Risk of Open Conversion During Robotic Gastrectomy for Gastric Cancer: Optimizing Patient Selection.

1
Introduction
Minimally invasive gastrectomy (MIG), by laparoscopic or robotic approach, has become standard‐of‐care for curative‐intent treatment of gastric adenocarcinoma (GC) at many high‐volume centers, particularly in East Asia. Despite higher instrument costs, the need for additional surgeon training and longer operative times, indications for MIG have expanded from early GC (EGC) to advanced GC (AGC)—defined as cT2 or greater with consistently better surgical outcomes, including faster recovery and lower complication rates. Intraoperative loss of safety and appropriate oncologic techniques are concerns that can lead to open conversion (OC) and poorer outcomes.
Laparoscopic gastrectomy (LG) for both EGC and AGC has improved recovery and lowered complication rates [1, 2, 3, 4, 5, 6]. Yet, the surgical procedure for AGC, which requires the recommended D2 lymphadenectomy can be complex and technically demanding, especially in more obese Western patient populations. Even for East Asian GC patients who have a lower average body mass index (BMI), laparoscopic D2 lymphadenectomy has a learning curve exceeding 50–100 cases. Thus, LG adoption is challenging in low‐volume centers or for patients with more advanced tumors or high BMI [7, 8]. In fact, LG for cancer has plateaued at 23%–26% in the past decade [9, 10], while robotic approach, with its easier instrument manipulation, better visualization, improved ergonomics, and a shorter learning curve [11, 12, 13, 14], has outpaced the laparoscopic approach in the United States [15, 16, 17, 18]. Compared to open gastrectomy (OG), robotic gastrectomy (RG) has equivalent postoperative complication rates [19, 20] and is superior to LG in regard to blood loss, length of stay (LOS), and lymph node harvest [21, 22, 23].
As RGs for cancer become more common, concerns rise about risk of OC due to loss of intraoperative safety and/or inability to maintain proper oncologic techniques. OC increases LOS and is associated with worse oncologic margin status [24]. Cancer and operation‐specific factors such as tumor size > 5 cm, high BMI, and more proximal tumors increase likelihood of OC during MIG [11, 24]. Varying MIG to OG conversion rates have been identified. Some Asian centers report rates as low as 0% [4, 25, 26], whereas Western centers report rates as low as 5.2% in small series [27] vs. 27.7% (61 of 220) at a single high‐volume US cancer center [11], with an overall 7.1% OC rate according to the National Cancer Database [28]. Understanding perioperative risk factors for OC may better inform surgical decision making yet, data regarding preoperative risk assessment for OC in Western GC patient population is scarce. In this study, we aimed to provide risk stratification for OC within a diverse GC population undergoing RG by identifying potential predictors based on patient, tumor, and preoperative management data.

2
Methods
2.1
Study Design and Patients
This is a retrospective study of gastric adenocarcinoma patients who underwent curative‐intent radical resection between January 2010 and October 2022 at a single, high‐volume comprehensive US cancer center. Data was retrospectively collected using electronic medical records before June 2015 and prospectively collected and maintained from July 2015. Patients were included based on biopsy‐proven gastric adenocarcinoma, as defined by the AJCC 8th edition. Patients were included if the operation used robotic assistance. Patients were excluded if (1) they underwent palliative gastrectomy (n = 5); (2) had a gastrectomy for an indication other than gastric adenocarcinoma (n = 9); 3) insufficient data (n = 8 robotic cases were logged but had insufficient data/documentation to extract more than a few predictor variables (e.g., age, gender, date of surgery); (4) an additional 43 gastrectomy cases were excluded after review of the operative note found that these were not robotic procedures but performed laparoscopically. The study was approved by the Institutional Review Board (IRB# 19187). This study followed recommended STROBE guidelines for observational studies [29].

