Supplementary Text R2 1
Additional details of methods and results
Content
Search strategy…………………………………………………………………………..3
Eligibility criteria………………………………………………………………………..3
Data extraction…………………………………………………………………………..4
Assessment of methodological quality………………………………………………….5
Search results……………………………………………………………………………5
Study characteristics…………………………………………………………………….6
Risk of bias……………………………………………………………………………...6
References to studies included in the multiple-treatments meta-analysis………………7
WinBUGS codes for random effects model and fixed effects model for binary outcome
data…………………………………………...................................................................19
WinBUGS codes for random effects meta-regression model for binary outcome
data……….......................................................................................................................22
Rankogram and SUCRA……………………………………………………………….24
Inconsistency…………………………………………………………………………...26
Cross validation………………………………………………………………………...26
Between-study heterogeneity…………………………………………………………..27
Supplementary Text R2 2
Publication bias…………………………………………………………………………28
References……………………………………………………………………………...29
Supplementary Text R2 3
Search strategy
This multiple treatment meta-analysis followed the Preferred Reporting Items for
Systematic Review and Meta analyses (PRISMA) statements (Supplementary PRISMA
Checklist).1 Using MEDLINE, the Web of Science, and the Cochrane Library, an
English literature search was carried out for randomized controlled trials (RCTs)
published from January 1990 to February 2013 that evaluated the clinical efficacy of
SPN, SEN, IMPN, and IMEN in human adult patients undergoing elective
gastrointestinal surgery. Also, bibliographic reviews and abstracts presented through
2013 were manually searched. The following text words and medical subject headings
(MeSH) terms were used for searching: “Enteral Nutrition” AND/OR “Parenteral
Nutrition” AND/OR “Immunomodulations” AND/OR “immunonutrition” AND/OR
“immuno-enhancing” AND/OR “Omega-3 Fatty Acids” AND/OR “Omega-6 Fatty
Acids” AND/OR “Arginine” AND/OR “Glutamine” AND/OR “Fish Oils” AND/OR
“RNA” AND/OR “Nutritional Support” combined with “Randomized Controlled Trial”
AND/OR “Gastrointestinal” AND/OR “Surgical Procedures” AND/OR “Perioperative
Period” AND/OR “Postoperative Period” AND/OR “Preoperative Period”. The
experimental designs taken into consideration were as follows: RCTs.
Eligibility criteria
Supplementary Text R2 4
SEN and IMEN were the perioperative delivery of any nutrient in solid or liquid form
(including usual food intake) that passed through any part of the digestive tract,
regardless of whether the patients received conventional oral diets with intravenous
fluids (standard care) or tube feeds. SPN was defined as administration of nutritional
liquids containing a minimum of glucose and amino acids that were perioperatively
administered through the central or peripheral venous system. IMPN was also defined
as administration of SPN with fish oil emulsions. If more than one version of the same
study was retrieved, the most recent study was used. Exclusion criteria are as follows:
(1) trials that investigated the efficacy of an oral nutritional supplement (sip feed); (2)
trials that evaluated the impact of nutrition only on nutritional or physiologic outcomes
(e.g., nitrogen balance or amino acid profile); (3) trials that treated patients receiving
home parenteral nutrition; (4) trials that included cardiopulmonary, head injury,
pediatric, gynecologic, urological, traumatic, emergency, transplantation surgery,
chemotherapy, radiotherapy, or critically ill patients.
Data extraction
Data were extracted on study design, setting, patient population, pathology of diseases,
site of surgery, the regimens, methods of nutritional support, and the outcome variables
listed above. Outcomes assessed were the incidence of any infection, overall
Supplementary Text R2 5
complication, mortality, wound infection, pneumonia, anastomotic leak, intra-abdominal
abscess, sepsis, and urinary tract infection for binary outcome data. Data were extracted
as the total number of patients affected by complications rather than the total number of
incidences of complications.
Assessment of methodological quality
Study quality was assessed using the Cochrane risk of bias tool, an established tool
based on the following domains: sequence generation for the randomization of subjects,
allocation concealment of treatment, blinding of participants, personnel, and outcome
assessors, incomplete outcome date, selective outcome reporting, and other sources of
bias—study design, early stopping, baseline imbalance, and some other problems. For
each study, the risk of bias was reported as “low risk”, “unclear risk”, or “high risk” in
the domains. Bias assessment was performed using Review Manager Version 5.2.3
(Cochrane Collaboration, UK).2
Search results
Seventy-four studies totaling 7,572 participants met all of the inclusion criteria; SPN
was compared with SEN in 29 studies; SPN was compared with IMPN in 18 studies;
SPN was compared with IMEN in 12 studies; SEN was compared with IMPN in 2
Supplementary Text R2 6
studies; SEN was compared with IMEN in 29 studies, and IMPN was compared with
IMEN in 2 studies (Supplementary Table 1 and Figure 2).
Study characteristics
Sixty studies stated the underlying pathology of the study participants, of which 48
studies (80%) were comprised of malignant status and the remaining 10 studies (20%)
included both malignant and benign status (Supplementary Table 1). No study included
only benign diseases. Twenty-two studies (30%) reported the number of patients with
malnutrition (Supplementary Table 1). Patients were fed through either a catheter tube
or orally with 21 kinds of EN (Supplementary Table 2); 5 kinds of IMEN and 15 kinds
of SEN (Supplementary Tables 3 and 4). Seven kinds of parenteral lipid emulsions were
administered; 3 kinds of IMPN, and 4 kinds of standard lipid emulsion (Supplementary
Tables 2 and 5). In fifty-seven studies, the nutrition was administered postoperatively;
preoperatively in 13 studies, and perioperatively in 12 studies (Supplementary Table 2).
