Key Points
Question
Is the use of robotic pancreatoduodenectomy (RPD) noninferior to open pancreatoduodenectomy (OPD) in terms of clinically relevant pancreatic fistula occurrence?
Findings
In this propensity score–matched analysis of 304 patients, RPD demonstrated similar clinically relevant pancreatic fistula rates compared with OPD. Robotic pancreatoduodenectomy was also noninferior to OPD in terms of the occurrence of any complication, severe complications (Accordion severity grading system grade ≥3), hospital stay, 30-day readmission, and 90-day mortality.
Meaning
Robotic pancreatoduodenectomy is noninferior to OPD in terms of clinically relevant pancreatic fistula development and other major postoperative outcomes.
Abstract
Importance
The adoption of robotic pancreatoduodenectomy (RPD) is gaining momentum; however, its impact on major outcomes, including pancreatic fistula, has yet to be adequately compared with open pancreatoduodenectomy (OPD).
Objective
To demonstrate that use of RPD does not increase the incidence of clinically relevant pancreatic fistula (CR-POPF) compared with OPD.
Design, Setting, and Participants
Data were accrued from 2846 patients who underwent pancreatoduodenectomies (OPDs, n = 2661; RPDs, n = 185), performed by 51 surgeons at 17 institutions worldwide (2003-2015). All RPDs were conducted at a high-volume, academic, pancreatic surgery specialty center—in a standardized fashion—by surgeons who had surpassed the RPD learning curve. Propensity score matching was used to minimize bias from nonrandomized treatment assignment. The RPD and OPD cohorts were matched by propensity scores accounting for factors significantly associated with either undergoing robotic surgery or CR-POPF occurrence on logistic regression analysis. These variables included pancreatic gland texture, pancreatic duct diameter, intraoperative blood loss, pathologic findings of disease, and intraoperative drain placement.
Interventions
Use of RPD or OPD.
Main Outcomes and Measures
The major outcome of interest was CR-POPF occurrence, which is the most common and morbid complication following pancreatoduodenectomy.
Results
The overall cohort was 51.5% male, with a median age of 64 years (interquartile range, 56-72 years). The propensity score–matched cohort comprised 152 RPDs and 152 OPDs; all covariate imbalances were alleviated. After adjusting for potential confounders, undergoing RPD was associated with a reduced risk for CR-POPF incidence (OR, 0.4 [95% CI, 0.2-0.7]; P = .002) relative to OPD. Other predictors of risk-adjusted CR-POPF occurrence included soft pancreatic parenchyma (OR, 4.7 [95% CI, 3.4-6.6]; P < .001), pathologic findings of high-risk disease (OR, 1.4 [95% CI, 1.1-1.9]; P = .01), small pancreatic duct diameter (vs ≥5 mm: 2 mm, OR, 2.1 [95% CI, 1.4-3.1]; P < .001; ≤1 mm, OR, 1.8 [95% CI, 1.0-3.0]; P = .03), elevated intraoperative blood loss (vs ≤400 mL: 401-700 mL, OR, 1.5 [95% CI, 1.1-2.0]; P = .01; >1000 mL, OR, 2.1 [95% CI, 1.4-2.9]; P < .001), omission of intraoperative drain(s) (OR, 0.5 [95% CI, 0.3-0.8]; P = .005), and octreotide prophylaxis (OR, 3.1 [95% CI, 2.3-4.0]; P < .001). Patients undergoing RPD demonstrated similar CR-POPF rates compared with patients in the OPD cohort (6.6% vs 11.2%; P = .23). This relationship held for both grade B (6.6% vs 9.2%; P = .52) and grade C (0% vs 2.0%; P = .25) POPFs. Robotic pancreatoduodenectomy was also noninferior to OPD in terms of the occurrence of any complication (73.7% vs 66.4%; P = .21), severe complications (Accordion grade ≥3, 23.05% vs 23.7%; P > .99), hospital stay (median: 8 vs 8.5 days; P = .31), 30-day readmission (22.4% vs 21.7%; P > .99), and 90-day mortality (3.3% vs 1.3%; P = .38).
Conclusions and Relevance
To our knowledge, this is the first propensity score–matched analysis of robotic vs open pancreatoduodenectomy to date, and it demonstrates that RPD is noninferior to OPD in terms of pancreatic fistula development and other major postoperative outcomes.
This study examines the effectiveness of robotic pancreatoduodenectomy vs open pancreatoduodenectomy for clinically relevant pancreatic fistula occurrence and other major postoperative outcomes.
Introduction
The use of minimally invasive approaches has increased for a wide variety of procedures in general surgery. Some proposed benefits associated with minimally invasive organ resection include smaller incisions, reduced operative blood loss, and shorter recovery time. Its application to pancreatoduodenectomy (PD), however, has thus far been tempered for a number of reasons, including concerns regarding oncologic effectiveness, cost control, training issues, and technical challenges associated with the operation. Some of the technical drawbacks associated with laparoscopic approaches for PD may be ameliorated through robotic techniques, which offer 3-dimensional vision and advanced degrees of freedom and stability in tool handling. Some propose that this added precision can improve the execution of the reconstruction aspect of PD, particularly for the detailed creation of the pancreatic or biliary anastomoses.
The most common and morbid complication following PD is postoperative pancreatic fistula. Clinically relevant pancreatic fistulas (CR-POPFs) (International Study Group of Pancreatic Fistula [ISGPF] grades B and C) significantly alter the patient’s optimal recovery pathway and account for more than one-third of mortalities following PD. Since most robotic PDs (RPDs) are currently performed in select centers by specialized surgeons, comparative analyses of CR-POPF occurrence between robotic and open approaches have been limited to small, single-center retrospective cohort studies of variable quality. These studies have been unable to detect significant differences in CR-POPF occurrence between operative approaches; however, each study has been underpowered (< 50 RPDs), and robotic cases were often limited to specific subsets of selected patients.