2.2
Robotic Gastrectomy Standardization
Surgeon‐based use of robot assistance for lymphadenectomy in GC surgery began in 2003 at our institution. In July 2015, RG was departmentally standardized. Our institution has established a dedicated gastroesophageal cancer program comprised of a multidisciplinary team of specialists with expertise in treating GC. A standardized best‐practice model, based on the National Comprehensive Cancer Network (NCCN) GC guidelines, with recommendations by the Korean Practice Guidelines for GC (KGCA), Japanese Gastric Cancer Guidelines (JGCA), and the European Society for Medical Oncology (ESMO) for GC patient management, and GI American Society of Clinical Oncology (ASCO), has been developed over the past decade for an optimized GC patient management pathway. Preoperatively, GC patients, particularly those with complex presentations are reviewed and discussed amongst team members, including the lead medical, surgical, and radiation oncologists and GI pathologist specializing in GC care. Together, each patient received a personalized approach to management decisions, including a thorough preoperative work‐up, discussion on the timing and type of perioperative systemic therapy regimen, surgical timing, extent, and approach, and goals of patient care. After which surgery was performed. All patients with planned curative‐intent surgery, including subtotal or completion gastrectomy, total gastrectomy, or esophagogastrectomy, were evaluated for RG with appropriate lymphadenectomy extent. GC patients with resectable disease were offered a robotic approach preferentially over laparoscopic or open procedures, if considered physically fit for RG by the surgeon and anesthesia team. Relative contraindications, such as multiorgan involvement and/or extensive nodal disease on preoperative imaging, varied by surgeon. Absolute contraindications included bleeding disorders, high cardiac risk, calcified left gastric artery on imaging, and patient refusal of robotic approach after informed consent.

2.3
Data Collection
Data collected included clinicopathologic characteristics, preoperative workup, staging, medical history, operation type, extent of gastrectomy and lymphadenectomy, estimated blood loss (EBL), anesthesia time, postoperative complications, and survival. All procedures were performed by surgical oncologists with extensive experience in minimally invasive surgery, with addition of a specifically RG‐trained surgeon in July 2015. Patient records were reviewed to identify preoperative workup, denotation of OC concern before surgery within the surgeon's notes, rationale for conversion based on operative reports, and postoperative complication based on Clavien–Dindo classifications with highest grade reported. Bulky nodes were identified either as noted specifically on imaging, endoscopy reports, or intraoperatively. Fibrosis was noted based on operative reports. Difficulty with esophagojejunostomy anastomoses were identified based on operative note descriptions.

2.4
Statistical Analysis

χ
2 and analysis of variance analyses for categorical and continuous variables were used, respectively, to investigate significance of patient demographics and comorbidities between outcome groups. Analysis of predictors was done for RG patients following RG standardization in 2015. Univariate analysis of predictor variable distribution by approach and outcome variables by approach was conducted using the glm function in R studio. Multivariable logistic regression models were constructed to identify perioperative risk factors that significantly increase the probability of conversion. Independent factors significant in univariable analyses were applied in a backward stepwise regression model using the glm function in R studio. The least significant univariate variable was removed in the multivariable regression model using a backward stepwise method using the Akaike Information Criterion for statistical accuracy to attain model significance [30, 31, 32]. Steps were taken to reduce the possibility of overcorrection and the introduction of bias in the model. The model variable elimination was performed following sensitivity analyses, including parsing possible confounding comorbidities (e.g., instead of only age, receipt of neoadjuvant, exact ASA grading, specific medical comorbidities, time between neoadjuvant/adjuvant and surgery, etc.). Moreover, a liberal p‐value of < 0.2 was utilized during elimination to avoid premature predictor discarding [33]. Variables included were only incorporated after being tested for non‐multicollinearity (correlation < 0.8). Results were reported as a risk‐adjusted odds ratio with a 95% confidence interval of the specified variable relative to the specified reference. Analysis was conducted on the entire RG cohort to identify predictors for unplanned conversion from robotic to open and consequences of unplanned conversion. A p‐value of < 0.05 was considered significant. All data set preparation and statistical analyses were performed using R studio Desktop (Version 2023.09.0 + 463). Visualization of multivariable logistic regression results was demonstrated using the forest plot package [34, 35].