Risk of bias
The risk of bias was adequate for 37 studies (50%) in the randomized sequence, clear
for 35 studies (47%) in the allocation concealment, adequate for 24 studies (32%) in the
double blinding, complete for 13 studies (18%) in the blinding of the outcome
assessment, low for 42 studies (57%) in the incomplete outcome data, low for 29 studies
Supplementary Text R2 7
(39%) in the selective reporting, and free of other bias for 22 studies (30%)
(Supplementary Figures 1A and 1B).
References to studies included in the multiple-treatments meta-analysis
1. Hamaoui E, Lefkowitz R, Olender L, et al. Enteral nutrition in the early
postoperative period: A new semi-elemental formula versus total parenteral
nutrition. J Parenter Enteral Nutr. 1990;14:501–507.
2. Schroeder D, Gillanders L, Mahr K, et al. Effects of immediate postoperative
enteral nutrition on body composition, muscle function, and wound healing. J
Parenter Enteral Nutr. 1991;15:376–383.
3. Von Meyenfeldt MF, Meijerink WJ, Rouflart MM, et al. Perioperative nutritional
support: a randomised clinical trial. Clin Nutr. 1992;11:180–186.
4. Reissman P, Teoh TA, Cohen SM, et al. Is early oral feeding safe after elective
colorectal surgery? A prospective randomised trial. Ann Surg. 1995;222:73–77.
5. Baigrie RJ, Devitt PG, Watkin DS. Enteral versus parenteral nutrition after
oesophagogastric surgery: A prospective randomized comparison. Aust N Z J Surg.
1996;66:668–670.
6. Beier-Holgersen R, Brandstrup B. Influence of early postoperative enteral nutrition
versus placebo on cell-mediated immunity, as measured with the Multitest CMI.
Supplementary Text R2 8
Scand J Gastroenterol. 1999;34:98–102.
7. Carr CS, Ling KD, Boulos P, et al. Randomised trial of safety and efficacy of
immediate postoperative enteral feeding in patients undergoing GI resection. BMJ.
1996;312:869–871.
8. Ortiz H, Armendariz P, Yarnoz C. Is early postoperative feeding feasible in elective
colon and rectal surgery? Int J Colorectal Dis.1996;11:119–121.
9. Hartsell PA, Frazee RC, Harrison JB, et al. Early postoperative feeding after
elective colorectal surgery. Arch Surg. 1997;132:518–521.
10. Reynolds JV, Kanwar S, Welsh FKS, et al. Does the route of feeding modify gut
barrier function and clinical outcome in patients after major upper GI surgery? J
Parenter Enteral Nutr. 1997;21:196–201.
11. Sand J, Luostarinen M, Matikainen M. Enteral or parenteral feeding after total
gastrectomy: Prospective randomised pilot study. Eur J Surg. 1997;163:761–766.
12. Shirabe K, Matsumata T, Shimada M, et al. A comparison of parenteral
hyperalimentation and early enteral feeding regarding systemic immunity after
major hepatic resection—the results of a randomized prospective study.
Hepatogastroenterology. 1997;44:205–209.
13. Stewart BT, Woods RJ, Collopy BT, et al. Early feeding after elective open
Supplementary Text R2 9
colorectal resections: A prospective randomized trial. Aust N Z J Surg.
1998;68:125–128.
14. Aiko S, Yoshizumi Y, Sugiura Y, et al. Beneficial effects of immediate enteral
nutrition after esophageal cancer surgery. Surg Today. 2001;31:971–978.
15. Bozzetti F, Braga M, Gianotti L, et al. Postoperative enteral versus parenteral
nutrition in malnourished patients with GI cancer: A randomized multicentre trial.
Lancet. 2001;358:1487–1492.
16. Braga M, Gianotti L, Gentilini O, et al. Early postoperative enteral nutrition
improves gut oxygenation and reduces costs compared with total parenteral
nutrition. Crit Care Med. 2001;29:242–248.
17. Pacelli F, Bossola M, Papa V, et al. EN-TPN Study Group. Enteral vs parenteral
nutrition after major abdominal surgery: An even match. Arch Surg. 2001;136:933–
936.
18. Page RD, Ooa AY, Russell GN, et al. Intravenous hydration versus naso-jejunal
enteral feeding after esophagectomy: A randomised study. Eur J Cardio-thorac
Surg. 2002;22:666–672.
19. Rayes N, Hansen S, Seehofer D, et al. Early enteral supply of fiber and Lactobacilli
versus conventional nutrition: A controlled trial in patients with major abdominal
Supplementary Text R2 10
surgery. Nutrition. 2002;18:609–615.
20. Feo CV, Romanini B, Sortini D, et al. Early oral feeding after colorectal resection:
A randomized controlled study. Aust NZ J Surg. 2004;74:298–301.
21. Petkova PS. Improved clinical outcome in patients with early enteral nutrition after
major abdominal surgery. Clin Nutr. 2004;23:1457–1458.
22. Wu GH, Liu ZH, Wu ZH, et al. Perioperative artificial nutrition in malnourished
gastrointestinal cancer patients. World J Gastroenterol. 2006;12:2441–2444.