Propensity score matching offers a method to minimize bias from nonrandomized treatment assignment and enables impartial comparisons of cohorts, which otherwise cannot be assessed in a randomized fashion. Applying this concept to the present study, the propensity score is a patient’s probability of undergoing RPD conditional on observed covariates (eg, pancreatic gland texture, duct diameter, pathologic findings of disease). By conditioning on the treatment score, this study would replicate some of the characteristics of a randomized clinical trial.
Owing to the current absence of level 1 evidence evaluating RPD and open PD (OPD), this study applied a propensity score–matching approach in a multicenter setting to demonstrate that RPD is noninferior to OPD in terms of CR-POPF development. Secondary end points included cohort comparisons in terms of the occurrence of any complication, mild to moderate complications, severe complications, 90-day mortality, 30-day readmission, and duration of hospital stay.
Methods
This study was approved by the institutional review boards at the University of Pennsylvania and the University of Pittsburgh Medical Center. In the overall series, 51 pancreatic surgeons contributed PDs from 17 high-volume, academic institutions. All OPDs were performed at 16 institutions from January 2003 to October 2015 by 48 surgeons who had surpassed the OPD learning curve. This learning curve has been previously defined as completion of 60 consecutive OPDs. Anastomotic techniques used in the OPD cohort were either end-to-side pancreaticojejunostomy (PJ) or pancreaticogastrostomy.
All RPDs were conducted at a single center (University of Pittsburgh Medical Center [UPMC]) from August 2011 to March 2015 by 3 surgeons (A.H.Z, M.E.H, and H.J.Z.) who had surpassed the RPD learning curve. This learning curve—as identified by inflexion points in several outcome metrics, such as operative time, conversion to open surgery, estimated blood loss, and POPF—has been previously defined by these surgeons as completion of 80 consecutive RPDs at UPMC since the inception of its robotic program in 2008. Selection criteria for RPD, by those 3 surgeons, has expanded significantly beyond the learning curve, such that the only contraindications to RPD during this study period were (1) the inability to tolerate pneumoperitoneum, (2) borderline resectable tumors that require an end-to-end venous reconstruction or conduit, and (3) unavailability of the robot owing to scheduling limitations or lack of “block time.” Within the study time period, a total 304 PDs were performed by the 3 surgeons at UPMC, of which 205 (67.4%) were approached robotically.
All RPDs were performed in a standardized fashion. Following laparoscopic exploration, the robot was docked and used for the entire resection and reconstructive phases of the procedure. The PJs were performed in a duct-to-mucosa fashion using a modified Blumgart technique. Briefly, this end-to-side PJ involves an outer (ie, pancreatic capsule to seromuscular jejunum) layer of 3 horizontally placed inverted mattress silk sutures. The same suture is used to create both the anterior and posterior outer layer of the anastomosis. Each suture begins by traversing the pancreas in full-thickness from its anterior to posterior surface, followed by a seromuscular bite of the jejunum, and back again to traverse the pancreas in full thickness fashion from the posterior to the anterior surface. The sutures are tied, but the needles are kept in situ so they can be used again for the anterior layer. An interrupted duct-to-mucosa anastomosis is then performed using 6 sutures of 5-0 absorbable material (PDS or Vicryl) over an internal stent (Hobbs Medical Inc). Finally, the same silk needles used for the posterior layer are used to create the anterior layer. All PJs were drained by a single anterior 10-mm Blake drain, and early drain removal (ie, on postoperative day 3 or 4) was used following serum and drain amylase measurements on postoperative day 3.
For both operative approaches, cases were included only if all 4 elements of the Fistula Risk Score (FRS), a validated risk assessment tool for CR-POPF, were documented. In addition, the placement of intraoperative drains, transanastomotic stents, and somatostatin analogs (ie, octreotide) were made at the surgeon’s discretion.
Pancreatic Fistula Classification
The primary outcome of interest was CR-POPF, which was graded in accordance with ISGPF standards. Transient, biochemical (ie, grade A) leaks were not studied owing to their lack of clinical impact on outcomes. Clinically relevant fistulas (ie, grades B and C) have clinically significant effects and significantly alter the patient’s recovery. Grade B POPFs are often treated with therapeutic agents, such as antibiotics, prophylactic somatostatin analogs, total parenteral nutrition, percutaneous drain placement, or prolonged operative drainage (ie, >3 weeks). Grade C POPFs are characterized by organ failure, reoperation, or death.
Assigning Fistula Risk
Fistula risk was determined using the validated, 10-point FRS. This established metric is based on the presence of certain risk factors for the development of CR-POPF: soft to normal pancreatic parenchyma, pathologic findings of high-risk disease (all pathologic abnormalities other than pancreatic ductal adenocarcinoma or pancreatitis), small pancreatic duct diameter, and elevated intraoperative blood loss. Individual scores are derived through the summation of each of the 4 weighted risk factors. Calculated scores are then discretized and assigned to 1 of 4 risk zones: (1) negligible risk, 0 points; (2) low risk, 1 to 2 points; (3) moderate risk, 3 to 6 points; or (4) high risk, 7 to 10 points.
Statistical Analysis
The study was designed to demonstrate noninferiority of RPD to OPD in terms of CR-POPF development. An overall CR-POPF rate of 15.0% was assumed for both groups, and the noninferiority margin was set at 10%. In addition, the α was set at .05 and β at .20, yielding a power of 80%. Therefore, to demonstrate that there is truly no difference between the standard (ie, OPD) and experimental (ie, RPD) treatment, then 316 patients would be required to be 80% sure that the upper limit of a 1-sided 95% CI—or equivalently a 90% 2-sided CI—would exclude a difference in favor of the standard group of more than 10%.
