Systematic review and meta-analysis of contemporary pancreas surgery with arterial resection

The PubMed/Medline, Cochrane Library, CINAHL, ClinicalTrials.​gov (clinical trials registry), and WHO ICTRP (clinical trials registry) databases were searched for this study through their respective online search engines. The search was performed on studies published between January 2011 (as the meta-analysis that set the standard on this topic dates back to 2011) [3] and January 2020 (the number of studies including neoadjuvant therapies considerably increased by that time). The optimal literature search for systematic reviews in surgery was followed [9].The search strategies used in the single databases are displayed in Supplementary Table 1. Furthermore, the reference lists of the available studies were manually searched to find relevant articles. The last search was conducted on 09 January 2020. Abstracts and full-text reviews were evaluated independently in an unblinded standardized manner by two authors (AR and IB) in order to assess eligibility for inclusion or exclusion. Disagreements between reviewers were resolved by consensus; if no agreement could be reached, a third author (JP) decided if the respective study was to be included.

Studies in English assessing resection of pancreatic cancer in curative intent including resection of major visceral arteries (celiac trunk and/or SMA and/or common hepatic artery), with or without a control group of patients undergoing pancreatic resection without arterial resection, were included. Studies reporting on splenic artery resection were excluded [10]. Studies published before 2011 were also not included, as for example the study by Bockhorn et al. [11]. Studies with an irrelevant abstract or title or with less than five patients were excluded, as were reviews, case reports, comments, and letters. Studies with no differentiation between venous and arterial resection were also excluded. When studies from the same authors including the same patient population were found, the study with a control group was selected. When no control group was described, the studies with fewer patients were excluded. The studies by Klompmaker et al. [12, 13] were excluded as they included data from other single-center studies which were already included in our analysis. Details of the study selection process are summarized in the flowchart of Fig. 1.

After performing a qualitative analysis, only studies including only patients operated from 2000 on with a control group regarding standard resection with curative intention were selected for meta-analysis.

Studies were analyzed and data was extracted separately by two authors (AR and IB) and presented in a tabular fashion (Tables 1, 2, and 3). The following descriptive data was documented for each selected study: first author, year of publication, inclusion period, sample size, arterial resection, neoadjuvant therapy, country, and median follow-up (Table 1). The following patient and operation characteristics were documented: age, gender, type of operation, type of artery resected, duration of surgery, and blood loss (Table 2). The following clinically relevant outcomes for pancreatic surgery were also extracted: morbidity (any type of complication, surgical and medical), pancreatic fistula [45], delayed gastric emptying [46], postoperative bleeding (International Study Group of Pancreatic Surgery definitions) [47], reoperation rate, mortality, length of hospital stay, median survival, actuarial survival (1, 2, 3, and 5 years), R0 resection rate (if possible using the Royal College of Pathologists definition [48]), histologic arterial invasion, and lymph node positivity (Table 3). Risk of bias was accessed using the ROBINS-I tool (risk of bias in nonrandomized studies of interventions) [49] (Table 4). No relevant articles in languages other than English were found. No contact with authors was made.

A meta-analysis was performed with morbidity using the Clavien–Dindo classification [50], perioperative mortality, 1 year survival, R0 resection [48], postoperative pancreatic fistula (defined according to the International Study Group on Pancreatic Fistula) [45], and delayed gastric emptying (DGE) rates (defined by the International Study Group of Pancreatic Surgery) [46] as outcome measures (random-effects model and fixed-effects model) using the Review Manager (RevMan) software, version 5.3 (Cochrane Collaboration, Oxford, UK). The magnitude of the effect estimate was visualized by forest plots. An odds ratio (OR) was calculated for binary data. The 95% confidence interval (CI), heterogeneity, and statistical significance were reported for each outcome. The χ² and the Kruskal–Wallis tests were used for the evaluation of statistical significance. p < 0.05 was considered to be statistically significant. When the studies did not report mean and standard deviation, these were calculated using the methods described by the guidelines of the Cochrane Collaboration [51] and Hozo et al. [52]. As not all studies report individual patient data or hazard ratios, the survival analysis was performed with weighted rates. Furthermore, a subgroup analysis was performed based on the rate of the patients receiving neoadjuvant chemotherapy. Studies with more than 50% of patients receiving neoadjuvant chemotherapy were included in the neoadjuvant subgroup. No subgroup analysis on planned versus not planned resection and arterial reconstruction versus no reconstruction and arterial resection plus venous resection versus arterial resection alone was performed because no control group was available.