23. Wachtler P, König W, Senkal M, et al. Influence of a total parenteral nutrition
enriched with omega-3 fatty acids on leukotriene synthesis of peripheral leukocytes
and systemic cytokine levels in patients with major surgery. J Trauma.
1997;42:191–198.
24. Kelbel I, Wagner F, Wiedeck-Suger H, et al. Effects of n-3 fatty acids on immune
function: a double-blind, randomized trial of fish oil based infusion in
post-operative patients. Clin Nutr. 2002;21:13–14.
25. Weiss G, Meyer F, Matthies B, et al. Immunomodulation by perioperative
administration of n-3 fatty acids. Br J Nutr. 2002;87:S89–S94.
26. Heller AR, Rössel T, Gottschlich B, et al. Omega-3 fatty acids improve liver and
pancreas function in postoperative cancer patients. Int J Cancer.2004;111:611–616.
Supplementary Text R2 11
27. Kłek S, Kulig J, Szczepanik AM, et al. The clinical value of parenteral
immunonutrition in surgical patients. Acta Chir Belg. 2005;105:175–179.
28. Senkal M, Geier B, Hannemann M, et al. Supplementation of omega-3 fatty acids
in parenteral nutrition beneficially alters phospholipid fatty acid pattern. J Parenter
Enteral Nutr. 2007;31:12–17.
29. Wichmann MW, Thul P, Czarnetzki HD, et al. Evaluation of clinical safety and
beneficial effects of a fish oil containing lipid emulsion (Lipoplus, MLF541): data
from a prospective, randomized, multicenter trial. Crit Care Med. 2007;35:700–
706.
30. Liang B, Wang S, Ye YJ, et al. Impact of postoperative omega-3 fatty
acid-supplemented parenteral nutrition on clinical outcomes and
immunomodulations in colorectal cancer patients. World J Gastroenterol.
2008;14:2434–2439.
31. Badía-Tahull MB, Llop-Talaverón JM, Leiva-Badosa E, et al. A randomised study
on the clinical progress of high-risk elective major gastrointestinal surgery patients
treated with olive oil-based parenteral nutrition with or without a fish oil
supplement. Br J Nutr. 2010;104:737–741.
32. Jiang ZM, Wilmore DW, Wang XR, et al. Randomized clinical trial of intravenous
Supplementary Text R2 12
soybean oil alone versus soybean oil plus fish oil emulsion after gastrointestinal
cancer surgery. Br J Surg. 2010;97:804–809.
33. Makay O, Kaya T, Firat O, et al. ω-3 Fatty acids have no impact on serum lactate
levels after major gastric cancer surgery. J Parenter Enteral Nutr. 2011;35:488–
492.
34. Wang J, Yu JC, Kang WM, et al. Superiority of a fish oil–enriched emulsion to
medium-chain triacylglycerols/long-chain triacylglycerols in gastrointestinal
surgery patients: A randomized clinical trial. Nutrition. 2012;28:623–629.
35. de Miranda Torrinhas RSM, Santana R, Garcia T, et al. Parenteral fish oil as a
pharmacological agent to modulate postoperative immune response: A randomized,
double-blind, and controlled clinical trial in patients with gastrointestinal cancer.
Clin Nutr. 2012;doi:10.1016/j.clnu.2012.12.008.
36. Han YY, Lain SL, Ko WJ, et al. Effects of fish oil on inflammatory modulation in
surgical intensive care unit patients. Nutr Clin Pract. 2012;27:91–98.
37. Ma CJ, Sun LC, Chen FM, et al. A double-blind randomized study comparing the
efficacy and safety of a composite vs. a conventional Intravenous fat emulsion in
postsurgical gastrointestinal tumor patients. Nutr Clin Pract. 2012;27:410–415.
38. Zhu X, Wu Y, Qiu Y, et al. Effect of parenteral fish oil lipid emulsion in parenteral
Supplementary Text R2 13
nutrition supplementation combined with enteral nutrition support in patients
undergoing pancreaticoduodenectomy. J Parenter Enteral Nutr. 2013;37:236–242.
39. Heslin MJ, Latkany L, Leung D, et al. A prospective, randomized trial of early
enteral feeding after resection of upper GI malignancy. Ann Surg. 1997;226:567–
580.
40. Gianotti L, Braga M, Nespoli L, et al. A randomized controlled
trial of preoperative oral supplementation with a specialized diet in
patients with gastrointestinal cancer. Gastroenterology. 2002;122:1763–1770.
41. Helminen H, Raitanen M, Kellosalo J. Immunonutrition in elective gastrointestinal
surgery patients. Scand J Surg. 2007;96:46–50.
42. Suzuki D, Furukawa K, Kimura F, S et al. Effects of perioperative immunonutrition
on cell-mediated immunity, T helper type 1 (Th1)/Th2 differentiation, and Th17
response after pancreaticoduodenectomy. Surgery. 2010;148:573–581.
43. Liu C, Du Z, Lou C, et al. Enteral nutrition is superior to total parenteral nutrition
for pancreatic cancer patients who underwent pancreaticoduodenectomy. Asia Pac J
Clin Nutr. 2011;20:154–160.
