Perioperative and oncological outcomes following minimally invasive versus open pancreaticoduodenectomy for pancreatic duct adenocarcinoma

Background The outcomes of minimally invasive pancreaticoduodenectomy have not been adequately compared with those of open pancreaticoduodenectomy in patients with pancreatic ductal adenocarcinoma. We performed a meta‐analysis to compare the perioperative and oncological outcomes of these two pancreaticoduodenectomy procedures specifically in patients with pancreatic ductal adenocarcinoma. Methods Before this study was initiated, a specific protocol was designed and has been registered in PROSEPRO (ID: CRD42020149438). Using the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines, PubMed, EMBASE, Web of Science, Cochrane Central Register, and ClinicalTrials.gov databases were systematically searched for studies published between January 1994 and October 2019. Overall survival, disease-free survival, and time to commencing adjuvant chemotherapy were the primary endpoint measurements, whereas perioperative and short-term outcomes were the secondary endpoints. Results The final analysis included 9 retrospective cohorts comprising 11,242 patients (1377 who underwent minimally invasive pancreaticoduodenectomy and 9865 who underwent open pancreaticoduodenectomy). There were no significant differences in the patients’ overall survival, operative time, postoperative complications, 30-day mortality, rate of vein resection, number of harvested lymph nodes, or rate of positive lymph nodes between the two approaches. However, disease-free survival, time to starting adjuvant chemotherapy, length of hospital stay, and rate of negative margins in patients who underwent minimally invasive pancreaticoduodenectomy showed improvements relative to those in patients who underwent open surgery. Conclusions Minimally invasive pancreaticoduodenectomy provides similar or even improved perioperative, short-term, and long-term oncological outcomes when compared with open pancreaticoduodenectomy for patients with pancreatic ductal adenocarcinoma. Electronic supplementary material The online version of this article (10.1007/s00464-020-07641-1) contains supplementary material, which is available to authorized users.

1 3 [1]. Despite the application of various neoadjuvant and adjuvant treatment protocols, pancreaticoduodenectomy (PD) remains the only curative treatment for patients with adenocarcinoma of the pancreatic head. PD is one of the most complex procedures in gastroenterological surgery, and requires extensive visceral organ dissection and complex reconstructive digestive anastomoses; therefore, it is normally performed using an open approach.
The first laparoscopic PD (LPD) was reported by Gagner and Pomp in 1994 [2], while the first robotic PD (RPD) was described by Giulianotti et al. in 2003 [3]. Since then, minimally invasive PD (MIPD), which encompasses LPD and RPD, is increasingly used worldwide; however, it remains a challenging procedure because of the technical limitations of laparoscopy or robot control as well as the steep learning curve when training physicians. Therefore, the feasibility and safety of the minimally invasive approach remain controversial.
Recent studies [4,5] and meta-analyses [6,7] showed that the outcomes of MIPD were similar to or more favorable than open PD (OPD) with respect to the incidence of postoperative morbidity, short-term oncologic outcomes, and long-term overall survival rates. Furthermore, MIPD is associated with lower estimated intraoperative blood loss, shorter length of hospital stay (LOS), higher rate of R0 resection, and the harvesting of a greater number of lymph nodes. However, these studies involved patients with a variety of disease histologies; therefore, objective conclusions regarding the oncologic outcomes of patients with this malignancy, especially PDAC, were unclear. Other related meta-analyses [8,9] involved relatively few patients with PDAC. According to our search, no meta-analysis was performed specifically to investigate the perioperative, shortterm, and long-term oncological outcomes of patients who underwent MIPD for PDAC. In this study, we carefully screened and selected studies that specifically investigated patients with PDAC. The aim of our meta-analysis was to meaningfully assess the perioperative, short-term, and longterm oncologic outcomes of these patients, with the primary investigative endpoints being overall survival (OS), diseasefree survival (DFS), and the time to starting postsurgical adjuvant chemotherapy.