Continuous variables are expressed as mean (SD) or median (interquartile range [IQR]), while categorical variables are presented as absolute numbers and percentages. For univariate comparisons, χ2 analysis or Fisher exact tests were used to evaluate categorical variables; alternatively, continuous variables were analyzed using t test and Wilcoxon rank sum test for normally and nonnormally distributed data, respectively. Stepwise, backward logistic regression analysis identified covariates associated with the use of RPD (P ≤ .05 for entry; P > .10 for removal).
Propensity score matching is a method used to minimize treatment selection bias when estimating causal treatment effects in nonrandomized studies. “Control” (ie, OPD) and “case” (ie, RPD) sets are matched on a set of variables that would otherwise confound comparisons between them. Once a matched sample has been formed, the treatment effect can be estimated by directly comparing outcomes (eg, CR-POPF) between control and case patients in the matched sample. In the primary analysis (model 1), propensity scores were developed accounting for all factors significantly associated with either undergoing RPD or CR-POPF occurrence on logistic regression analysis. Accordingly, individual propensity scores were calculated through logistic regression modeling based on the following 7 covariates: pancreatic gland texture, pancreatic duct diameter, intraoperative blood loss, pathologic findings of high-risk disease (pancreatic ductal adenocarcinoma or pancreatitis vs other), intraoperative drain placement, octreotide prophylaxis, and transanastomotic stent placement. The OPD and RPD patients were then paired 1:1 on these propensity scores using exact matching. A standard caliper size of 0.2 × log [SD of the propensity score] was used. Standardized differences were estimated before and after matching to evaluate the balance of covariates; small absolute values (<0.1) indicate balance between treatment groups.
A secondary propensity score–matched analysis (model 2) was conducted accounting only for factors significantly associated with undergoing RPD. In this analysis, individual propensity scores were calculated through logistic regression modeling based on the following covariates: pancreatic gland texture, pancreatic duct diameter, and transanastomotic stent placement. The OPD and RPD patients were once again paired 1:1 on propensity scores using exact matching. The caliper size and subsequent analyses used to compare cohorts were the same as those used for the initial model.
Following 1:1 propensity score–matching, CR-POPF occurrence between matched OPD and RPD patients was examined by McNemar test. Secondary end points included mild to moderate complications (modified Accordion severity grading system grade 1-2), severe complications (modified Accordion grade ≥3), and any complication (modified Accordion grade ≥1). In congruence with CR-POPF follow-up, these were assessed out to 90 days following the index procedure. Other outcomes studied were duration of hospital stay, 30-day readmission, and 90-day mortality. All binary outcome comparisons between propensity score–matched OPD and RPD cohorts were carried out using McNemar test, while continuous outcomes used the paired t test or Wilcoxon rank-sum test. P ≤ .05 was considered statistically significant; all tests were 2-sided. All statistical computations were performed using SPSS statistical software (version 21.0; IBM Corp) with propensity score matching performed using the SPSS extension program developed by Felix Thoemmes.
Results
Factors Predicting Undergoing RPD
The overall cohort was 51.5% male, with a median age of 64 years (IQR, 56-72 years). Before propensity score matching, 2846 patients—185 in the RPD set (6.5%) and 2661 in the OPD set (93.5%)—met study criteria. Significantly different variables on univariate comparisons between RPD and OPD cohorts were entered into a multivariable model to identify predictors of undergoing RPD (Table 1). The following variables were independently associated with undergoing RPD: soft gland texture (odds ratio [OR], 2.3 [95% CI, 1.5-3.4]; P < .001), smaller pancreatic duct diameter (vs ≥5 mm: 4 mm, OR, 0.3 [95% CI, 0.2-0.6]; P < .001; 3 mm, OR, 0.4 [95% CI, 0.2-0.6]; P < .001; 2 mm, OR, 0.5 [95% CI, 0.3-0.9]; P = .02), and internal transanastomotic stent placement (OR, 71.5 [95% CI, 36.0-141.8]; P < .001). Notably, reduced intraoperative blood loss was not associated with undergoing RPD.
Table 1. Comparison of Preoperative and Intraoperative Variables Between Cohorts Undergoing OPD or RPD Before and After Propensity Score-Matching (Model 1).
| Variable | No. (%) | |||||
|---|---|---|---|---|---|---|
| Before Propensity Score-Matching | After Propensity Score-Matching | |||||
| OPD | RPD | P Value | OPD | RPD | P Value | |
| Patients | 2661 (93.5) | 185 (6.5) | 152 (50.0) | 152 (50.0) | ||
| Gland texture | <.001 | >.99 | ||||
| Firm/hard | 1305 (49.0) | 61 (33.0) | 58 (38.2) | 58 (38.2) | ||
| Soft | 1356 (51.0) | 124 (67.0) | 94 (61.8) | 94 (61.8) | ||
| Pathologic findings of high-risk disease? | .27 | >.99 | ||||
| No, PDAC/pancreatitis | 1405 (52.8) | 90 (48.6) | 81 (53.3) | 81 (53.3) | ||
| Yes, other pathologic findings | 1256 (47.2) | 95 (51.4) | 71 (46.7) | 71 (46.7) | ||
| Pancreatic duct diameter, mm | <.001 | >.99 | ||||
| ≥5 | 730 (27.4) | 67 (36.2) | 57 (37.5) | 57 (37.5) | ||
| 4 | 527 (19.8) | 20 (10.8) | 18 (11.8) | 18 (11.8) | ||
| 3 | 739 (27.8) | 41 (22.2) | 37 (24.3) | 37 (24.3) | ||
| 2 | 539 (20.3) | 38 (20.5) | 32 (21.1) | 32 (21.1) | ||
| ≤1 | 126 (4.7) | 19 (10.3) | 8 (5.3) | 8 (5.3) | ||
| Intraoperative blood loss, mL | <.001 | >.99 | ||||
| ≤400 | 1431 (53.8) | 147 (79.5) | 123 (80.9) | 123 (80.9) | ||
| 401-700 | 608 (22.8) | 21 (11.4) | 16 (10.5) | 16 (10.5) | ||
| 701-1000 | 287 (10.8) | 9 (4.9) | 7 (4.6) | 7 (4.6) | ||
| >1000 | 335 (12.6) | 8 (4.3) | 6 (3.9) | 6 (3.9) | ||
| Operative factors | ||||||
| Type of reconstruction | .04 | >.99 | ||||
| Pancreaticojejunostomy | 2569 (96.5) | 185 (100) | 152 (100) | 152 (100) | ||
| Pancreaticogastrostomy | 91 (3.4) | 0 | 0 | 0 | ||
| None, occlusion | 1 (0.04) | 0 | 0 | 0 | ||
| Intraoperative drain placement? | <.001 | |||||
| No | 205 (7.7) | 0 | 0 | 0 | ||
| Yes | 2456 (92.3) | 185 (100) | 152 (100) | 152 (100) | ||
| Transanastomotic stent placement? | <.001 | >.99 | ||||
| No | 1500 (56.4) | 9 (4.9) | 9 (5.9) | 9 (5.9) | ||
| Internal | 548 (20.6) | 176 (95.1) | 143 (94.1) | 143 (94.1) | ||
| External | 613 (23.0) | 0 | 0 | 0 | ||
| Prophylactic octreotide? | .001 | >.99 | ||||
| No | 1826 (68.6) | 105 (56.8) | 100 (65.8) | 100 (65.8) | ||
| Yes | 835 (31.4) | 80 (43.2) | 52 (34.2) | 52 (34.2) | ||
Abbreviations: OPD, open pancreatoduodenectomy; PDAC, pancreatic ductal adenocarcinoma; RPD, robotic pancreatoduodenectomy.