44. Daly JM, Lieberman MD, Goldfine J, et al. Enteral nutrition with supplemental
arginine, RNA, and omega-3 fatty acids in patients after operation: immunologic,
Supplementary Text R2 14
metabolic, and clinical outcome. Surgery. 1992;112:56–67.
45. Daly JM, Weintraub FN, Shou J, et al. Enteral nutrition during multimodality
therapy in upper gastrointestinal cancer patients. Ann Surg. 1995;221:327–338.
46. Wachtler P, Hilger RA, König W, et al. Influence of a pre-operative enteral
supplement on functional activities of peripheral leukocytes from patients with
major surgery. Clin Nutr. 1995;14:275–282.
47. Kenler AS, Swails WS, Driscoll DF, et al. Early enteral feeding in postsurgical
cancer patients. Fish oil structured lipid-based polymeric formula versus a standard
polymeric formula. Ann Surg. 1996;223:316–333.
48. Senkal M, Mumme A, Eickhoff U, et al. Early postoperative enteral
immunonutrition: clinical outcome and cost-comparison analysis in surgical
patients. Crit Care Med. 1997;25:1489–1496.
49. McCarter MD, Gentilini OD, Gomez ME, et al. Preoperative oral supplement with
immunonutrients in cancer patients. J Parenter Enteral Nutr. 1998;22:206–211.
50. Braga M, Gianotti L, Radaelli G, et al. Perioperative immunonutrition in patients
undergoing cancer surgery: results of a randomized double-blind phase 3 trial. Arch
Surg. 1999;134:428–433.
51. Senkal M, Zumtobel V, Bauer KH, et al. Outcome and cost-effectiveness of
Supplementary Text R2 15
perioperative enteral immunonutrition in patients undergoing elective upper
gastrointestinal tract surgery: a prospective randomized study. Arch Surg.
1999;134:1309–1316.
52. Erdem NZ, Kulaçoĝlu İH, Temel NA, et al. Perioperative Oral Supplement with
Immunonutrients in Gastrointestinal Cancer Patients. Turk J Med Sci. 2001;31:79–
86.
53. Braga M, Gianotti L, Nespoli L, et al. Nutritional approach in malnourished
surgical patients: a prospective randomized study. Arch Surg. 2002;137:174–180.
54. Jiang XH, Li N, Zhu WM, et al. Effects of postoperative immune-enhancing enteral
nutrition on the immune system, inflammatory response, and clinical outcome.
Chinese medical Journal. 2004;117:835–839.
55. Farreras N, Artigas V, Cardona D, et al. Effect of early postoperative enteral
immunonutrition on wound healing in patients undergoing surgery for gastric
cancer. Clin Nutr. 2005;24:55–65.
56. Guoxiang Y, Xinbo X, Xingpei L, et al. Effects of postoperative enteral
immune-enhancing diet on plasma endotoxin level, plasma endotoxin inactivation
capacity and clinical outcome. J Huazhong Univ Sci Technolog Med Sci.
2005;25:431–434.
Supplementary Text R2 16
57. Lobo DN, Williams RN, Welch NT, et al. Early postoperative jejunostomy feeding
with an immune modulating diet in patients undergoing resectional surgery for
upper gastrointestinal cancer: a prospective, randomized, controlled, double-blind
study. Clin Nutr. 2006;25:716–726.
58. Xu J, Zhong Y, Jing D, et al. Preoperative enteral immunonutrition improves
postoperative outcome in patients with gastrointestinal cancer. World J Surg.
2006;30:1284–1289.
59. Finco C, Magnanini P, Sarzo G, et al. Prospective randomized study
on perioperative enteral immunonutrition in laparoscopic colorectal surgery. Surg
Endosc. 2007;21:1175–1179.
60. Klek S, Kulig J, Sierzega M, et al. Standard and immunomodulating enteral
nutrition in patients after extended gastrointestinal surgery--a prospective,
randomized, controlled clinical trial. Clin Nutr. 2008;27:504–512.
61. Gunerhan Y, Koksal N, Sahin UY, et al. Effect of preoperative immunonutrition
and other nutrition models on cellular immune parameters. World J Gastroenterol.
2009 28;15:467–472.
62. Okamoto Y, Okano K, Izuishi K, et al. Attenuation of the systemic inflammatory
response and infectious complications after gastrectomy with preoperative oral
Supplementary Text R2 17
arginine and omega-3 fatty acids supplemented immunonutrition. World J Surg.
2009;33:1815–1821.
63. Sodergren MH, Jethwa P, Kumar S, et al. Immunonutrition in patients undergoing
major upper gastrointestinal surgery: a prospective double-blind randomised
controlled study. Scand J Surg. 2010;99:153–161.
64. Fujitani K, Tsujinaka T, Fujita J, et al. Prospective randomized trial of preoperative
enteral immunonutrition followed by elective total gastrectomy for gastric cancer.
Br J Surg. 2012;99:621–629.
65. Hübner M, Cerantola Y, Grass F, et al. Preoperative immunonutrition in patients at
nutritional risk: results of a double-blinded randomized clinical trial. Eur J Clin
Nutr. 2012;66:850–855.
66. Sultan J, Griffin SM, Di Franco F, et al. Randomized clinical trial of omega-3 fatty
acid-supplemented enteral nutrition versus standard enteral nutrition in patients
undergoing oesophagogastric cancer surgery. Br J Surg. 2012;99:346–355.