Among the 425 articles, 31 studies from 9 countries were included in the qualitative analysis (Tables 1 and 2). Publication years were from 2011 to 2019. The inclusion periods of patients ranged from 1970 to 2018. Within these 30 studies, a total of 841 patients underwent pancreatic surgery with arterial resection or reconstruction and 6270 patients underwent a procedure without arterial resection.

Among the AR groups from all studies, 50% of patients received neoadjuvant chemotherapy (data from 26/31 studies).

The overall morbidity rate was 66.8% (data from 29/31 studies) in the AR group versus 93% (data from 10/11 studies) in the no AR group (p < 0.001). The overall postoperative pancreatic fistula (POPF) rate (all grades combined) was 27% in the AR group (data from 26/31 studies) versus 14% in the no AR group (data from 10/11 studies, p < 0.001). Regarding DGE (all grades combined), a rate of 19% in the AR group (data from 15/31 studies) versus 13% in the no AR group (data from 7/11 studies, p < 0.001) was observed. Postoperative bleeding (all grades combined) was 12.6% in the AR group (data from 11/31 studies) versus 3% in the no AR group (data from 5/11 studies, p < 0.001). A reoperation was required in 11% of patients in the AR group (data from 14/31 studies) versus 4.6% of patients in the no AR group (data from 5/11 studies, p < 0.001). No grade differentiation could be performed for POPF, DGE, and postoperative bleeding because stratification was not available in all studies.

As the goal of the meta-analysis was to review contemporary pancreas surgery with arterial resection from the last 20 years, we excluded the studies from Takahashi et al. [37], Glebova et al. [22], and Yamamoto et al. [42] because of the inclusion period starting in 1993, 1970, and 1991, respectively. Moreover, the study from Addeo et al. [14] was not included, as it reports and compares the outcomes of surgical resection of borderline resectable (BL) and locally advanced (LA) “unresectable” pancreatic cancer after neoadjuvant chemotherapy and not directly an arterial resection group with a standard resection group. The study from Rehders et al. [33] was also excluded because it only reported less than five patients who underwent arterial resection. The control group from the Del Chiaro study [21] was excluded because it included patients undergoing palliative surgery.

From the 31 studies, seven cohort studies with a total of 579 patients were included in the meta-analysis. In the risk of bias assessment, only the Loveday et al. [25] study was classified as low risk (Table 3).

As depicted in Table 5, there were 110 patients in the AR group and 469 patients in the control group of the included studies. Thirty-eight percent of the patients who underwent arterial resection received neoadjuvant chemotherapy. Seventy-three percent of the patients in the AR group underwent distal pancreatectomy, 17% pancreaticoduodenectomy, and 10% total pancreatectomy. Ninety-one percent underwent a CA or hepatic artery branches (HAB) resection and 9% an SMA resection. Sixty percent of the arterial resections were planned and only 34% of all patients in the AR group underwent arterial reconstruction.

Regarding the duration of the operation, almost all studies demonstrated that pancreatic surgery with AR was longer than standard surgery. In the random-effects model, the operation time was shorter in the standard group with a mean difference of 98 min (95% CI [77.42, 116.96], p < 0.001) (Fig. 2a). In all included studies, blood loss was higher in the AR group with a mean difference of 319 mL (95% CI [150.02, 487.2], p < 0.001) (Fig. 2b). The study from Ham et al. was excluded as no SD of the mean was provided.