Materials and methods
Before this study was initiated, we designed a specific protocol which has been registered in PROSEPRO. The ID is CRD42020149438 (details of registration is included in Supplementary Materials). Thus, this study was performed in accordance with the recommendations of the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines [10]. This article does not contain any studies with human participants performed by any of the authors; therefore, there was no requirement for IRB approval.

Data sources and search methods
We systematically searched the PubMed, EMBASE, Web of Science, Cochrane Central Register, and ClinicalTrials.gov databases for studies published in English between 1994 and October 2019. The search Medical Subject Heading (MeSH) terms were "laparoscopic pancreaticoduodenectomy" OR "robotic pancreaticoduodenectomy" AND "pancreatic ductal adenocarcinoma," as well as all associated entry words retrieved using the MeSH index (details of our search strategy are included in Supplementary Materials). There was no language restriction. We also reviewed the introduction and discussion sections of the retrieved manuscripts, relevant review articles, and published meta-analyses to identify additional trials. Two authors (Rui Sun and Jiawen Yu) independently conducted the literature search, screened the abstracts, and selected the relevant trials.

Inclusion criteria
Studies published between January 1994 and October 2019 were considered eligible if they met the following inclusion criteria: (1) the investigated population comprised patients with PDAC; (2) the interventions compared were LPD or RPD versus OPD; and (3) the study investigated at least one of the following outcomes: operative time, intraoperative blood transfusion, postoperative morbidity and mortality, LOS, rate of vein and R0 resection, number of lymph nodes retrieved, time to starting adjuvant treatment, DFS, and OS.

Exclusion criteria
The exclusion criteria are as follows: (1) review articles; (2) meeting abstracts; (3) irrelevant studies such as those that investigated only a single surgical technique; (4) insufficient information available in the English abstract; (5) studies with no comparative data; and (6) studies without PDAC data. If papers had overlapping data, those describing the smaller-scale studies were excluded.

Quality assessment
We adopted the Newcastle-Ottawa Scale (NOS) [11], which is designed specifically for observational investigations, to assess the quality of the selected studies. The NOS focuses on 3 separate sections of a case-control or cohort study, with the number of stars representing the assessment score. The maximum achievable score under the NOS is 9 stars, including 4 for the selection process, 2 for comparability, and 3 for 1 3 exposure and outcome. A score of ≥ 6 stars is considered indicative of high quality. Two investigators independently assessed the selected studies.

Data extraction
Two investigators independently extracted the following information: first author, year of publication, study type, mean age, population size, tumor size, and main outcomes; the latter included operative time, intraoperative blood transfusion, postoperative morbidity and mortality, LOS, rate of vein and R0 resection, number of lymph nodes retrieved, time to adjuvant treatment, DFS, and OS. The evaluators resolved any disputes via consensus during the screening processes.

Statistical analysis
Statistical analyses were conducted using Review Manager 5.3 (Cochrane Collaboration). The relative risk (RR) and mean difference (MD) with the 95% confidence interval (CI) were used as the measures of dichotomous and continuous variables, respectively. Natural logarithm hazard ratios (HRs) and standard errors were used as summary statistics for DFS and OS data. Using the methods of Wan et al. [12] and Luo et al. [13], medians with ranges as well as the mid-quartile range were converted into means with standard deviations. For studies in which natural logarithm HRs and standard errors or corresponding 95% CIs were not available, estimates from the published survival curves were calculated using the method suggested by Tierney et al. [14]. P-values less than 0.05 indicated statistical significance. Heterogeneity was quantified by the I 2 statistic; a study with an I 2 less than 50% was considered to have no heterogeneity, and the fixed effects model was then applied to pool the results; otherwise, the random effects model was used.

