Introduction

Following pancreaticoduodenectomy (PD), delayed gastric emptying (DGE) remains one of the most frequent postoperative complications ranging between 14 and 40% [1,2,3,4], associated with increased hospital length of stay [5], need for parenteral nutrition, increased hospitals costs [6,7,8] and hospital readmissions [9]. The exact mechanism of DGE after PD is not well understood however the configuration of gastroenteric reconstruction may influence the likelihood of developing DGE [10,11,12].

Several gastroenteric reconstruction techniques are currently in use including antecolic Billroth II reconstruction, retrocolic Billroth II reconstruction and a Roux-en-Y reconstruction route of gastroenteric anastomosis [13,14,15]. The Roux-en-Y reconstruction is frequently used with a view to isolate the pancreatojejunostomy [13, 14] and the gastroenteric anastomosis [15,16,17], hence attempting to reduce incidence of postoperative pancreatic fistula as well as DGE. Previous meta-analyses comparing antecolic and retrocolic reconstruction showed reduced rates of DGE with an antecolic gastroenteric anastomosis [13, 14]. Further RCTs comparing Roux-en-Y reconstruction with antecolic reconstruction showed comparable DGE rates between the two reconstruction techniques [16, 17]. In both studies, the gastroenteric anastomosis was performed through an antecolic route.

To date, it remains unclear about the best technique for gastroenteric reconstruction following PD to reduce DGE. This systematic review and network meta-analysis aimed at comparing different gastroenteric reconstruction techniques to identify the technique with the lowest rates of DGE.

Methods

Search strategy

The study was registered with the PROSPERO database at inception (Registration CRD42019 135718). The PubMed, Embase, MEDLINE and Cochrane library databases were searched using the PRISMA [18] framework by two independent investigators (SKK and SP). The search was completed on 04 May 2019. The search terms used were ‘delayed gastric emptying’ and ‘trials’, ‘randomised’ or ‘randomised controlled trial’, and ‘Whipple procedure’ or ‘Pancreaticoduodenectomy’ or ‘Whipple surgery’ individually or in combination. Supplementary Table 1 summarises the search terms used. A ‘related articles’ function was used to capture recent reviews on the topic, and a further manual search of reference lists was carried out to make the search comprehensive.

Inclusion and exclusion criteria

Inclusion criteria were: (1) randomised controlled trials (RCTs) comparing gastroenteric reconstruction technique in patients undergoing PD, where overall DGE was the primary endpoint and (2) studies published in the English language. Exclusion criteria were: (1) non-randomised controlled studies and case reports and (2) publications with a heterogeneous patient population with a combination of other types of pancreatic resections, i.e. distal pancreatectomy which could not be separated from those of patients undergoing PD. The identified abstracts from the search were reviewed by two authors (SKK and SP). Full-text article review of potentially relevant abstracts was conducted. A further manual search of the references of the included articles was performed to identify potentially relevant studies. Any disagreement regarding the included articles between the two authors (SKK, SP) was resolved after discussion with the co-authors.

Study outcomes

The primary outcome was overall delayed gastric emptying (DGE) defined according to the International Study Group of Pancreatic Surgery (ISGPS) [4]. Secondary outcome measures were grade B/C DGE, time to starting solids and liquids, duration of nasogastric tube, overall complications, overall (grade A, B and C pancreatic fistula) [19] and grade B/C pancreatic fistula [20], postoperative pancreatic haemorrhage [21], intraabdominal abscess and length of stay.

Data extraction

One author (SKK) retrieved data on study characteristics, patient demographics, types of reconstruction techniques and relevant clinical outcomes. The drawings of the methods of reconstruction for the techniques evaluated are presented in three included articles [16, 22, 23].

Assessment of methodological quality

The Cochrane Risk of Bias tool for randomised controlled trials was used to assess the methodological quality of included studies [24]. The Cochrane Risk of Bias tool is a domain-based evaluation which allows assessment of selection bias, performance bias, detection bias, attrition bias and reporting bias separately.