3
Results
3.1
Robotic Gastrectomy
Two hundred and eighty‐nine (n = 109 before; n = 180 after July 2015) gastrectomies for GC were performed between January 2010 and October 2022. Five gastrectomies were for palliative intent; 9 were for non‐gastric adenocarcinoma pathologies; and 43 were performed laparoscopically. One hundred and thirty‐three (46.0%) gastrectomies performed for gastric adenocarcinoma were planned as robot‐assisted procedures, and 107 (80.5%) were completed fully robotically, including intracorporeal anastomoses (fully robotic RGs: 11/19 before 2015; 96/114 after 2015). The RG OC rate was 42.1% before RG standardization in 2015, after which the rate was 0% between 2015 and 2019. However, after expansion of RG criteria in 2019 to include patients deemed more technically demanding by surgeons, including more advanced stage and post‐neoadjuvant patients, the OC rate increased to 15.8%. The overall OC rate for the study period was 19.5% (n = 26 of 133) (Table 1).

3.2
Patient Characteristics
Notable characteristic differences were observed between fully RG patients and those whose RG converted to open. Patients were older in the OC group, with an average age of 68 ± 11.4 vs. 60 ± 14.3 years for the fully RG group (p = 0.01) (Table 1). Also, patients in the OC group had a significantly higher incidence of hypertension (HTN) (OC—69.2% vs. RG—36.4%; p = 0.01). However, other comorbidities and functional status were similar (p > 0.05). Smoking status between OC and RG groups was similar (p > 0.05). A significantly higher proportion of OC patients had a history of prior abdominal surgeries (34.6% vs. 10.3%) than the RG group (p < 0.01). However, the proportion of OC versus RG patients who underwent neoadjuvant chemotherapy (61.5% vs. 42.1%) or neoadjuvant radiation (0% vs. 4.7%) did not significantly differ (p > 0.05). Interestingly, significantly more fully RG patients received preoperative endoscopic ultrasound (EUS) evaluation (75.7%, n = 81) compared to OC patients (50.0%, n = 13) (p < 0.01).

3.3
Tumor Characteristics
Tumor characteristics between OC and RG groups were different. There were significantly more patients with AGC in the OC group (n = 23, 95.8%) than in the fully RG group (n = 72, 67.3%) (p = 0.03) (Table 1). Lymphovascular invasion was more frequently noted in OC than RG patient tumors (38.4% vs. 15.0% respectively; p = 0.04). More tumors were in the proximal third of the stomach in the OC group (57.7%, n = 15) than the RG group (24.3%, n = 26) (p = 0.02), congruent with more robotic distal gastrectomies in the fully RG group (53.3%, n = 57) than OC patients (26.9%, n = 7) (p < 0. 01) (Table 1). However, histologic grades, signet ring cell presence (29.2% vs. 24.3%), and perineural invasion (34.6% vs. 15.9%) rates were not significantly different between groups (p > 0.05).

3.4
Intraoperative Events
Intraoperatively, more OC patients experienced complications (19.2%) than RG patients (1.9%) and required intraoperative transfusions (11.5% vs. 1%, respectively; p < 0.05 both) due to blood loss; more OC than fully RG patients had combined resections involving other organs (26.9% vs. 1.9% respectively, p < 0.01). However, there were no differences in rates of D2 lymphadenectomy completion (p > 0.05) or jejunostomy tube placement between two groups. Moreover, number of nodes retrieved was similar, as was the radicality (> 95% R0) of the operation (p > 0.05). Anesthesia time did not significantly differ between the OC (416 min) and RG (388 min) groups (Table 1).

3.5
Open Conversion Rationale
Three OC patients were preoperatively documented as high risk for OC based on comorbidities and CT scan imaging demonstrating extensive nodal disease. Analysis of converted cases revealed several reasons for unplanned OC in three categories—patient factors: hemodynamic instability upon insufflation (n = 2, 7.7%); tumor factors: bulky N2 station nodes (n = 7, 26.9%) and tumor invasion and/or fibrosis (n = 10, 38.5%); and procedure factors: difficulty with esophagojejunostomy (EJ) (n = 6, 23.1%) (Table 2).