Predictors of CR-POPF Occurrence
The impact of undergoing RPD on CR-POPF development was examined in the unmatched cohort (2846) using logistic regression modeling. After adjusting for potential confounders, undergoing RPD was associated with a reduced risk for CR-POPF incidence (OR, 0.4 [95% CI, 0.2-0.7]; P = .002) relative to OPD (Table 2). Other predictors of risk-adjusted CR-POPF occurrence included soft pancreatic parenchyma (OR, 4.7 [95% CI, 3.4-6.6]; P < .001), pathologic findings of high-risk disease (OR, 1.4 [95% CI, 1.1-1.9]; P = .01), small pancreatic duct diameter (vs ≥5 mm: 2 mm, OR, 2.1 [95% CI, 1.4-3.1]; P < .001; ≤1 mm, OR, 1.8 [95% CI, 1.0-3.0]; P = .03), elevated intraoperative blood loss (vs ≤400 mL: 401-700 mL, OR, 1.5 [95% CI, 1.1-2.0]; P = .01; >1000 mL, OR, 2.1 [95% CI, 1.4-2.9]; P < .001), omission of intraoperative drain(s) (OR, 0.5 [95% CI, 0.3-0.8]; P = .005), and octreotide prophylaxis (OR, 3.1 [95% CI, 2.3-4.0]; P < .001).
Table 2. Logistic Regression Model for Clinically Relevant Pancreatic Fistula (CR-POPF) Occurrence in 2846 Unmatched Patients Undergoing OPD or RPD.
| Variable | CR-POPF, No. (%) | Odds Ratio (95% CI) | P Value | |
|---|---|---|---|---|
| No | Yes | |||
| Operative approach | ||||
| Open | 2310 (86.8) | 351 (13.2) | 1 [Reference] | .002 |
| Robotic | 169 (91.4) | 16 (8.6) | 0.4 (0.2-0.7) | |
| Gland texture | ||||
| Firm/hard | 1309 (95.8) | 57 (4.2) | 1 [Reference] | <.001 |
| Soft | 1170 (79.1) | 310 (20.9) | 4.7 (3.4-6.6) | |
| Pathologic findings of high-risk disease ? | ||||
| No, PDAC/pancreatitis | 1378 (92.2) | 117 (7.8) | 1 [Reference] | .01 |
| Yes, Other pathologic findings | 1101 (81.5) | 250 (18.5) | 1.4 (1.1-1.9) | |
| Pancreatic duct diameter, mm | ||||
| ≥5 | 739 (92.7) | 58 (7.3) | 1 [Reference] | |
| 4 | 488 (89.2) | 59 (10.8) | 1.4 (0.9-2.1) | .11 |
| 3 | 682 (87.4) | 98 (12.6) | 1.4 (0.9-2.1) | .06 |
| 2 | 454 (78.7) | 123 (21.3) | 2.1 (1.4-3.1) | <.001 |
| ≤1 | 116 (80.0) | 29 (20.0) | 1.8 (1.0-3.0) | .03 |
| Intraoperative blood loss, mL | ||||
| ≤400 | 1403 (88.9) | 175 (11.1) | 1 [Reference] | |
| 401-700 | 544 (86.5) | 85 (13.5) | 1.5 (1.1-2.0) | .01 |
| 701-1000 | 257 (86.8) | 39 (13.2) | 1.4 (0.9-2.1) | .14 |
| >1000 | 275 (80.2) | 68 (19.8) | 2.1 (1.4-2.9) | <.001 |
| Operative factors | ||||
| Type of reconstruction | ||||
| Pancreaticojejunostomy | 2406 (87.4) | 348 (12.6) | 1 [Reference] | |
| Pancreaticogastrostomy | 73 (80.2) | 18 (19.8) | .07 | |
| None, occlusion | 0 | 1 (100) | .19 | |
| Intraoperative drain placement? | ||||
| No | 163 (79.5) | 42 (20.5) | 1 [Reference] | |
| Yes | 2316 (87.7) | 325 (12.3) | 0.5 (0.3-0.8) | .005 |
| Transanastomotic stent placement? | ||||
| No | 1294 (85.8) | 215 (14.2) | 1 [Reference] | |
| Internal | 623 (86.0) | 101 (14.0) | 1.2 (0.9-1.7) | .27 |
| External | 562 (91.7) | 51 (8.3) | 0.7 (0.5-1.0) | .06 |
| Prophylactic octreotide | ||||
| No | 1790 (92.7) | 141 (7.3) | 1 [Reference] | |
| Yes | 689 (75.3) | 226 (24.7) | 3.1 (2.3-4.0) | <.001 |
Abbreviations: OPD, open pancreatoduodenectomy; PDAC, pancreatic ductal adenocarcinoma; RPD, robotic pancreatoduodenectomy.