Statistical analysis

The statistical analysis of the network meta-analysis was carried out according to the recommendations of the Cochrane Library and PRISMA guidelines [18]. Odds ratios (ORs) were used for analysis of categorical variables. A random effects (DerSimonian–Laird method) model was preferred for the reporting the outcomes of the meta-analysis. Publication bias was assessed by visually inspecting the funnel plots of included studies. A network meta-analysis comparing the type’s reconstruction techniques was conducted by pooling the included studies. A network map was obtained for each individual outcome with graphical representations of treatments (nodes) and comparisons (lines). The network maps with closed loops were then entered into the network analyses. Direct and indirect treatment effects were compared after inconsistency was excluded in the networks. This was performed by examining for overall inconsistency throughout the entire network and then by the fitting of node splitting models checking for loop inconsistency, within all three-way treatment comparison loops [25]. The null hypothesis was accepted when P values were > .05, assuming consistency and networks were entered into consistency modelling and network plots were generated. The network plots were further supplemented with interval plots of pooled effect estimates. P-score was used to rank the reconstruction techniques using the netmeta package, as reported previously to be similar to the surface under the cumulative ranking areas for all outcomes. The probability of superiority of each treatment was assessed using the P-score [26,27,28]. Statistical significance was considered when P < 0.05. R Foundation Statistical software (R 3.2.1) was used to perform the statistical analyses [29, 30].

Results

Study characteristics

Eight studies comparing all relevant gastroenteric reconstruction techniques after PD were included (Fig. 1). The baseline data on the randomised trials are presented in Table 1 and Supplementary Table 2. These eight studies contributed 761 patients for analysis. Six studies [23, 31,32,33,34,35] comprising 580 patients compared antecolic and retrocolic Billroth II reconstruction and two studies [16, 17] comprising 181 patients compared antecolic Billroth II with Roux-en-Y reconstruction. One study was a multi-centre study [23], while the others were single-centre studies [16, 17, 31,32,33,34,35]. Six studies reported pylorus preserving PD (PPPD) [17, 23, 31, 33,34,35], of which in four studies all patients had PPPD [9, 17, 33,34,35]. The remaining two studies only included classical PD [32], and another only included subtotal stomach-preserving pancreaticoduodenectomy (SSPPD) [16]. The summary of the Cochrane Risk of Bias assessment is presented in Supplementary Table 3.

Fig. 1
figure 1

PRISMA diagram of included studies

Full size image
Table 1 Characteristics of included studies
Full size table

Delayed gastric emptying

Primary outcome measure

The results of all pairwise comparisons of the different reconstruction techniques for all postoperative outcomes are summarised in Table 2. Overall delayed gastric emptying was reported in all studies comprising 761 patients. Antecolic Billroth II reconstruction ranked first followed by Roux-en-Y reconstruction (Table 3). Grade B/C delayed gastric emptying was reported in six studies comprising 657 patients. Antecolic Billroth II reconstruction ranked first followed by Roux-en-Y (Table 3). A funnel plot for the risk of bias presented in Fig. 2 shows a symmetrical plot indicating minimal risk of bias.

Table 2 Summary of postoperative outcomes
Full size table
Table 3 Ranking of postoperative outcomes
Full size table
Fig. 2
figure 2

Funnel plot for risk of bias in main outcome for delayed gastric emptying

Full size image

Time to solid and liquid diets

Secondary outcome measures

Time to solid and time to liquid diets were reported in six and two studies, respectively; hence, the latter outcome was not included in the network meta-analysis. There were no significant differences in reconstruction techniques between antecolic Billroth II and retrocolic Billroth II (MD:  − 0.57 days, CI95%:  − 2.47 to 1.34, P = 0.6) and Roux-en-Y (MD:  − 3.50 days, CI95%:  − 7.41 to 0.41, P = 0.079) for time to solid diets. Antecolic Billroth II was ranked first followed by retrocolic Billroth II for time to solid diet (Table 3).

Duration of nasogastric (NG) tube

Duration of NG tube was reported in six studies. There were no significant differences in duration of NG tube in reconstruction techniques between antecolic Billroth II reconstruction and retrocolic Billroth II reconstruction (MD: 0.06 days, CI95%:  − 0.62 to 0.74, P = 0.9) for the duration of NG tube. Retrocolic Billroth II reconstruction ranked first followed by antecolic Billroth II for the duration of NG tube (Table 3).

Postoperative pancreatic fistula

Overall POPF was reported in all included studies. Roux-en-Y was ranked first followed by antecolic Billroth II for overall POPF (Table 3). Grade B/C POPF was reported in five studies. Roux-en-Y was ranked first followed by antecolic Billroth II for grade B/C POPF (Table 3).

Postpancreatectomy haemorrhage (PPH)

Overall postoperative pancreatic haemorrhage was reported in four studies. There were no significant differences in reconstruction techniques between antecolic Billroth II and retrocolic Billroth II (OR: 0.79, CI95%: 0.30–2.04, P = 0.7) and Roux-en-Y (OR: 1.76, CI95%: 0.33–9.36, P = 0.5) for the rates of PPH. Roux-en-Y reconstruction ranked first followed by antecolic Billroth II reconstruction for overall PPH (Table 3).