Five studies provided information about overall morbidity. A total of 42 patients who underwent pancreatic surgery with AR were included in this analysis. There was no statistically significant difference between the AR and no AR groups (OR 1.15, 95% CI [0.58, 2.28], p = 0.69) (Fig. 3). The overall morbidity rate was 48% in the AR group and 39% in the standard resection group (p = 0.1). In the subgroup analysis for neoadjuvant therapy, the results were not significantly different between the neoadjuvant group (OR 1.28, 95% CI [0.49, 3.32]) and the upfront surgery group (OR 1.12, 95% CI [0.35, 3.57], p = 0.86).

Regarding postoperative pancreatic fistula, there was no statistically significant difference in the analysis of 110 patients undergoing pancreatic surgery with AR and 469 undergoing standard surgery (OR 0.77, 95% CI [0.39, 1.52], p = 0.45) (Fig. 4a). DGE was assessed in four studies. There was no significant difference in patients receiving AR versus the standard procedure (OR 2.30, 95% CI [0.36, 14.57], p = 0.08) (Fig. 4b). Meta-analysis for postoperative bleeding was not performed because this outcome was only reported in two of the selected studies.

Six of the included studies reported data on perioperative mortality. Mortality was nonsignificantly higher in the AR group (OR 2.55, 95% CI [0.69, 9.42], p = 0.16) (Fig. 5). The weighted mortality rate was 3.2% in the AR group and 1.5% in the standard resection group (p = 0.27).

Five studies reported on R0/R1 rates although with heterogeneous definitions. Patients undergoing arterial resection had lower R0 resection rates compared to the no AR group (69% vs 89%, OR 0.24, 95% CI [0.11, 0.54], p < 0.001). In the subgroup analysis concerning neoadjuvant therapy, patients undergoing upfront surgery with AR showed lower R0 rates when compared with those undergoing standard surgery (50% vs 86%, OR 0.17, 95% CI [0.08, 0.36], p < 0.001). Patients undergoing arterial resection after neoadjuvant chemotherapy had no statistically significant difference of R0 rates than the ones undergoing standard resection (92% vs 92%, OR 1.04, 95% CI [0.08, 13.31], p = 0.98) (Fig. 6).

Four studies were included in the meta-analysis regarding lymph node involvement. Lymph node positivity was observed in 58% of the patients with AR and 60% in the standard group (p = 0.69). No significant difference was found in the meta-analysis of the four studies (OR 1.39, 95% CI [0.66, 2.92], p = 0.38) (Fig. 7).

Concerning 1-year survival, the meta-analysis showed no statistically significant difference between the two groups (78% vs 77%, OR 0.92, 95% CI [0.41, 2.09], p = 0.85) (Fig. 8). In the subgroup analysis for neoadjuvant therapy, there was no statistical significance neither in the neoadjuvant group (OR 0.47, 95% CI [0.16, 1.40], p = 0.18) nor in the upfront surgery group (OR 1.49, 95% CI [0.46, 4.88], p = 0.16) between the AR and the no AR groups.

In the qualitative analysis, the overall morbidity rate was significantly lower in the AR group. This may be explained by a high morbidity rate of 97% in the standard resection group of the study by Glebova et al. [22]. Here, there were 5591 patients in the control group, which reflects an overrepresentation. In fact, morbidity rates of 41.8–77.5% have been reported in two recent RCTs for major pancreatic resections [53, 54], which is similar to the observed 66.8% morbidity observed in the AR group.