Characteristics of Propensity Score–Matched Cohorts
To better control for confounding in the selection of undergoing RPD vs OPD, patients were matched 1:1 based on the likelihood of either undergoing RPD or factors associated with CR-POPF occurrence in the unmatched cohort (Table 2). The propensity score–matched cohort from the primary analysis (model 1) comprised 304 patients—152 in the RPD group (50.0%) and 152 in the OPD group (50.0%). Covariate differences between cohorts were compared before and after matching (Table 1). Previously observed covariate imbalances between RPD and OPD cohorts with respect to pancreatic gland texture, pancreatic duct diameter, intraoperative blood loss, type of reconstruction, intraoperative drain placement, transanastomotic stent placement, and administration of octreotide prophylaxis were alleviated after matching. As such, matching was effective in reducing the absolute standardized difference to less than 10% for all covariates. In fact, exact matching was achieved for every patient (Table 1); therefore, RPD and OPD cohorts had identical mean (SDs) (3.5 [2.3] vs 3.5 [2.3]; P > .99) and median (4 [IQR, 1-5] vs 4 [IQR, 1-5]; P > .99) FRSs.
Effect of Undergoing RPD on CR-POPF Occurrence in Propensity Score–Matched Cohorts (Model 1)
In the overall propensity score–matched cohort (model 1), 27 patients (8.9%) developed a CR-POPF (grade B, 7.9%; grade C, 1.0%). Comparisons of the RPD and OPD matched cohorts revealed no significant differences in CR-POPF occurrence (6.6% vs 11.2%; P = .23). Further comparisons, stratified by FRS risk zone, confirmed noninferiority of RPD to OPD in terms of CR-POPF incidence (Table 3). In addition, no differences were observed between matched cohorts in terms of grade B (RPD vs OPD: 6.6% vs 9.2%; P = .52) or grade C (RPD vs OPD: 0 vs 2.0%; P = .25) POPF development.
Table 3. Impact of Undergoing RPD vs OPD for Clinically Relevant Pancreatic Fistula (CR-POPF) Occurrence and Other Postoperative Outcomes in Propensity Score-Matched Cohorts (Model 1)a.
| Variable, No. (%) | No. (%) | P Value | |
|---|---|---|---|
| OPD | RPD | ||
| Patients | 152 (50.0) | 152 (50.0) | |
| CR-POPF | 17 (11.2) | 10 (6.6) | .23 |
| FRS risk zone | |||
| Negligible, FRS 0 | 0 | 1 (4.3) | >.99 |
| Low, FRS 0-2 | 2 (6.7) | 0 | .50 |
| Moderate, FRS 3-6 | 12 (14.1) | 5 (5.9) | .14 |
| High, FRS 7-10 | 3 (21.4) | 4 (28.6) | >.99 |
| Grade B POPF | 14 (9.2) | 10 (6.6) | .52 |
| Grade C POPF | 3 (2.0) | 0 | .25 |
| Complications | |||
| Any, Accordion severity grading system grade ≥1 | 101 (66.4) | 112 (73.7) | .21 |
| Mild/moderate, Accordion score 1-2 | 65 (42.8) | 77 (50.7) | .24 |
| Severe, Accordion severity grading system grade ≥3 | 36 (23.7) | 35 (23.0) | >.99 |
| Readmission, 30-d | 33 (21.7) | 34 (22.4) | >.99 |
| Duration of hospital stay, d | |||
| Mean (SD) | 11.8 (10.6) | 10.5 (6.9) | .22 |
| Median, IQR | 8.5 (7-12) | 8 (7-12) | .31 |
| Mortality, 90-d | 2 (1.3) | 5 (3.3) | .38 |
Abbreviations: FRS, Fistula Risk Score; IQR, interquartile range; OPD, open pancreatoduodenectomy; PDAC, pancreatic ductal adenocarcinoma; POPF, postoperative pancreatic fistula; RPD, robotic pancreatoduodenectomy.
CR-POPF outcome comparisons were also stratified across FRS risk zones.
Effect of Undergoing RPD on Other Outcomes in Propensity Score–Matched Cohorts (Model 1)
Other postoperative outcomes were compared between the RPD and OPD matched cohorts. The RPD cohort had noninferior outcomes compared with the OPD cohort in terms of the occurrence of any complication (Accordion grade ≥1: 73.7% vs 66.4%; P = .21), mild to moderate complications (Accordion grade 1-2: 50.7% vs 42.8%; P = .24), severe complications (Accordion grade ≥3: 23.0% vs 23.7%; P > .99), readmission (22.4% vs 21.7%; P > .99), and mortality (3.3% vs 1.3%; P = .38) (Table 3). In addition, median duration of hospital stay was similar between RPD and OPD patients (8 days [IQR, 7-12 days] vs 8.5 days [IQR, 7-12 days]; P = .31).