Search results and characteristics of the included studies
A flowchart of our analysis protocol is shown in Fig. 1. Our analysis included 9 retrospective cohort studies [15][16][17][18][19][20][21][22][23] comparing MIPD and OPD in a total of 11,242 patients with PDAC (1377 and 9865 underwent MIPD and OPD, respectively). The characteristics and qualities of these 9 studies are listed in Table 1, while the main results are shown in Table 2. All the analysis results including operative time, intraoperative blood transfusion, LOS, postoperative complications and mortality, and short-and long-term oncological outcomes are shown in Figs. 2, 3, 4, and 5. Among the 9 studies, 8 [15,[17][18][19][20][21][22][23] compared LPD and OPD and only 1 [16] compared RPD and OPD. The means of the patients' ages in each study ranged from 65.4 to 79.6 years with a majority in both the MIPD and OPD cohorts being of comparable age; only patients of both groups in Chapman et al.'s study (79.6 versus 79.6 years; P = 0.99) [22] were older than those in other studies. There were no significant differences between the proportions of male and female patients with PDCA.

Operative time
Three studies [15,21,23] that encompassed 913 patients investigated operative times (including 228 and 685 patients who underwent MIPD and OPD, respectively). The operative time was longer for the MIPD group in 2 of these studies, and was longer for the OPD group in the third. The pooled estimates of these studies showed that the duration of surgery was not significantly different between the MIPD and OPD groups (MD 50.09; 95% CI − 25.21 to 125.38; P = 0.11). The analysis found statistically significant heterogeneity (I 2 = 97%); thus, a random effects model was adopted ( Fig. 2A). No differences in the results and no heterogeneity were found on sensitivity analysis.

Intraoperative blood transfusion
Three studies encompassing 913 patients (228 who underwent MIPD and 685 who underwent OPD) compared intraoperative blood transfusion rates [15,21,23]. The pooled results showed a higher rate of intraoperative blood transfusion in the OPD group; the pooled RR (0.58; 95% CI 0.44 to 0.77; P = 0.0002) showed a significant difference in the intraoperative blood transfusion rate between the two groups. Heterogeneity was not significant (I 2 = 0%) (Fig. 2B).

LOS
Six trials [15,17,[20][21][22][23] with a total of 10,863 patients (1319 and 9544 who underwent MIPD and OPD, respectively) investigated the LOS. Five studies showed the LOS to be significantly lower in the MIPD group, whereas 1 showed it to be lower in the OPD group. The analysis found significant heterogeneity (I 2 = 98%), and a random effects model was adopted. The pooled mean difference (MD = − 4.66; 95% CI − 9.13 to − 0.19; P = 0.04) indicated a significantly shorter LOS in the MIPD group (Fig. 2C). The results and heterogeneity were not significantly different in the sensitivity analysis.

Time to starting adjuvant chemotherapy after surgery
Four studies [15,[20][21][22] gathered data on the time to starting adjuvant chemotherapy after surgery, including a total of 10,494 patients (1242 and 9252 underwent MIPD and OPD, respectively). The pooled results indicated that the time to adjuvant chemotherapy was significantly shorter in the MIPD group (MD = − 15.35; 95% CI − 30.33 to − 0.36; P = 0.04) (Fig. 4E). The heterogeneity was high (I 2 = 98%), and a random effects model was adopted.

Publication bias
Begg's funnel plot was used to assess any publication bias present in the articles. As shown in the funnel plot of OS (Fig. 6), no evidence of significant publication bias was found.