3.6
Postoperative Course
The average LOS was 7.1 days for all patients. Patients who successfully underwent a fully RG had a significantly shorter LOS than those undergoing OC (6.5 days vs. 10.2 days, p < 0.01, respectively). The overall 30‐day and 90‐day readmission rates were 10% and did not differ between groups (p > 0.05, both). The overall morbidity rate (Clavien–Dindo > 2) was 17.3% [16.8% (n = 18) RG vs. 19.2% (n = 5) OC]. One mortality occurred secondary to postoperative myocardial infarction. There was no difference in proportion of patients who underwent adjuvant chemotherapy, time to adjuvant chemotherapy, or radiation after robotic or OC gastrectomy (p > 0.05). Average recurrence‐free survival interval was 20.7 months (median 14.5 mo, IQR 8.5–19.3 mo) vs. 20.5 months (median 16 mo, IQR 11.5–24.8 mo) for OC vs. RG patients (p > 0.05); average overall survival (OS) was 30 months (median 18 mo, IQR 10–36 mo) for OC vs 55 months (median 58 mo, IQR 33–76.5 mo) for RG (p < 0.01) (Table 1).

3.7
Postoperative Outcomes Associated With Conversion to Open Gastrectomy
Conversion from robotic to OG was associated with poorer short‐term postoperative outcomes across several measures on univariate analysis. Slightly increased odds of a longer LOS (OR 1.02) with conversion (p = 0.01) were identified. The ORs of postoperative morbidity, 30 and 90‐day readmission were 1.10, 1.15, and 1.03, respectively, if there was a conversion from robotic to OG, but were insignificant (p > 0.05). Moreover, conversion did not significantly influence the odds of recurrence (OR 1.14, p > 0.05) for the OC cohort (Table 3), yet this group had significantly shorter OS with an average of 30 months vs. 55 months for the RG group (p < 0.01) (Table 1) despite indeterminate (OR range includes 1, p < 0.01) (Table 3) influence of OC on OS.

3.8
Multivariate Predictors of Conversion From Robotic to Open Gastrectomy
The following significant univariate variables were determined to be relevant after stepwise selection for the multivariable model: HTN, prior abdominal surgeries, preoperative EUS, EGC status, combined resection, tumor size, AJCC stage, lymphovascular invasion, intraoperative transfusion, and tumor location. Age was not included as it was the eliminated least‐significant univariate significant variable.
First, prior abdominal surgeries on multivariate analysis significantly increased the risk of OC (OR 1.31, p = 0.02). Obtaining a preoperative EUS was associated with significantly reduced OC risk by 22.5% (OR 0.78, p < 0.05). Other factors, such as HTN (OR 1.16, p = 0.05), presence of lymphovascular invasion (OR 1.19, p = 0.07), and intraoperative transfusion needs (OR 1.43, p = 0.07) were not significant on multivariate analysis. Notably, relative to being in the proximal third of the stomach, tumors located in the middle or distal third were not an independent predictor for OC (OR 0.92 and 0.91, respectively, p > 0.05) (Figure 1).