In the meta-analysis, morbidity was not significantly increased in the arterial resection group. This may be attributed to different reasons. One is that the majority of studies in the meta-analysis reported results of the so-called modified Appleby procedure, e.g., distal pancreatectomy with common hepatic artery/celiac trunk resections. Risk of complications is lower in such resections because there is no need for reconstruction—as compared with pancreaticoduodenectomy, for example. In our own experience, morbidity may be significantly increased when performing arterial resections with reconstructions. Different approaches are internationally chosen to these procedures, such as avoiding a pancreatic anastomosis and rather performing total pancreatectomy. Because of this clinical perception, we have designed a multicentric exploratory study for a staged resection in cases with arterial infiltration or arterial stenosis (PREVADER study; protocol currently under review; also, see https://​clinicaltrials.​gov/​ct2/​show/​NCT04136769). Here, patients with borderline resectable or locally advanced pancreatic ductal adenocarcinoma (PDAC) undergo a first operation—called a visceral debranching procedure—aiming at bypassing the arterial invasion/stenosis, followed by neoadjuvant chemotherapy. If there is no tumor progression (and patency of the arterial bypass) on restaging, patients undergo re-exploratory laparotomy and—if possible—tumor resection. In the second stage, arterial resection can be performed en bloc with the tumor resection without the need of arterial reconstruction.

Unlike the past meta-analysis on this topic [3], we observed no statistically significant difference between the arterial resection and standard surgery groups, although the weighted mortality rate was 3.2% in the arterial resection group and 1.5% in the standard resection group. Possibly, the statistical power of our analysis was not sufficient for such a small absolute difference. Nevertheless, our analysis reports relevantly lower absolute mortality rates compared with those reported by Mollberg et al. of 11.8% [3]. Mortality was differently defined across the studies included in this meta-analysis (in hospital, 30-day and 90-day mortality). Also, these low mortality rates suggest that only highly selected patients were included, which is supported by mortality rates in three recent RCTs of 3–10% [53‐55]. The results are comparable to a recent work from Klompmaker et al. [12]. In an analysis of 240 patients, in whom distal pancreatectomy with celiac axis resection (DP-CAR) was combined with (neo-)adjuvant chemotherapy, the 90-day mortality rate was 3.5% [12]. On the other hand, a retrospective cohort study from the same group included 68 patients from 20 hospitals in 12 countries and reported 30-day and 90-day mortality rates after DP-CAR of 10% and 16%, respectively [13].

There is a high risk of bias because patient selection may explain these significant differences between studies and may not reflect the reality even in high-volume centers with less stringent patient selection. Nevertheless, in the light of the presented data, pancreatic surgery with arterial resection can be performed in selected patients with reasonable mortality rates.

Morbidity could not be stratified in grades according to the Clavien–Dindo classification owing to nonavailability of pertinent data. Concerning morbidity, slightly better results could be observed in the standard resection group, but no significant difference was shown. Of note is the significant heterogeneity among the included studies.

Regarding postoperative pancreatic fistula, which is the major cause of morbidity after pancreatic resection, no significant difference was observed between the two groups. It is assumed that most studies reported clinical relevant fistula even though a differentiation of CR-POPF (grade B/C) from POPF-A/biochemical leak was not always made across the included studies. This is supported by CR-POPF rates in three recent RCTs ranging from 6 to 29% as compared with our data of 27% versus 14% [53‐55].

Interestingly, lower fistula incidence was reported among the patients with arterial resection, probably due to more advanced tumors and hence more solidified postobstructive fibrosis [56].

Planning the arterial resection may have a positive effect on postoperative morbidity as compared to unplanned resections [31]. Meticulous assessment of preoperative CT scans for detection of possible arterial encasement is therefore essential.

According to our meta-analysis, arterial resection in selected patients does not have a relevant effect on perioperative morbidity, especially on postoperative pancreatic fistula or delayed gastric emptying.

This data has to be interpreted carefully because of the high risk of bias in the included studies as only studies including AR were searched for.