Effect of Undergoing RPD on Outcomes in Propensity Score–Matched Cohorts (Model 2)
Outcomes were also compared between RPD (n = 171) and OPD (n = 171) matched cohorts from the secondary propensity score–matched analysis (model 2). Similar to results from model 1, there were no significant differences between RPD and OPD cohorts in terms of the development of CR-POPF (7.0% vs 12.3%; P = .10), grade B POPF (7.0% vs 9.9%; P = .33), or grade C POPF (0% vs 2.3%; P = .12). In addition, the RPD cohort had noninferior outcomes compared with the OPD cohort in terms of the occurrence of any complication (Accordion grade ≥1: 74.3% vs 68.4%; P = .23), mild to moderate complications (Accordion grade 1-2: 49.7% vs 43.3%; P = .23), severe complications (Accordion grade ≥3: 24.6% vs 25.1%; P = .90), readmission (21.6% vs 17.9%; P = .38), and mortality (2.9% vs 3.5%; P = .76). Median duration of hospital stay was also similar between RPD and OPD patients (8 days [IQR, 7-12 days] vs 8 days [IQR, 8-12 days]; P = .17).
Discussion
This study demonstrates that RPD is noninferior to OPD for CR-POPF occurrence. To control more comprehensively for biases associated with the selection of a particular operative approach for PD, we stringently matched RPD and OPD patients using a propensity score–matching method. In matched patients, noninferiority of RPD was observed for CR-POPF occurrence. Secondary analyses also identified a similar incidence of any complication, mild to moderate complications, severe complications, readmission, mortality, and duration of hospital stay between RPD and OPD.
The importance of a focused analysis comparing CR-POPF outcomes between RPD and OPD is manifold. First, CR-POPFs are the greatest contributor to major morbidity and mortality following PD; therefore, this complication is strongly associated with the overall incidence of severe complications, duration of hospital stay, readmission, and mortality. Second, the risk factors for this complication have been well defined, which enabled the present study to match RPD and OPD patients on characteristics relevant to the outcome of interest—thus eliminating bias.
The first study comparing CR-POPF outcomes between RPD and OPD showed equivalency between operative approaches. Despite demonstrating similarities in CR-POPF outcomes and mortality, Zhou and colleagues reported lower rates of overall complications and reduced hospital stay with RPD. Although the present study also showed similar rates of CR-POPF between cohorts following propensity score matching, the rates of overall complications and hospital stay were, instead, noninferior between RPD and OPD. Potential explanations for these discrepancies include the study by Zhou et al being too underpowered to make an impartial assessment—only 8 patients were included for each operative approach—and the generalizability of their reported hospital stay values are questionable. In the study by Zhou et al, the mean duration of stay for RPD was 16.4 days vs 24.3 days for OPD; conversely, in the present study RPD and OPD patients had mean hospital stays of 10.5 days and 11.8 days, respectively.
Another single-center, retrospective study comparing RPD and OPD was conducted by Buchs et al. Robotic pancreatoduodenectomy demonstrated equivalent outcomes compared with OPD in terms of CR-POPF, overall complications, postoperative mortality, and duration of hospital stay. Although that study reported the largest number of RPDs for a comparative analysis of RPD and OPD to date, it was still limited to just 44 RPDs and 39 OPDs. Two additional retrospective studies have compared these operative approaches. Their findings mirrored those reported by Zhou et al: equivalency in terms of CR-POPF occurrence, overall complications, and mortality; however, RPD was associated with reduced hospital stay. Both studies were underpowered and influenced by selection bias. Lai and colleagues performed RPDs only for patients with an American Society of Anesthesiologists score of 3 or lower and no vascular invasion, while Chalikonda et al performed RPDs on a greater proportion of patients with pancreatitis. Conversely, in the present study, RPD and OPD patients were exactly matched based on characteristics associated with CR-POPF development and undergoing RPD. This eliminated treatment selection bias and replicated some of the characteristics of a randomized clinical trial.
Although the current study focused on comparing robotic and open approaches, laparoscopy is another minimally invasive technique for PD. The efficacy of laparoscopic PD (LPD) vs OPD has been investigated in numerous studies and meta-analyses. These studies were unable to detect significant differences in the frequency of pancreatic fistula formation between operative approaches. The equivalency in outcomes between LPD and OPD observed in previous studies, combined with the findings of the present study, suggest that minimally invasive approaches are feasible for PD and produce noninferior outcomes when compared with open approaches.
Limitations
To our knowledge, this propensity score–matched analysis of RPD and OPD represents the largest study on the subject to date, but it has several limitations that warrant emphasis. First, the institutions included in this analysis are all high-volume, academic, pancreatic surgery specialty centers; consequently, our findings may not be generalizable to lower-volume, nonacademic centers. Second, this study is retrospective, and, although many cases were consecutive, this did not apply across all institutions because all components of the FRS were required for inclusion in the analysis. Third, all RPDs were performed by a highly skilled and experienced group of surgeons who had surpassed the RPD learning curve before contributing to the current study; therefore, these findings may not be generalizable to less experienced surgeons who have not surpassed the learning curve for RPD. Fourth, despite adjusting for baseline differences between RPD and OPD patients using a rigorous propensity score–matching approach, it is possible that unmeasured confounders were not accounted for, resulting in residual treatment selection bias. Finally, it was determined that each cohort would require 158 patients (316 total) to demonstrate noninferiority of RPD to OPD (power = 80%, α = .05), but only 152 patients were matched for each operative technique (304 total) in model 1. Therefore, the initial analysis was slightly underpowered; however, the secondary propensity score–matched analysis (model 2)—which matched patients only on factors associated with use of RPD—was sufficiently powered, with 171 patients in each cohort (342 total). Future prospective randomized clinical trials will be necessary to determine whether one approach is superior to the other; however, an adequately powered superiority trial would require a substantial number of patients. For example, to demonstrate a one-third decrease in CR-POPF incidence would require 1368 patients (684 per cohort).
With comparable perioperative outcomes to OPD, this study provides an impetus to continue exploring the potential benefits—or lack thereof—of RPD. In the current climate of value driven health care, the cost of using the robotic platform for complex procedures, such as PD, needs to be carefully examined. While this study focused on traditional perioperative outcome metrics, studies examining the impact of this platform on indirect costs including quality of life, receipt and completion of full-dose multidrug adjuvant chemotherapy—rather than time to adjuvant therapy—and survival are urgently needed. As such, this study adds to a growing body of evidence that dispels reports of inferior perioperative outcomes for minimally invasive PD, provided these procedures are performed by high-volume pancreatic surgeons in a systematic and structured setting, and lays the foundation for longitudinal patient-centered assessments of various approaches to PD.