67. Giger-Pabst U, Lange J, Maurer C, et al. Short-term preoperative supplementation
of an immunoenriched diet does not improve clinical outcome in well-nourished
patients undergoing abdominal cancer surgery. Nutrition. 2013;29:724–729.
68. Braga M, Vignali A, Gianotti L, et al. Immune and nutritional effects of early
Supplementary Text R2 18
enteral nutrition after major abdominal operations. Eur J Surg. 1996;162:105–112.
69. Schilling J, Vranjes N, Fierz W, et al. Clinical outcome and immunology of
postoperative arginine, omega-3 fatty acids, and nucleotide-enriched enteral
feeding: A randomized prospective comparison with standard enteral and low
calorie/low fat i.v. solutions. Nutrition. 1996;12:423–429.
70. Gianotti L, Braga M, Vignali A, et al. Effect of route of delivery and formulation of
postoperative nutritional support in patients undergoing major operations for
malignant neoplasms. Arch Surg. 1997;132:1222–1229.
71. Di Carlo V, Gianotti L, Balzano G, et al. Complications of pancreatic surgery and
the role of perioperative nutrition. Dig Surg. 1999;16:320–326.
72. Braga M, Gianotti L, Vignali A, et al. Preoperative oral arginine and n-3 fatty acid
supplementation improves the immunometabolic host response and outcome after
colorectal resection for cancer. Surgery. 2002;132:805–814.
73. Klek S, Kulig J, Sierzega M, et al. The impact of immunostimulating nutrition on
infectious complications after upper gastrointestinal surgery: a prospective,
randomized, clinical trial. Ann Surg. 2008;248:212–220.
74. Klek S, Sierzega M, Szybinski P, et al. Perioperative nutrition in malnourished
surgical cancer patients–a prospective, randomized, controlled clinical trial. Clin
Supplementary Text R2 19
Nutr. 2011;30:708–713.
WinBUGS code for random effects and fixed effects model for binary outcome
data
A hierarchical model with random effects was used to account for between-study
variance, in which data was formatted in a binomial likelihood with a logit link
function.3 The probability of an event in arm 𝑘 reported in study 𝑖 is denoted by 𝑝𝑖𝑘 .
The number of events 𝑟𝑖𝑘 in arm 𝑘 of study 𝑖 has the following binomial likelihood,
𝑟𝑖𝑘 ~𝐵(𝑝𝑖𝑘 , 𝑛𝑖𝑘 ) where 𝑛𝑖𝑘 is the sample size. For the binomial likelihood, we model
the probabilities 𝑝𝑖𝑘 on the logit scale, logit(𝑝𝑖𝑘 ) = 𝜇𝑖 + 𝛿𝑖𝑘 where 𝜇𝑖 is a random
parameter for the baseline. Then, the random effects are distributed
normally𝛿𝑖𝑘 ~𝑁(𝜇𝑖𝑘 , 𝜎𝑘 2 ), where 𝜇𝑖𝑘 is a study-specific logarithm of the odds ratio
and 𝜎𝑘 2 is the between study variance. The 𝜇𝑖𝑘 is expressed by 𝑑𝑘 (treatment
effects) and 𝑑1 (reference treatment effect); that is, 𝜇𝑖𝑘 = 𝑑𝑘 − 𝑑1. Prior distribution
needs to be set for 𝜇𝑖 , 𝑠𝑑(standard deviation), and 𝑑𝑘 : 𝜇𝑖 ~𝑁(0, 0.0001),
𝑠𝑑~𝑈(0, 5), and 𝑑𝑘 ~𝑁(0, 0.0001).4 The estimated odds ratio (OR) of a treatment 𝑐
versus a treatment 𝑘 is derived as: OR 𝑐𝑘 = exp(𝑑𝑐 − 𝑑𝑘 ), where 𝑑1 = 0 for the
treatment that has been denoted as the reference treatment. For a fixed effects model,
the between study variance (𝜎 2 ) is set to zero. WinBUGS codes were available at:
Supplementary Text R2 20
http://www.nicedsu.org.uk.