Discussion
To our knowledge, this is the first meta-analysis designed to specifically evaluate the perioperative and oncological outcomes of patients with PDAC who underwent MIPD and compare the findings with those for patients who underwent OPD. Overall, the pooled results revealed no significant difference in OS between the two groups, although patients who underwent MIPD had longer DFS periods and commenced adjuvant chemotherapy sooner after surgery. Moreover, we found no significant difference in operative time, postoperative complications, 30-day mortality, vein resection, number of harvested lymph nodes, or number of positive lymph nodes between the two approaches. There was a significant improvement in margin-negative resection with MIPD, as well as a shorter LOS and a lower intraoperative blood transfusion volume. Thus, our findings suggest that the outcomes of MIPD are at least equivalent to, if not better than, those of OPD in patients with PDAC.
It was previously thought that operating time would be longer with MIPD; this method is certainly challenging and has a steep learning curve. A recent retrospective multicenter analysis by Wang et al. [24] found that the learning curve for MIPD had 3 phases, with peaks evident after the completion of 40 and 104 cases. They also concluded that the operative time significantly decreased from phase I to phase III and was even comparable with that of OPD, consistent with the findings of van Hilst et al. [25] and Boone et al. [26]. In the present study, no significant difference between the 2 approaches was observed in terms of operative time. Croome et al. [15] excluded 10 patients who underwent MIPD in the first 6 months to avoid the early segment of their learning curve, which may have introduced bias with regard to our pooled results; however, we found no significant change in our results when excluding their study from our analysis. Therefore, our findings suggest that the operating time for MIPD is comparable to that of OPD once the learning curve is achieved.
The feasibility and safety of MIPD have been established given that enthusiasm for this procedure continues to grow quickly. MIPD has the potential to markedly reduce intraoperative blood loss and transfusion, as well as LOS; these attributes were corroborated in our own study. Although there were no sufficient data for analyzing intraoperative blood loss, the rate of blood transfusion was lower in patients undergoing MIPD in the present study, which ought to reflect lower intraoperative blood loss. Allogeneic blood transfusion has been confirmed to be an independent negative predictor of DFS and OS among patients with PDAC who underwent PD [27,28]. Therefore, MIPD may potentially be related to improved survival compared to OPD in patients with this disease. A shorter LOS could also translate into faster recovery, which may help introduce subsequent treatments sooner. LOS is also an important component of healthcare costs [29], and whether opting for MIPD can significantly reduce such costs over OPD is an important area of active investigation [30]. In the present study, total and major complications attributed to MIPD occurred at a similar rate to those attributed to OPD, including postoperative pancreatic fistulae, delayed gastric emptying, postpancreatectomy hemorrhage, and 30-day mortality. Other studies have also observed such outcomes [25,31,32]. Therefore, the minimally invasive approach does not appear to alter the risk of severe complications and death compared to OPD. This is highly important because most resections are performed to treat malignant diseases. Moreover, Stauffer et al. showed that major complications may be an independent predictor of poorer survival [21]. It is highly probable that major postoperative complications would at least delay the administration of subsequent chemotherapy or limit the patient's ability to tolerate a full course.
Our study also found that surrogate oncologic indicators including the number of resected veins, number of harvested lymph nodes, and rate of positive nodes were similar in the 2 groups, although the rate of R0 resection was higher in the MIPD group than in the OPD group. MIPD was contraindicated in patients who required vascular resection and reconstruction in a previous study [33]. However, Croome et al. [34] suggested that MIPD with major vascular resection is feasible and safe, and it can achieve similar results in terms of morbidity and mortality rates as well as of oncologic outcomes as does OPD with major vascular resection. The latter notion is supported by our data. Nevertheless, the acquisition of considerable experience for performing OPD both with and without vascular resection and for performing MIPD without vascular resection is strongly recommended before attempting minimally invasive major vein resection and reconstruction [35]. Previous studies found that the lymph node and margin statuses were significant predictors of DFS and OS [21,36]. Our pooled results revealed that MIPD  Begg's funnel plot for assessing publication bias produced oncologic outcomes similar to, or even more favorable than, OPD did. Moreover, Stauffer et al. [21] found that the proportion of positive lymph nodes in patients who underwent MIPD was lower than that in patients who underwent OPD, and that the negative margin rate was higher; however, neither finding was statistically significant. These results may indicate that the minimally invasive approach allows for complete and adequate lymphadenectomy and pancreatic resection as it provides superior and magnified views of the tumor.