Conclusions
To our knowledge, this study is the largest of its kind to compare the efficacy of robotic vs open PD, as well as the first to use propensity score–matching methodology. The findings herein demonstrate that robotic PD is noninferior to open PD in terms of pancreatic fistula development and other major postoperative outcomes.
References
- 1.Clinical Outcomes of Surgical Therapy Study Group A comparison of laparoscopically assisted and open colectomy for colon cancer. N Engl J Med. 2004;350(20):2050-2059. [DOI] [PubMed] [Google Scholar]
- 2.Huscher CG, Mingoli A, Sgarzini G, et al. . Laparoscopic versus open subtotal gastrectomy for distal gastric cancer: five-year results of a randomized prospective trial. Ann Surg. 2005;241(2):232-237. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Fleshman J, Sargent DJ, Green E, et al. ; Clinical Outcomes of Surgical Therapy Study Group . Laparoscopic colectomy for cancer is not inferior to open surgery based on 5-year data from the COST Study Group trial. Ann Surg. 2007;246(4):655-662. [DOI] [PubMed] [Google Scholar]
- 4.Buunen M, Veldkamp R, Hop WC, et al. ; Colon Cancer Laparoscopic or Open Resection Study Group . Survival after laparoscopic surgery versus open surgery for colon cancer: long-term outcome of a randomised clinical trial. Lancet Oncol. 2009;10(1):44-52. [DOI] [PubMed] [Google Scholar]
- 5.Kim HH, Hyung WJ, Cho GS, et al. . Morbidity and mortality of laparoscopic gastrectomy versus open gastrectomy for gastric cancer: an interim report: a phase III multicenter, prospective, randomized Trial (KLASS Trial). Ann Surg. 2010;251(3):417-420. [DOI] [PubMed] [Google Scholar]
- 6.Biere SS, van Berge Henegouwen MI, Maas KW, et al. . Minimally invasive versus open oesophagectomy for patients with oesophageal cancer: a multicentre, open-label, randomised controlled trial. Lancet. 2012;379(9829):1887-1892. [DOI] [PubMed] [Google Scholar]
- 7.Correa-Gallego C, Dinkelspiel HE, Sulimanoff I, et al. . Minimally-invasive vs open pancreaticoduodenectomy: systematic review and meta-analysis. J Am Coll Surg. 2014;218(1):129-139. [DOI] [PubMed] [Google Scholar]
- 8.Boggi U, Signori S, De Lio N, et al. . Feasibility of robotic pancreaticoduodenectomy. Br J Surg. 2013;100(7):917-925. [DOI] [PubMed] [Google Scholar]
- 9.McMillan MT, Vollmer CM Jr. Predictive factors for pancreatic fistula following pancreatectomy. Langenbecks Arch Surg. 2014;399(7):811-824. [DOI] [PubMed] [Google Scholar]
- 10.Vollmer CM Jr, Lewis RS, Hall BL, et al. . Establishing a quantitative benchmark for morbidity in pancreatoduodenectomy using ACS-NSQIP, the Accordion Severity Grading System, and the Postoperative Morbidity Index. Ann Surg. 2015;261(3):527-536. [DOI] [PubMed] [Google Scholar]
- 11.McMillan MT, Soi S, Asbun HJ, et al. . Risk-adjusted outcomes of clinically relevant pancreatic fistula following pancreatoduodenectomy: a model for performance evaluation. Ann Surg. 2016;264(2):344-352. [DOI] [PubMed] [Google Scholar]
- 12.McMillan MT, Christein JD, Callery MP, et al. . Comparing the burden of pancreatic fistulas after pancreatoduodenectomy and distal pancreatectomy. Surgery. 2016;159(4):1013-1022. [DOI] [PubMed] [Google Scholar]
- 13.Bassi C, Dervenis C, Butturini G, et al. ; International Study Group on Pancreatic Fistula Definition . Postoperative pancreatic fistula: an international study group (ISGPF) definition. Surgery. 2005;138(1):8-13. [DOI] [PubMed] [Google Scholar]
- 14.Pratt WB, Maithel SK, Vanounou T, Huang ZS, Callery MP, Vollmer CM Jr. Clinical and economic validation of the International Study Group of Pancreatic Fistula (ISGPF) classification scheme. Ann Surg. 2007;245(3):443-451. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.McMillan MT, Vollmer CM Jr, Asbun HJ, et al. . The characterization and prediction of ISGPF grade C fistulas following pancreatoduodenectomy. J Gastrointest Surg. 2016;20(2):262-276. [DOI] [PubMed] [Google Scholar]
- 16.Buchs NC, Addeo P, Bianco FM, Ayloo S, Benedetti E, Giulianotti PC. Robotic versus open pancreaticoduodenectomy: a comparative study at a single institution. World J Surg. 2011;35(12):2739-2746. [DOI] [PubMed] [Google Scholar]
- 17.Zhou NX, Chen JZ, Liu Q, et al. . Outcomes of pancreatoduodenectomy with robotic surgery versus open surgery. Int J Med Robot. 2011;7(2):131-137. [DOI] [PubMed] [Google Scholar]
- 18.Lai EC, Yang GP, Tang CN. Robot-assisted laparoscopic pancreaticoduodenectomy versus open pancreaticoduodenectomy: a comparative study. Int J Surg. 2012;10(9):475-479. [DOI] [PubMed] [Google Scholar]
- 19.Chalikonda S, Aguilar-Saavedra JR, Walsh RM. Laparoscopic robotic-assisted pancreaticoduodenectomy: a case-matched comparison with open resection. Surg Endosc. 2012;26(9):2397-2402. [DOI] [PubMed] [Google Scholar]