4
Discussion
Robotic surgery is being increasingly used for curative‐intent gastrectomy for AGC. We provide novel real‐life practice outcomes of reduced RG OC rates after standardization of RG practice, and present preoperatively identifiable risk factors predictive of RG OC for GC.
Our results show that following standardization, the rate of institutional RG utilization increased from 17.4% to 63.3% even with expanded RG criteria from EGC to AGC. Simultaneously, OC declined from 42.1% pre‐2015 standardization to 0% until 2019. Further, when RG for GC criteria expanded to include more advanced tumors and complex patients, the RG OC rate increased to 15.8%.
We identified potential OC risk‐reducing factors for GC RG patients. Although not an independent predictor, EGC was a significant factor on univariate analysis for an odds reduction of 20%. Similarly, having middle or distal third tumors had lower odds of OC (OR 0.79 and OR 0.82, respectively; p < 0.01); yet, on multivariate analysis, tumor location was not an independent predictor, consistent with another analysis [11].
The sole significant independent predictor that was associated with reduced likelihood of OC was a preoperative EUS. The AJCC recommends EUS as part of preoperative staging and considers it a primary imaging modality for locoregional GC staging for T and N categories [36, 37, 38]. Obtaining an EUS, as recommended by current cancer guidelines, provides several advantages. Most importantly, it allows for distinction between early (cT1a/b) versus more advanced T‐stages, guiding whether a patient undergoes a D2 lymphadenectomy and/or receives neoadjuvant treatment. Moreover, EUS identifies nodal disease, including location and biopsy [39]. Thus, while EUS itself does not technically prevent OC, as there is no therapeutic intervention or change in the patient clinco‐demographics or tumor nature, it could reduce OC risk due to refined RG candidate selection. EUS perhaps adds value as a second imaging modality to enhance understanding of tumor size, more granular lymph node evaluation, and relationship of the malignancy and associated nodes in relation to adjacent structures. This is supported by reported operative findings, where the surgeon's decision to convert to open from RG primarily occurred due to factors limiting procedure progress, including tumor invasion, adhesions/fibrosis, bulky nodes prohibiting safe identification of vascular structures, or esophagojejunal anastomotic difficulties. Thus, improving guideline‐concordant preoperative EUS may be a modifiable aid in decreasing RG OC rates.
Several factors were associated with but not independent predictors of OC, such as age, COPD (OR 2.21) and HTN (OR 1.21) (Tables S1 and Table 1). These discrepancies are not unexpected as age and comorbidities like HTN, diabetes, and the development of lung disease are linked [40, 41]. Thus, these factors could be confounded by correlation. However, COPD and HTN have in general been linked to conversion overall for robotic procedures [42, 43, 44] These comorbidities may have physiologic roles in conversion reasons noted in our study, such as intolerance to insufflation, for example, with lack of respiratory compliance and hemodynamic lability, respectively, for COPD and HTN. While HTN was not a significant factor for OC as seen by multivariate analysis, the overall incidence of hemodynamic instability as a driver for OC was low. Thus, it is inscrutable as to whether HTN may be a significant predictor even in a larger study, if its effect was impacted by a relatively lower sample size in this study.
Prior abdominal surgery was a strong independent predictor of increased OC risk as observed by multivariate regression. OC odds increased by 31.1% with a history of prior abdominal surgery. This could be due to operative challenges with RG related to adhesions or difficulty with safely mobilizing for resection. These findings are consistent with studies for minimally invasive abdominal surgeries in other procedures that identified prior abdominal surgery as a relative contraindications. Preoperatively assessing patients with prior abdominal surgery as high risk for OC may improve RG selection, patient and surgeon counseling for operative expectations, and postoperative outcomes. Additionally, although not an independent predictor on multivariate analysis, univariate analysis showed that AGC stage was significantly associated with increased OC risk. Preoperative modification based on these findings may include more emphasis on downstaging cancers, despite no independent predictive value for neoadjuvant therapy before surgery in a Western cohort.
Importantly, our analysis supports existing literature in that conversion negatively impacts OS, with significantly longer survival of 55 months in the RG cohort vs. 30 months in the OC cohort. OC was associated with increased hospitalization, consistent with robotic surgery conversion trends in the United States [45, 46]. Conversion did not impede R0 resection, D2 lymphadenectomy, nodes harvested, nodal positivity, margins, or even time to adjuvant therapy in the setting of similar high‐risk histologic features. Despite poorer postoperative outcomes, our data suggest that conversion itself did not compromise oncologic operation quality. Further larger studies are needed to comprehend details for poorer outcomes, as overall patient clinic‐demographics and oncologic outcomes did not seem to directly contribute to this survival disparity in our data.
There are certain limitations to our study. Firstly, the RG cohort and primary operating RG surgeon evolved over time, along with the integration of procedure standardization and establishment of post‐gastrectomy ERAS protocol [47]. The overall conversion rate factored in all surgeons at the institution performing RG for GC in this study, regardless of overall gastrectomy volume. Conversion rates widely varied among operating surgeons, with the highest volume surgeon also having the highest number of AGC cases (Table 4). Secondly, our study time frame incorporated the COVID‐19 pandemic. There were no conversions in RG between mid‐2015 and mid‐2019, with all conversions occurring from late 2019 onwards. These conversions could not be attributed to increasing use of neoadjuvant, as neoadjuvant use was 54.2% (26 of 48) for RG cases after 2019 vs. a neoadjuvant use rate of 44.7% pre‐2019 (38 of 85) (p = ns). Hence, the RG OC rates were likely not representative of typical RG currently, with some degree of variability secondary to the mentioned circumstances. Additionally, this was a retrospective study. Although patient data was prospectively collected, missing data, selection, and interpretation bias cannot be fully excluded as possibilities.
Our findings contribute in several ways to understanding patient selection for RG in AGC and identify preoperative, mitigating factors that influence OC. Outcomes for RG in GC are superior to those from LG and have improved at high‐volume centers, as surgical experience in treating GC and familiarity with RG are important [23, 48, 49]. We compared demographics and tumor characteristics of OC in RG for GC to show how to (1) select appropriate RG patients, (2) highlight surgical experience to anticipate unpredictable intraoperative difficulties, and (3) delineate intervenable preoperative patient and work‐up factors to lower risk and poorer outcomes associated with conversion. Our data adds to evidence that RG is an adequate treatment approach for AGC, as we included a large proportion of patients receiving neoadjuvant treatment—a population previously excluded in clinical studies analyzing RG safety and efficacy [4, 22]. Some studies suggest that neoadjuvant treatment increases the likelihood of OC, while others posit that robotic distal gastrectomy is advantageous after neoadjuvant chemotherapy for AGC patients [50, 51]. Thus, our finding that neoadjuvant treatment is not a significant or independent predictor of RG OC for AGC adds to the literature.
Single‐center LG OC rates are as low as 0.63% to 6.3%, with conversion primarily due to bulky tumors, adhesions, reconstruction, and bleeding difficulties intraoperatively, reasons similar to our RG OC conversion experience [52, 53]. However, majority of low‐conversion rate LG studies are from Eastern centers, with Western centers reporting laparoscopic conversion rates of around 14.5% [54]. Comparatively, RG conversion rates in smaller single‐center studies are as low as 0% [25], and in high‐volume centers as high as 27.7% when including learning curve time [11], but with an overall pooled conversion rate of 3.9% to 7.8% in Western centers when combining EGC and AGC patients [24, 55]. Our single‐institution high‐volume center study offers realistic insight into OC rates in the setting of RG implementation and standardization, alongside evolving AGC inclusion criteria in the last several years.