The primary results of this study showed a significant improvement in the time to starting adjuvant chemotherapy and in DFS among patients with PDAC who underwent MIPD; Conrad et al. [31] reported similar outcomes in their study. According to a recent evaluation of the European Study Group for Pancreatic Cancer, completion of all 6 cycles of adjuvant chemotherapy following resection of PDAC improves OS if the chemotherapy is started within 12 weeks [37]. Croome et al. [15] reported that the lack of chemotherapy within 90 days of surgery was a strong predictor of poorer OS; this suggests that the inability to initiate or complete chemotherapy in a timely manner will ultimately have a negative influence on survival. Our data do indicate a longer DFS in the MIPD group than in the OPD group, which may reflect the difference in the timing of adjuvant treatment initiation. Therefore, we posit that MIPD may enhance the ability of patients to receive subsequent treatments in a timely fashion after surgery and to complete their dosing schedules as well.
Another primary result in our study was the lack of a significant difference in OS between the 2 approaches. Chapman et al. [22] reported that patients undergoing MIPD had a noticeably longer median survival time than did those undergoing OPD (19.8 vs. 15.6 months) as well as an improved OS (HR 0.85, 95% CI 0.69 to 10.3) after adjusting for patient-and tumor-related characteristics; these findings were similar to those of Croome et al. [15] and Conrad et al. [31]. Furthermore, Stauffer et al. [21] also found that the 5-year survival rates of patients who underwent MIPD and OPD were 32% and 15.34%, respectively. However, the differences in OS in these aforementioned studies were not significant. It remains to be determined whether MIPD can result in a significantly different OS among patients with PDCA as additional patients are accrued and follow-up periods are lengthened. Nevertheless, in our opinion, one of the key objectives for achieving superior oncologic outcomes with existing treatment modalities is to improve the ability of patients to receive both complete tumor resection and chemotherapy, and we believe that MIPD is a realistic method to meet this objective.
Two other meta-analyses that were similar to ours were identified in our search [8,38]. Although the stated aims of both were to assess LPD versus OPD in patients with PDAC, they included patients with ampullary carcinoma, some benign tumors, and even chronic pancreatitis in addition to those with PDACs. Moreover, the pooled results of short-term oncological outcomes such as R0 resection and number of lymph nodes retrieved were different from ours, since we only included patients with PDAC. Even though their findings regarding OS were similar to ours, they found 4 of the studies that were also included in our analysis [15,[20][21][22] to be heterogeneous (although we did not), whereas 8 other studies were not heterogeneous in their analyses (as consistent with our data). Consequently, their meta-analyses may prevent achieving objective conclusions regarding the oncologic safety of MIPD.
There were some limitations to our study that should be taken into account when considering the results. First, all included studies were retrospective and conducted in high-volume hospitals. Numerous studies have shown that hospital volume is significantly correlated with the incidence of perioperative outcomes [24]. Kantro et al. [20] reported that the 30-day mortality rate for patients who underwent MIPD was higher in low-volume hospitals than in high-volume institutions; therefore, there was a risk of selection bias even though such confounders could not be avoided. Second, the numbers of patients in some of the studies were too small, leading to low-power analyses. Third, only one of the included studies [16] compared RPD and OPD, which may have introduced bias because robotic and laparoscopic approaches have substantial differences, and the number of patients who underwent RPD was small and accounted for a small proportion in the analysis of results regarding the R0 resection rate, number of harvested lymph nodes, rate of positive nodes, and OS. However, the results and heterogeneity for all these variables showed no significant change after exclusion of the study comparing RPD and OPD. Nevertheless, some of our conclusions, particularly those pertaining to RPD, should be interpreted with caution. In the future, when more data are available, subgroup analysis should be performed to determine whether is any difference between the laparoscopic and robotic approaches. Finally, the heterogeneity in some of the results was high. Although we conducted sensitivity analyses, the results and heterogeneity were barely changed. Thus, some of our results should be interpreted with caution. Overall, additional prospective and multicenter randomized controlled trials with longer follow-up periods are warranted to compare the oncological outcomes of MIPD and OPD.
In conclusion, the results of our meta-analysis indicate that MIPD appears to be safe and feasible with perioperative, short-term, and long-term outcomes that are similar to those of OPD in the setting of PDAC. Moreover, MIPD may also provide advantages such as lower transfusion rates, higher rates of margin-negative resection, shorter time to starting 1 3 adjuvant chemotherapy, and longer DFS when compared to OPD in patients with PDAC.