# Random effects model for multi-arm trials for binary outcome data
model{
for (i in 1:ns){
# adjustment for multi-arm trials is zero for control arm
w[i,1] <- 0
# treatment effect is zero for control arm
delta[i,1] <- 0
# vague priors for all trial baselines
mu[i] ~ dnorm(0,0.0001)
for (k in 1:na[i]) {
# binomial likelihood
r[i,k] ~ dbin(p[i,k],n[i,k])
# model for linear predictor
logit(p[i,k]) <- mu[i] + delta[i,k]
# expected value of the numerators
rhat[i,k] <- p[i,k] * n[i,k]
# deviance contribution
dev[i,k] <- 2 * (r[i,k] * (log(r[i,k])-log(rhat[i,k])) + (n[i,k]-r[i,k]) * (log(n[i,k]-r[i,k]) log(n[i,k]-rhat[i,k])))}
# summed residual deviance contribution for this trial
resdev[i] <- sum(dev[i,1:na[i]])
for (k in 2:na[i]) {
# trial-specific LOR distributions
delta[i,k] ~ dnorm(md[i,k],taud[i,k])
# mean of LOR distributions (with multi-arm trial correction)
md[i,k] <- d[t[i,k]] - d[t[i,1]] + sw[i,k]
# precision of LOR distributions (with multi-arm trial correction)
taud[i,k] <- tau *2*(k-1)/k
# adjustment for multi-arm RCTs
w[i,k] <- (delta[i,k] - d[t[i,k]] + d[t[i,1]])
# cumulative adjustment for multi-arm trials
sw[i,k] <- sum(w[i,1:k-1])/(k-1)}}
# total residual deviance
Supplementary Text R2 21
totresdev <- sum(resdev[])
# treatment effect is zero for reference treatment
d[1]<-0
# vague priors for treatment effects
for (k in 2:nt){
d[k] ~ dnorm(0,0.0001)}
# vague prior for between-trial SD
sd ~ dunif(0,5)
# between-trial precision = (1/between-trial variance)
tau <- pow(sd,-2)
# between-trial variance
var <- pow(sd,2)
# ranking on relative scale
for (k in 1:nt){
# assume events are "good"
rk[k]<-nt+1-rank(d[],k)
# assume events are "bad"
rk[k]<-rank(d[],k)
# calculate probability that treat k is best
best[k]<-equals(rk[k],1)}
# ranking of treatments
for (k in 1:nt) {
#
when the outcome is positive - omit
'nt+1-'
order[k]<- rank(d[],k)
# when the outcome is negative
most.effective[k]<-equals(order[k],1)
for (j in 1:nt) {
effectiveness[k,j]<- equals(order[k],j)}}
for (k in 1:nt) {
for (j in 1:nt) {
cumeffectiveness[k,j]<- sum(effectiveness[k,1:j])}}
# SUCRAS
for (k in 1:nt) {
SUCRA[k]<- sum(cumeffectiveness[k,1:(nt-1)]) /(nt-1)}
# pairwise ORs and LORs for all possible pair-wise comparisons, if nt>2
for (c in 1:(nt-1)) {
Supplementary Text R2 22
for (k in (c+1):nt) {
or[c,k] <- exp(d[k] - d[c])
lor[c,k] <- (d[k]-d[c])}}}
# Fixed effects model for multi-arm trials for binary outcome data
model{
for(i in 1:ns){
# vague priors for all trial baselines
mu[i] ~ dnorm(0,.0001)
for (k in 1:na[i]){
# binomial likelihood
r[i,k] ~ dbin(p[i,k],n[i,k])
# model for linear predictor
logit(p[i,k]) <- mu[i] + d[t[i,k]] - d[t[i,1]]
# expected value of the numerators
rhat[i,k] <- p[i,k] * n[i,k]
#Deviance contribution
dev[i,k] <- 2 * (r[i,k] * (log(r[i,k])-log(rhat[i,k])) +
(n[i,k]-r[i,k]) * (log(n[i,k]-r[i,k]) -
log(n[i,k]-rhat[i,k])))}
# summed residual deviance contribution for this trial
resdev[i] <- sum(dev[i,1:na[i]])}
# Total residual deviance
totresdev <- sum(resdev[])
# treatment effect is zero for reference treatment
d[1]<-0
WinBUGS code for random effects meta-regression model for binary outcome data
To assess the influence of covariates on the true efficacy of immunonutrition, a
meta-regression model was considered. The meta-regression analysis was performed
concerning the following 3 covariates: published year of articles, timing of the
administration of nutrition (preoperatively, postoperatively, or perioperatively), and
country of origin (Europe, North America, or Asia). The number of patients with
Supplementary Text R2 23
malnutrition was considered as a relevant covariate; but it could not be used due to the
reasons that only 34 studies out of the total of 74 (34%) reported the number of patients
with malnutrition. The main body of the WinBUGS code for the random effects
meta-regression model is similar to the random effects model; however, a subgroup and
a continuous covariate are incorporated into the model. In the case of a continuous
covariate, the analysis used centered covariate values (x[i]-mx). This is achieved by
subtracting the mean covariate value from each covariate.5 WinBUGS codes were
available at: http://www.nicedsu.org.uk.
# Random effects meta-regression model for multi-arm trials
model{
for(i in 1:ns){
# adjustment for multi-arm trials is zero for control arm
w[i,1] <- 0
# treatment effect is zero for control arm
delta[i,1] <- 0
# vague priors for all trial baselines
mu[i] ~ dnorm(0,0.0001)
for (k in 1:na[i]) {