Overall complications

Overall complications were reported in three studies. There were no significant differences in reconstruction techniques between antecolic Billroth II and retrocolic Billroth II (OR: 1.20, CI95%: 0.67–2.16, P = 0.6) and Roux-en-Y (OR: 0.62, CI95%: 0.24–1.63, P = 0.3). Antecolic Billroth II was ranked first followed by Roux-en-Y for overall complications (Table 3).

Other postoperative outcomes

Bile leak, surgical site infection, reoperation, in-hospital mortality and length of stay were also evaluated as presented in Tables 2, 3. There were no significant differences in any of the reconstruction techniques on these outcomes. Antecolic Billroth II were ranked first for all these outcomes except reoperation for which Roux-en-Y was ranked first.

Discussion

This is the first network meta-analysis of randomised controlled trials to evaluate incidence of DGE after various gastroenteric reconstruction techniques following PD. This study only included RCTs where overall DGE was the primary endpoint allowing reliable assessment of well-powered RCTs. In this review, three reconstruction techniques were evaluated such as antecolic or retrocolic Billroth II and Roux-en-Y. Antecolic Billroth II reconstruction was found to be the best technique for overall and grade B/C DGE. This observation is a result of including RCTs where DGE was the primary endpoint, limiting inclusion of more studies where POPF was the primary endpoint [22, 36]. Antecolic Billroth II was also ranked first for other outcomes such as intraabdominal abscess, bile leaks, surgical site infection, in-hospital mortality and length of stay.

Several technical variations of gastroenteric reconstruction have been previously proposed based on the extent of gastroduodenal resection, such as PD with pyloric preservation [37], pyloric ring resection [10, 38] or subtotal gastric preservation [39], yet none have demonstrated superiority in reducing DGE. The most commonly reported reconstruction technique of anastomoses after PD technique is a single-loop reconstruction, whereby the pancreatojejunostomy, hepaticojejunostomy and the gastroenteric anastomosis with a Billroth II reconstruction are performed sequentially [40,41,42]. Furthermore, various modifications of the reconstruction to reduce the incidence of DGE have been described in the literature. A gastroenteric anastomosis with Roux-en-Y separating it from pancreatojejunostomy has been associated with lower rates of DGE [43]. Murakami et al. demonstrated lower rates of DGE after Roux-en-Y reconstruction compared to Billroth I [43]. Similarly, Barakat et al. also reported a lower incidence of DGE for Roux-en-Y compared with Billroth II reconstruction (10% vs 57%) [10]. Other modifications to the reconstruction including addition of a Braun enteroenterostomy to the Roux-en-Y reconstruction have shown a further reduction of DGE rates [44, 45]. However, these studies are non-randomised, limiting their validity. The two RCTs comparing Roux-en-Y reconstruction with Billroth II antecolic reconstruction reported conflicting results with one trial showing comparable DGE rates [16], while the other reported significantly lower DGE rates with Billroth II antecolic reconstruction compared to Roux-en-Y reconstruction (6% vs 20%).[17].

While DGE remains an important complication after PD, its aetiology remains largely unknown, and management is difficult. Published RCTs suggest a role for the type of gastroenteric reconstruction in the prevention of DGE. The present study appears to demonstrate that an antecolic Billroth II gastroenteric reconstruction after PD is associated with a lower incidence of postoperative DGE. The high incidence of clinically relevant DGE in patients developing a POPF after PD suggests that the prevention of such complications is paramount in reducing the incidence of DGE. The theoretical reason for increased DGE with retrocolic Billroth II technique is that the reduced motility may be secondary to decreased venous drainage of the jejunal limb from a retrocolic reconstruction which in turn can lead to oedema of the jejunal limb itself and the gastrojejunal anastomosis, thereby effecting gastric emptying [46, 47]. An antecolic gastrojejunostomy theoretically avoids the mechanical problems as the descending jejunal loop is likely to more mobile than a retrocolic reconstruction [48].

This study has several limitations to the present review. It only includes eight RCTs comparing three different techniques. This highlights the need for further studies evaluating the types of gastroenteric reconstruction in reducing the incidence of DGE. Furthermore, the studies included in this review differ in the method of PD used, i.e. classical PD, PPPD and SSPPD, which limits assessment of the best technique. This does emphasise the importance of standardising the resection technique in future RCTs.

Conclusion

In summary, this network meta-analysis highlights that single-loop antecolic Billroth II reconstruction appears to be the best technique to reduce the incidence of DGE after pancreatoduodenectomy.