- 20.Austin PC. A critical appraisal of propensity-score matching in the medical literature between 1996 and 2003. Stat Med. 2008;27(12):2037-2049. [DOI] [PubMed] [Google Scholar]
- 21.Tseng JF, Pisters PW, Lee JE, et al. . The learning curve in pancreatic surgery. Surgery. 2007;141(5):694-701. [DOI] [PubMed] [Google Scholar]
- 22.Boone BA, Zenati M, Hogg ME, et al. . Assessment of quality outcomes for robotic pancreaticoduodenectomy: identification of the learning curve. JAMA Surg. 2015;150(5):416-422. [DOI] [PubMed] [Google Scholar]
- 23.Callery MP, Pratt WB, Kent TS, Chaikof EL, Vollmer CM Jr. A prospectively validated clinical risk score accurately predicts pancreatic fistula after pancreatoduodenectomy. J Am Coll Surg. 2013;216(1):1-14. [DOI] [PubMed] [Google Scholar]
- 24.Pratt WB, Callery MP, Vollmer CM Jr. Risk prediction for development of pancreatic fistula using the ISGPF classification scheme. World J Surg. 2008;32(3):419-428. [DOI] [PubMed] [Google Scholar]
- 25.Blackwelder WC. “Proving the null hypothesis” in clinical trials. Control Clin Trials. 1982;3(4):345-353. [DOI] [PubMed] [Google Scholar]
- 26.Austin PC. An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivariate Behav Res. 2011;46(3):399-424. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Austin PC. Optimal caliper widths for propensity-score matching when estimating differences in means and differences in proportions in observational studies. Pharm Stat. 2011;10(2):150-161. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Porembka MR, Hall BL, Hirbe M, Strasberg SM. Quantitative weighting of postoperative complications based on the accordion severity grading system: demonstration of potential impact using the American College of Surgeons National Surgical Quality Improvement Program. J Am Coll Surg. 2010;210(3):286-298. [DOI] [PubMed] [Google Scholar]
- 29.Thoemmes F. Propensity score matching in SPSS. University of Tubingen, 2012. http://arxiv.org/ftp/arxiv/papers/1201/1201.6385.pdf. Accessed June 2, 2016.
- 30.Kent TS, Sachs TE, Callery MP, Vollmer CM Jr. Readmission after major pancreatic resection: a necessary evil? J Am Coll Surg. 2011;213(4):515-523. [DOI] [PubMed] [Google Scholar]
- 31.Gumbs AA, Grès P, Madureira FA, Gayet B. Laparoscopic vs. open resection of noninvasive intraductal pancreatic mucinous neoplasms. J Gastrointest Surg. 2008;12(4):707-712. [DOI] [PubMed] [Google Scholar]
- 32.Pugliese R, Scandroglio I, Sansonna F, et al. . Laparoscopic pancreaticoduodenectomy: a retrospective review of 19 cases. Surg Laparosc Endosc Percutan Tech. 2008;18(1):13-18. [DOI] [PubMed] [Google Scholar]
- 33.Cho A, Yamamoto H, Nagata M, et al. . Comparison of laparoscopy-assisted and open pylorus-preserving pancreaticoduodenectomy for periampullary disease. Am J Surg. 2009;198(3):445-449. [DOI] [PubMed] [Google Scholar]
- 34.Zureikat AH, Breaux JA, Steel JL, Hughes SJ. Can laparoscopic pancreaticoduodenectomy be safely implemented? J Gastrointest Surg. 2011;15(7):1151-1157. [DOI] [PubMed] [Google Scholar]
- 35.Asbun HJ, Stauffer JA. Laparoscopic vs open pancreaticoduodenectomy: overall outcomes and severity of complications using the Accordion Severity Grading System. J Am Coll Surg. 2012;215(6):810-819. [DOI] [PubMed] [Google Scholar]
- 36.Kuroki T, Adachi T, Okamoto T, Kanematsu T. A non-randomized comparative study of laparoscopy-assisted pancreaticoduodenectomy and open pancreaticoduodenectomy. Hepatogastroenterology. 2012;59(114):570-573. [DOI] [PubMed] [Google Scholar]
- 37.Mesleh MG, Stauffer JA, Bowers SP, Asbun HJ. Cost analysis of open and laparoscopic pancreaticoduodenectomy: a single institution comparison. Surg Endosc. 2013;27(12):4518-4523. [DOI] [PubMed] [Google Scholar]
- 38.Bao PQ, Mazirka PO, Watkins KT. Retrospective comparison of robot-assisted minimally invasive versus open pancreaticoduodenectomy for periampullary neoplasms. J Gastrointest Surg. 2014;18(4):682-689. [DOI] [PubMed] [Google Scholar]
- 39.Langan RC, Graham JA, Chin AB, et al. . Laparoscopic-assisted versus open pancreaticoduodenectomy: early favorable physical quality-of-life measures. Surgery. 2014;156(2):379-384. [DOI] [PubMed] [Google Scholar]
- 40.Speicher PJ, Nussbaum DP, White RR, et al. . Defining the learning curve for team-based laparoscopic pancreaticoduodenectomy. Ann Surg Oncol. 2014;21(12):4014-4019. [DOI] [PubMed] [Google Scholar]
- 41.Doula C, Kostakis ID, Damaskos C, et al. . Comparison between minimally invasive and open pancreaticoduodenectomy: a systematic review. Surg Laparosc Endosc Percutan Tech. 2016;26(1):6-16. [DOI] [PubMed] [Google Scholar]
- 42.Liao CH, Wu YT, Liu YY, et al. . Systemic review of the feasibility and advantage of minimally invasive pancreaticoduodenectomy. World J Surg. 2016;40(5):1218-1225. [DOI] [PubMed] [Google Scholar]