Risk of Bias
Based on a priori defined criteria,11,12 only one study was found to have a major methodological flaw resulting in a high risk of bias.21
Three RCTs explicitly reported the process used to maintain allocation concealment,19, 20,21 whilst the remaining five were unclear. One study achieved blinding using a placebo intervention.18 Three studies documented failure to follow-up all randomized patients, with one reporting loss of 6.3% (7/111)20 of randomized patients, a second reporting 9.1% (8/88) loss 25 and the third documenting loss of 21.5% (11/51) of randomized patients.21 Complete results of the risk of bias assessment are presented in eFigure 1.
The funnel plot of the primary outcome did not reveal publication bias (eFigure 2).
Mortality
Eight RCTs enrolling 657 patients were included in the analysis of mortality. Two studies reported mortality at study day 30,18, 20 one reported mortality at study day 60,22 with the remaining five studies reporting mortality at time of hospital discharge. Mortality data for the trial by Mulrooney et al 22 was abstracted from the systematic review by Herbert et al.6 Herbert et al. reported corresponding with Mulrooney et al to obtain this mortality information.
Compared with later (traditional) feeding, commencing an early oral protein-containing diet resulted in a statistically significant reduction in mortality (OR 0.31, 95% Confidence Interval [CI] 0.12 to 0.80, P=0.02, Figure 2), with no important heterogeneity detected (Pheterogeneity=0.95, I2=0%).
Physical function
Three studies reported measures of physical function. Carr et al. reported change in handgrip strength,19 Stewart et al. documented time to mobilization after surgery25 and Minig et al. assessed physical function at study day 30 using the European Organization for Research and Treatment of Cancer (EORTC) C-30.21
Carr et al. reported a mean 9.6 (Standard Deviation [SD] 2.1) kg loss in handgrip strength in control patients, and a 6.7 (3.2) kg loss in handgrip strength in patients who received an early diet containing protein.19 Assessed using a standard t-test for differences between groups, patients who received an early diet containing protein experienced significantly less handgrip strength loss (2.9 kg less, 95% CI 0.9 to 4.9kg, P=0.01).
Minig et al. failed to find a significant difference between groups with regards to physical function assessed at study day 30 using EORTC C-30 (76.1±14.3 early vs. 62.1±25.3 traditional, P=0.146).21
Stewart et al. also failed to find a significant difference between groups with regards to time to mobilization after surgery.25 Stewart et al. did not report the actual times to mobilization for each group.
Because of the differences in outcome metrics reported, these measures of physical function could not be pooled.
Quality of life
One RCT reported formal measures of quality of life.21 Minig et al. failed to find a significant difference in EORTC OV-28 assessed at study day 30.
Duration of hospital stay
All eight included RCTs reported duration of hospital stay. One trial reported differences in median hospital stay,18 with the remaining seven reporting mean (SD). Mean (SD) hospital stay data for the trials by Minig et al., Mulrooney et al., and Stewart et al. was abstracted from the systematic review by Herbert et al.6 Herbert et al. reported corresponding with these authors to obtain this additional mean (SD) information regarding hospital stay.
Using a non-parametric test, the individual RCT conducted by Beier-Holgersen et al. reported a trend towards a reduction in median hospital stay for patients who received an early diet containing protein: median 8 vs. 11.5 days (P=0.08).18
Meta-analysis of the mean (SD) length of stay data reported by seven RCTs recruiting 598 patients demonstrated a significantly shorter hospital stay for patients randomized to receive an early oral protein-containing diet (-2.12 days, 95% CI -2.74 to -1.49 days, P<0.00001, eFigure 3), however, important heterogeneity was detected (Pheterogeneity=0.00006, I2=75%). Sources of this heterogeneity are investigated further with stratified analysis (see later in Results).
Intensive Care Unit admission
Three studies explicitly reported Intensive Care Unit (ICU) admission rates after surgery.18,21,23 Meta-analysis did not reveal any significant difference between groups (OR 0.61, 95% CI 0.24 to 1.53, P=0.29, eFigure 4), with no important heterogeneity detected (Pheterogeneity=0.55, I2=0%).
Surgical site infections
All eight RCTs documented surgical site infection rates,18,19,20,21,22,23,24,25 however one combined reporting of surgical site infections with urinary tract infections and therefore could not be included in this pooled analysis.19 Surgical site infection data for the trial by Mulrooney et al. was abstracted from the systematic review by Herbert et al.6 Herbert et al. reported corresponding with Mulrooney et al. to obtain this additional information.
The seven RCTs that explicitly reported surgical site infections enrolled 625 patients. Patients who received an early oral protein-containing diet were significantly less likely to experience a surgical site infection (OR 0.39, 95% CI 0.21 to 0.71, P=0.002, Figure 3), with no important heterogeneity detected (Pheterogeneity=0.19, I2=32%).
Anastomotic leak/dehiscence
Five RCTs enrolling 439 patients reported anastomotic leak,18,21 anastomotic leak/dehiscence,22,25 or anastomotic breakdown.23 None of these trials reported using explicit and objective criteria to diagnose a leak/dehiscence/breakdown. Anastomotic leak/dehiscence data for the trial by Mulrooney et al. was abstracted from the systematic review by Herbert et al.6 Herbert et al. reported corresponding with Mulrooney et al. to obtain this additional information.
Meta-analysis did not find any significant difference between groups (OR 0.74, 95% CI 0.30 to 1.87, P=0.53, eFigure 5) and no important heterogeneity was detected (Pheterogeneity=0.10, I2=49%).