# binomial likelihood
r[i,k] ~ dbin(p[i,k],n[i,k])
# model for linear predictor
logit(p[i,k]) <- mu[i] + delta[i,k] + (beta[t[i,k]]-beta[t[i,1]]) * (x[i]-mx)
# expected value of the numerators
rhat[i,k] <- p[i,k] * n[i,k]
# deviance contribution
dev[i,k] <- 2 * (r[i,k] * (log(r[i,k])-log(rhat[i,k])) + (n[i,k]-r[i,k]) * (log(n[i,k]-r[i,k])
-log(n[i,k]-rhat[i,k])))}
# summed residual deviance contribution for this trial
Supplementary Text R2 24
resdev[i] <- sum(dev[i,1:na[i]])
for (k in 2:na[i]) {
# trial-specific LOR distributions
delta[i,k] ~ dnorm(md[i,k],taud[i,k])
# mean of LOR distributions (with multi-arm trial correction) covariate effect relative to treat in arm 1
md[i,k] <- d[t[i,k]] - d[t[i,1]] + sw[i,k]
# precision of LOR distributions (with multi-arm trial correction)
taud[i,k] <- tau *2*(k-1)/k
# adjustment for multi-arm RCTs
w[i,k] <- (delta[i,k] - d[t[i,k]] + d[t[i,1]])
# cumulative adjustment for multi-arm trials
sw[i,k] <- sum(w[i,1:k-1])/(k-1)}}
# total residual deviance
totresdev <- sum(resdev[])
# treatment effect is zero for reference treatment
d[1]<-0
# covariate effect is zero for reference treatment
beta[1] <- 0
for (k in 2:nt){
# vague priors for treatment effects
d[k] ~ dnorm(0,0.0001)
# common covariate effect
beta[k] <- B}
# vague prior for covariate effect
B ~ dnorm(0,0.0001)
# vague prior for between-trial standard deviation
sd ~ dunif(0,5)
# between-trial precision = (1/between-trial variance)
tau <- pow(sd,-2)
# pairwise ORs and LORs for all possible pair-wise comparisons
for (c in 1:(nt-1)){
for (k in (c+1):nt){
or[c,k] <- exp(d[k] - d[c])
lor[c,k] <- (d[k]-d[c])}}}
Rankogram and SUCRA plot
Supplementary Text R2 25
Bayesian posterior probabilities rank the treatments for each outcome according to the
estimated effect size; that, the proportion of each Markov chain Monte Carlo cycle in
which a given treatment is ranked first out of the total number of cycles gives the
probability that the treatment is ranked in the best. Rank probabilities (rankograms) are
plotted against possible rank for all treatments (Supplementary Figure 2). However,
rankograms are unlikely to provide a ranking measure when many treatments are
competing and the probabilities to achieve each one of the possible ranks are the same
among the treatments.6,7 Alternatively, a cumulative probability that the treatment is
among the top 𝑏 = 1, 2, ⋯ , 𝑎 (anywhere between first and bth rank) may be useful.
The cumulative probability can be plotted against the possible rank. The larger the
surface under the cumulative probability ranking curve (SUCRA), the more probable
are the lowest rank—the more efficacious or acceptable treatment. If a treatment is
always ranks first, then SUCRA is 1, and if it always ranks last it is 0; that is SUCRA
would be 1 when a treatment is certain to be the best and 0 when a treatment is certain
to be the worst (Supplementary Figure 3). 6 The order of treatment 𝑘 in every Markov
chain Monte Carlo cycle is calculated as 𝑜𝑟𝑑𝑒𝑟𝑘 = ∑𝑎𝑏=1 𝐼(𝑑𝑏 ≤ 𝑑𝑘 ) where
𝐼(𝑑𝑏 ≤ 𝑑𝑘 ) = 1 if 𝑑𝑏 ≤ 𝑑𝑘 and 0 otherwise. The probability of treatment 𝑘 to be at
the 𝑏 order is estimated from 𝑒𝑓𝑓𝑒𝑐𝑡𝑖𝑣𝑒𝑛𝑒𝑠𝑠𝑘,𝑏 ; and, the cumulative probability is
Supplementary Text R2 26
derived from 𝑐𝑢𝑚𝑒𝑓𝑓𝑒𝑐𝑡𝑖𝑣𝑒𝑛𝑒𝑠𝑠𝑘,𝑏 . The SUCRA for treatment 𝑘 is 𝑆𝑈𝐶𝑅𝐴𝑘 =
∑𝑎−1
𝑏=1
𝑐𝑢𝑚𝑒𝑓𝑓𝑒𝑐𝑡𝑖𝑣𝑒𝑛𝑒𝑠𝑠𝑘,𝑏 6
.
𝑛𝑡−1
Inconsistency
A consistency between the direct and indirect evidence is an important assumption of
network meta-analysis.8,9 Consistency was assessed using the ifplot command
implemented in software package Stata Version 12.1 (Stata Corporation, College
Station, Texas)10. This command identified consistency within all first-order (triangles)
and second-order (quadrilaterals) closed loops in a network formed by multiple
treatments. Within each loop, the command defines the direct estimates and indirect
estimates for a randomly chosen contrast; then, the inconsistency factor (IF) is defined
as the difference between the direct and indirect estimates. The command presents the
absolute IF value and its CI (truncated to 0). Under the null hypothesis that there is no
inconsistency (H0: IF = 0; between-study variance, σ2), the approximate test can be
obtained as 𝑧 =
̂
IF
̂
𝜎
~𝑁(0, 1). A loop is defined as statistically inconsistent when |𝑧| >
1.96. When the 95% CI seems to include 0, the direct estimate of the summary effect
does not differentiate from the indirect estimate. Inconsistency was not significantly
recognized in a total of 33 loops (Supplementary Table 7).