The second important aspect emerging from our meta-analysis is that pancreatic surgery with arterial resection is associated with a lower 1-year survival rate than surgery without arterial resection. In a large systematic review and meta-analysis involving 18 studies, DP-CAR had a better 1-year survival rate compared to palliative treatments (pooled HR for OS 0.38 (95% CI 0.25–0.58, p < 0.01) [57]. These results are comparable to the weighted average of 1-year survival in our analysis, which was 78% in the AR group. In a study from the Dutch Pancreatic Cancer Group that involved a national cohort of 36,453 patients with PDAC, the 1-year survival of patients who received palliative chemotherapy improved along the last years from 13.3 to 21.2% (p < 0.001) [58]. This data is only partial comparable to ours, because this cohort also included patients with advanced metastatic disease (n = 4074). According to this data, arterial resection can be considered in selected patients instead of palliative chemotherapy.

Considering weighted median survival in both groups, patients undergoing arterial resection had a worse prognosis with 18.6 months compared with 32 months in patients undergoing surgery without arterial resection. In a recent systematic review that included 240 patients undergoing DP-CAR, the weighted median survival (14.4 months) was comparable to our analysis [12]. Our data are also comparable to data from a multicenter retrospective cohort study regarding patients undergoing arterial resection (18 months median survival) [13].

This finding may be attributed to unfavorable tumor biology with more advanced tumor stages and more aggressive growth in tumors affecting visceral arteries and requiring arterial resection. Longer operative time, lower R0 resection rates, and fewer patients with neoadjuvant treatment in the arterial resection group are additional factors. Nevertheless, arterial resection patients have better prognosis when compared to those undergoing palliative treatments.

In our analysis, despite heterogeneous definitions across the studies, the R0 resection rate was significantly lower in the patients undergoing pancreatic surgery with arterial resection. This can be explained by the local extent of tumor growth and hence the surgical complexity in the arterial resection group. Moreover, neoadjuvant treatment was associated with higher R0 resection rates. Kluger et al. showed in their study that a tumor-free resection margin could be achieved in 80% of the cases after neoadjuvant treatment in locally advanced pancreatic cancer with arterial encasement [59]. In the present series, the weighted average of R0 resection in arterial resection patients within the neoadjuvant subgroup was 92% compared with 50%, in the patients undergoing arterial resection without neoadjuvant chemotherapy (p < 0.001). Although neoadjuvant chemotherapy was not associated with prolonged long-term survival, our analysis suggests that it is crucial for achieving negative resection margins in this setting [12, 14, 59]. It has to be emphasized that in the analysis of neoadjuvant therapy, cohorts were grouped according to the predominantly administered therapy. Thus, for example, in the neoadjuvant therapy group, the majority, but not all patients, received neoadjuvant therapy.

The role of alternative therapies like chemoradiotherapy or staged resection needs further investigation. In a phase 3 randomized trial involving 449 patients that underwent either chemotherapy alone or radiochemotherapy, no significant difference between overall survivals of both groups was observed [60]. In a multicenter phase II trial in four hospitals in the Netherlands that enrolled 50 patients, stereotactic body radiotherapy was reported as feasible, and in 12% of the patients, a potentially curative resection could be performed [61]. In another retrospective cohort study, neoadjuvant radiotherapy was associated with increased pathologic downstaging and R0 resection rates [62].

The main drawback of this meta-analysis is that it is based exclusively on nonrandomized and partially retrospective studies. Among these studies, twenty-one had less than 30 patients with arterial resections and only two single-center studies included more than 100 arterial resections. Furthermore, only 34 patients had an arterial resection with arterial reconstruction. The MOOSE guidelines were followed to ensure transparency and standardized reporting, but the risk of bias is still considerable because of the nature of studies included in the meta-analysis. Furthermore, as most of the studies did not report individual patient data or hazard ratios, the survival analysis was performed with weighted rates or weighted median survival which is rather inaccurate surrogate measures for meta-analyses of survival outcomes [63]. Thus, the data should be carefully interpreted.