published in Chinese and the remainder were in English, three8,12,24 were multi-center and the others were single-center.9–11,13,23,25,33,34 The number of subjects enrolled were largely different between included trials and varied from 78 to 3957. The subjects in all trials were critically ill patients admitted to medical ICU, surgical ICU, or both, however, the primary diagnosis led to ICU admission were diverse. The subjects in all trials received EN as the main nutrition supply, other nutrition regimen, including PN, were also indicated if need. Among the included trials, eight studies8–13,25,33 directly compared two different dose of caloric intake, yet the remaining three trials23,24,34 did not. The aim of the trial by Allingstrup et al.23 was to assess the effectiveness of early goal-directed nutrition as compared to standard nutrition care, the trial by Braunschweig et al.34 was to assess the effectiveness of intensive medical nutrition therapy as compared to standard nutrition therapy, and the trial by Chapman et al.24 was to assess the effectiveness of two different energy-dense enteral nutrition in critically ill patients. Although, the interventions in the three studies resulted in a dose difference in daily caloric intake which met our criteria, hence we also included the three trials in this meta-analysis. The proportion of daily caloric intake to goal caloric requirements in hypocaloric feeding group ranged from 15% to 69%, and in standard feeding group were from 71% to 103%. The dose of protein intake was largely diverse between trials and between groups, four trials8,13,24,33 received a similar dose of protein intake and four trials10,11,23,34 received a significantly different dose of protein intake between the two groups, the others9,12,25 did not reported data on protein intake. The detailed characteristics and clinical outcomes of the individual studies are described in the supplementary material (Tables S2 and S3).

The details of the overall risk of bias of individual trials are summarized in Table 3. Summarily, two studies13,24 had a low risk of bias in all domains and are adjudicated as an overall low risk of bias, and the remaining studies8–12,23,25,33,34 were deemed as an overall high risk of bias due to at least one domain with an unclear or high risk of bias.

There are 10 trials8–12,23–25,33,34 consisting of 6855 participants presented data on incident of nosocomial infection, all but one24 were judged as have high risk of bias. Overall, we found no difference in the incident of nosocomial infection between the hypocaloric feeding and standard feeding group, with an RR of 1.01 (95% CI: 0.95–1.07, I2=13.3%) (Fig. 5) and TSA-adjusted 95% CI of 0.90–1.12 (D2=83%), meanwhile, the Z-curve reached the required information size (Fig. 6A), indicating sufficient evidence to reject a relative risk reduction of 15% in nosocomial infection. A similar results was found in sub-analysis of trials received similar dose of protein8,24,33 (Figs. 5 and 6B), trials received different dose of protein10,11,23,34 (Fig. 5), single-center trials9–11,23,25,33,34 and multi-center trials8,12,24 (Figs. S3 and S4).

Figure 5.

Forest plot of sub-analysis of trials received similar or different dose of protein for the incident of nosocomial infection. RR, relative risk.

(0.33MB).

Figure 6.

Trial sequential analysis for the incident of nosocomial infection. Trial sequential analysis using random-effects model with an adjusted family-wise error rate of 3.3%, power of 80%, for a relative risk reduction of 15% in control event proportion. (Panel A) In all included trials, control event proportion of 61.1%, D2 of 83%. The cumulative Z-curve cross the futility area and reach the required information size of 6035 participants. The TSA-adjusted 95% CI for an RR of 1.01 is 0.90–1.12. (Panel B) In trials received similar dose of protein, control event proportion of 74.4%, D2 of 20% (the actual measured D2 was 0%), the cumulative Z-curve cross the futility area and reach the required information size of 762 participants. The TSA-adjusted 95% CI for an RR of 0.99 is 0.93–1.05. RR, relative risk; TSA, trial sequential analysis.

(0.23MB).

We evaluated the publication bias for the primary outcomes. All the funnel plots were visually symmetrical (Fig. S5). The Begg's and Egger's tests for the short-term mortality (P=0.474 and P=0.510, respectively) and the incident of nosocomial infection (P=0.474 and P=0.298, respectively) indicated no evidence of publication bias.

The incident of bloodstream infection was reported in five studies,8,12,23,24,33 the incident of pneumonia was in five studies,8,11,12,23,33 and the incident of hypoglycemia was in six.8,13,23,24,33,34 The cumulative meta-analyses demonstrated that hypocaloric feeding, when compared with standard feeding, had no effects on the incident of bloodstream infection (Figs. S8 and S9), pneumonia (Figs. S10 and S11), or hypoglycemia (Figs. S12 and S13), regardless of the trials quality, the design type of trials and whether the dose of protein intake between groups was similar or different. However, these results were not robust because no boundaries were crossed in the TSA for the three outcomes, indicated that these results could be changed when future trials were included.

We identified five studies8,10–12,24 reported data on the incident of GI intolerance, one24 trial had a low risk of bias. Hypocaloric feeding could result in a lower incident of GI intolerance compared to standard feeding (RR=0.78, 95% CI: 0.70–0.87, p<0.001, I2=11.4%) (Fig. S14), which was confirmed by TSA (TSA-adjusted 95% CI: 0.67–0.90, D2=21%) and the trial sequential monitoring boundary for benefit was crossed (Fig. S15), a similar results were found in the sub-analysis of multi-center trials.8,12,14 In the sub-analysis of trials received similar dose of protein8,24 and trials received different dose of protein10,11 and single-center trials,10,11 we also found a lower incident of GI intolerance in hypocaloric feeding group, however this benefits disappeared after adjustment with TSA (Fig. S15).

In the meta-analyses of 10 trials8,9,11–13,23–25,33,34 involving the duration of ICU stay, 7 trials8,9,11,23,24,33,34 involving in-hospital stay, and 10 trials8–13,24,25,33,34 involving duration of MV, hypocaloric feeding was not superior to standard feeding in reducing the duration of MV (Fig. S16), ICU stay (Fig. S17), or in-hospital stay (Fig. S18). Furthermore, these results were not changed in sub-analysis of trials stratified based on the trials quality, the design type of trials and the dose of protein intake, however, hypocaloric feeding was associated with a longer ICU stay in sub-analysis of multi-center trials8,12,24 (Fig. S17C).

The overall quality of evidence was evaluated for all outcomes, we presented the summary of findings and the reasons for downgrading the quality in Table 2.