Postoperative nausea and vomiting
Three studies reported nausea.18,20,25 Using a Likert scale, Lau et al. reported a significantly lower nausea score in the early diet containing protein group compared to the traditional group (2.5 early vs. 4.7, P=0.01).20 However, Stewart et al. failed to find a significant difference between groups using a visual-analogue nausea scale (29 early vs. 31, reported as Not Significant[NS]),25 and Beier-Holgersen et al. also failed to find a significant difference between groups in the incidence of nausea (19/30 early vs. 22/30, NS).18 Due to the different measures used to assess nausea, pooled analysis could not be undertaken.
Three RCTs reported the number of patients who vomited,18,20,25 two RCTs reported the number of patients with ‘nausea and vomiting’ combined,19,21 and one study provided a graphical representation of vomiting on each study day but did not explicitly report rates.23
Pooled analysis of the five RCTs enrolling 312 patients that explicitly reported rates demonstrated a significant reduction in postoperative nausea and vomiting attributable to the provision of an early oral protein-containing diet (OR 0.62, 95% CI 0.38 to 0.99, P=0.04, eFigure 6), with no important heterogeneity detected (Pheterogeneity=0.17, I2=37%).
Pneumonia
Six RCTs enrolling 585 patients reported pneumonia.18,20,22,23,24,25 Pneumonia data for the trials by Mulrooney et al. and Stewart et al. was abstracted from the systematic review by Herbert et al.6 Herbert et al. reported corresponding with Mulrooney et al. and Stewart et al. to obtain this additional information.
Pooled analysis failed to demonstrate a significant difference between groups (OR 0.73, 95% CI 0.32 to 1.66, P=0.45, eFigure 7), with no important heterogeneity detected (Pheterogeneity=0.60, I2=0%).
Need for re-operation
Three RCTs enrolling 204 patients explicitly reported need for reoperation.18,20,21 Pooled analysis failed to demonstrate a significant difference between groups (OR 0.49, 95% CI 0.16 to 1.51, P=0.22, eFigure 8), with no important heterogeneity detected (Pheterogeneity=0.84, I2=0%).
Intra-abdominal abscess/peritonitis
Six RCTs reported intra-abdominal abscess/peritonitis.18,20,21,22,23,24 Data for the trial by Mulrooney et al. was abstracted from the systematic review by Herbert et al.6 Herbert et al. reported corresponding with Mulrooney et al. to obtain this additional information.
Pooled analysis of these six RCTs enrolling 545 patients demonstrated a significant reduction in the onset of intra-abdominal abscess/peritonitis in patients who received an early oral protein-containing diet (OR 0.20, 95% CI 0.06 to 0.66, P=0.008, eFigure 9), with no important heterogeneity detected (Pheterogeneity=0.98, I2=0%).
Number of patients with serious postoperative complications
Six RCTs enrolling 552 patients explicitly reported the total number of patients in each study group who had at least one serious postoperative complication.18,20,21,23,24,25 Serious postoperative complications included acute myocardial infarction, anastomotic leak/dehiscence, unexpected return to surgery, hospital readmission within 30 days of discharge, surgical site infection, peritonitis, intestinal obstruction, and other postoperative infections. Because mortality served as the primary outcome for this study, it was not included in this analysis of serious postoperative complications.
Pooled analysis of these six RCTs revealed that the provision of an early oral protein-containing diet resulted in significantly fewer patients developing a serious postoperative complication (OR 0.60, 95% CI 0.40 to 0.89, P=0.01, eFigure 10). Despite reporting of a different subset of serious postoperative complications by each study, there was no important heterogeneity detected (Pheterogeneity=0.25, I2=25%).
Number of patients with a postoperative infection
Four RCTs enrolling 210 patients reported the number of patients who had at least one postoperative infection.18,19,21,24 Based on pooled analysis of these four RCTs, the provision of an early oral protein-containing diet resulted in a significant reduction in the number of patients who experienced a postoperative infection (OR 0.17, 95% CI 0.08 to 0.37, P<0.0001, eFigure 11), with no important heterogeneity detected (Pheterogeneity=0.83, I2=0%).
Sensitivity analysis
Focused on the primary outcome, the sensitivity analysis considered trials with less certainty regarding protein content of the intervention group’s early nutrition. Thirteen clinical trials enrolling 1,216 patients were identified for inclusion in the sensitivity analysis.
These RCTs described their early nutrition intervention as “water”,26 “5% glucose”,27,28 “clear liquid”,29,30 “allowed to drink”,31 “oral liquids”,32 an “oral liquid diet”,33 a “fluid diet”,34 “fluids”,35 a “liquid diet”,36 a “semi-fluid diet”,37 or “filtrate liquids”.38
Inclusion of these 13 RCTs in a sensitivity analysis failed to find an impact of early non-protein liquid diets on mortality (OR 1.01, 95% CI 0.29 to 3.51, P=0.99). Furthermore, there was important heterogeneity between RCTs that evaluated a protein-containing diet and RCTs that evaluated non-protein liquid diets, suggesting these different interventions have different effects on mortality (Pheterogeneity=0.14, I2=54.7%, Figure 4).
Heterogeneity and stratified analysis
The only statistically significant result demonstrating important heterogeneity was the analysis of duration of hospital stay (I2=75%, eFigure 3). Stratified analysis based on study intervention (enteral nutrition/solid diet/protein drink supplement) meaningfully reduced heterogeneity within each strata (eFigure 12). Interpretation of results within each strata revealed that early enteral nutrition did not have any effect on duration of hospital stay (1.05 days, 95% CI -0.077 to 2.87 days, P=0.26, I2=0%), whilst both an early solid protein-containing diet (-1.86 days, 95% CI -2.73 to -1.00, P<0.000001, I2=35%) and early use of protein drink supplements (-3.54 days, 95% CI -4.59 to -2.49, P<0.000001, I2=0%) significantly reduced duration of hospital stay.