Cross validation
Supplementary Text R2 27
We used a leave-one-out cross validation (LOO-CV) technique5 for detection of an
outlier; the LOO-CV removes a single data set from the original data set, and then
compare the observed treatment effect from the original data set to the posterior
predictive distribution of effects expected from the remaining data set. The predictive
probability of each outcome in a future study 𝑝𝑛𝑒𝑤 is given by 𝑙𝑜𝑔𝑖𝑡(𝑝𝑛𝑒𝑤 ) =
𝑙𝑜𝑔𝑖𝑡 (𝑝𝑏𝑎𝑠𝑒 ) + 𝛿𝑛𝑒𝑤 , where 𝑝𝑏𝑎𝑠𝑒 is the probability of reference treatment and 𝛿𝑛𝑒𝑤
is the predictive treatment effect in a future study. The predictive number of events
𝑟𝑛𝑒𝑤 in the treatment arm of a future study of the same size (𝑛) as the omitted study
can be drawn from a binomial distribution, 𝑟𝑛𝑒𝑤 ~𝐵𝑖𝑛(𝑝𝑛𝑒𝑤, 𝑛). The 𝑟𝑛𝑒𝑤 is compared
with the observed number of the omitted study (𝑟) to obtain a Bayesian p-value
Pr(𝑟𝑛𝑒𝑤 > 𝑟): the probability obtains a value as extreme as that observed in the omitted
study. Extra lines of code were added to the random effects model. The LOO-CV
showed that 4 studies were outliers: the studies of Badia-Tahull31 (P = 0.02),
Beier-Holgersen6 (P = 0.002), and Daly44 (P = 0.02) for any infection; the studies of
Beier-Holgersen6 (P = 0.02) and Daly44 (P = 0.02) for overall complication; the studies
of Beier-Holgersen6 (P = 0.006) and Senkal51 (P = 0.003) for wound infection; the study
of Daly44 (P = 0.008) for pneumonia (Supplementary Table 8).
Between-study heterogeneity (𝛔𝟐 )
Supplementary Text R2 28
Heterogeneity is characterized as between-study variation within treatment contrasts—
the result of an uneven distribution of treatment effect modifiers. The presence of
between-study heterogeneity was assessed as estimates of the median between-study
variance (σ2 ) for each comparison. The σ2 values of ≤ 0.04, 0.14, and 0.40 ≥
indicate evidence of low, moderate, and high heterogeneity, respectively.11 The σ2
estimates were low for all outcomes, except for any infection, overall complication,
mortality, and sepsis with moderate σ2 . Supplementary Table 10 shows the value of
median σ2 with 95% Crl for all outcomes.
Publication bias
Publication bias was assessed visually using a netfunnel command—a
comparison-adjusted funnel plot—implemented in software package Stata version
12.1(Stata Corporation, College Station, Texas). In a pair-wise meta-analysis, a funnel
plot is a scatter plot of standard error vs. effect estimates for each study. Different
treatment comparisons have their own summary estimates in network meta-analysis;
therefore, there is not a common line of symmetry for all the studies. In the
comparison-adjusted funnel plot, the horizontal axis is each study's i observed logOR
(𝑦𝑖𝑋𝑌 ) centered to the comparison’s mean logOR obtained from the pairwise
meta-analysis (𝜇𝑋𝑌 ); and, the vertical axis is the inverted standard error of the effect
Supplementary Text R2 29
sizes as used in a standard funnel plot. If all studies appear symmetrically around the
zero line, the comparison-adjusted funnel plot suggests no evidence of small-study
effects in the network.12This result is shown in Supplementary Figure 4, which shows
that there is no evidence of publication bias for any outcome because we cannot be sure
if it is present or not using these methods.
References
1. Moher D, Liberati A, Tetzlaff J, et al. Preferred reporting items for systematic
reviews and Meta-analysis: The PRISMA statement. PLOS Med. 2009; 6:
e1000097.
2. Higgins JPT, Altman DG. Assessing risk of bias in included studies. In: Higgins
JPT, Green S, eds. Cochrane handbook for systematic reviews of interventions.
England: John Wiley & Sons Ltd, 2008:187–241.
3. Smith TC, Spiegelhalter DJ, Thomas A. Bayesian approaches to random-effects
meta-analysis: a comparative study. Stat Med. 1995;14:2685–2699.
4. Dias S, Welton NJ, Sutton AJ, et al. NICE technical support document 2: A
generalized linear modeling framework for pairwise and network meta-analysis of
randomized controlled trials. Available at: http://www.nicedsu.org.uk/. Accessed
February 14, 2013.
Supplementary Text R2 30
5. Dias S, Sutton AJ, Welton NJ, et al. NICE technical support document 3:
Heterogeneity: subgroup, meta-regression, bias and bias-adjustment. Available at:
http://www.nicedsu.org.uk/. Accessed February 14, 2013.
6. Salanti G, Ades AE, Ioannidis JPA. Graphical methods and numerical summaries
for presenting results from multiple-treatment meta-analysis: an overview and
tutorial. J Clin Epidemiol. 2011;64:163–171.
7. Cipriani A, Barbui C, Salanti G, et al. Comparative efficacy and acceptability of
antimanic drugs in acute mania: a multiple-treatments meta-analysis. Lancet.
2011;378:1306–1315.
8. Lu G, Ades AE. Assessing evidence inconsistency in mixed treatment comparisons.
J Am Stat Assoc. 2006;101:447–459.
9. Salanti G. Indirect and mixed-treatment comparison, network, or multiple
treatments meta-analysis: many names, many benefits, many concerns for the next
generation evidence synthesis tool. Res Syn Meth. 2012;3:80–97.
10. Veroniki AA, Vasiliadis HS, Higgins JPT, et al. Evaluation of inconsistency in
networks of interventions. Int J Epidemiol. 2012;42:332–345.
11. Spiegelhalter DJ, Abrams KR, Myles JP. Bayesian approaches to clinical trials and
health-care evaluation. West Sussex: John Wiley & Sons Ltd, 2004.
Supplementary Text R2 31
12. Chaimani A, Salanti G. Using network meta-analysis to evaluate the existence of
small-study effects in a network of interventions. Res Syn Meth. 2012;3:161–176.