5
Conclusion
For the first time, we identified preoperatively identifiable predictors of increased OC risk in RG, including advanced age, HTN, prior abdominal surgeries, and advanced staging. We also found that a preoperative EUS can be a risk‐reduction procedure. Improved preoperative planning combined with surgical experience in anticipation of difficult tumor dissection and reconstruction may improve patient selection for RG and OC rates.

Author Contributions
Conceptualization: Yanghee Woo, Yuman Fong, and Courtney Chen. Methodology: Courtney Chen, Tiffany Liu, Annie Yang, Ian Lau, Ryuhei Aoyama, Michael Sullivan, and Aaron Lewis. Formal analysis: Courtney Chen and Tiffany Liu. Investigation: Yanghee Woo, Courtney Chen, Annie Yang, Ian Lau, Kelly Mahuron, Ryuhei Aoyama, Michael Sullivan, Bradford Kim, Aaron Lewis, and Laleh Melstrom. Resources: Yanghee Woo, Yuman Fong, I Benjamin Paz, and Laleh Melstrom. Writing – Original Draft: Yanghee Woo and Courtney Chen. Writing: review and editing: Courtney Chen, Yanghee Woo, Tiffany Liu, Annie Yang, Ian Lau, Kelly Mahuron, Michael Sullivan; Ryuhei Aoyama, Bradford Kim, Aaron Lewis, Laleh Melstrom, I. Benjamin Paz, and Yuman Fong. Visualization: Yanghee Woo. Supervision: Yanghee Woo. Funding acquisition: Yanghee Woo and Yuman Fong.

Ethics Statement
The study was approved by the Institutional Review Board (IRB# 19187).

Conflicts of Interest
Yanghee Woo—Scientific advisor—Imugene; Paid consultant—J&J Ethicon, Virtual Incisions, AstraZeneca; Yuman Fong—Scientific consultant—Medtronic, Johnson & Johnson, Imugene; Royalties—Merck, Imugene. All other authors declare no conflicts of interest.

Synopsis
Hypertension, prior abdominal surgeries, and advanced staging increase the risk of open conversion during robotic gastrectomy (RG). Preoperative endoscopic ultrasound reduces risk. Surgical experience, procedure, preoperative patient conditions, and workup thoroughness can help optimize patients.

Supporting information
Supplemental Table 1.