The total operative time/learning curve analysis in this study demonstrates improvement in time over the study period (Fig. 1). The median operative time in the learning phase of 340 min was higher than the 298.5 min reported in the ROLARR trial [18] and 339.2 min reported by Kim et al. [17]. The median time in our proficiency phase was much lower than both at 260 min. In addition, we included co-cases/multi-visceral resection in our data, which can often significantly impact operative time.

A CUSUM analysis was then performed to demonstrate the learning curve. Most CUSUM analyses included three learning phases, including a plateau phase between learning and expert/proficiency phases. There was a short plateau period from case 64–79, after which the curve drops quickly. Given the small number of cases in this potential plateau phase, these cases were considered part of the learning phase rather than a separate phase. We consider the sharp drop after case 79 as the cutoff between the learning and proficiency phases. The number of cases required to get to proficiency/expert level in the available literature ranges widely, from 10’s to > 100 for some surgeons [4,5,6,7,8,9,10,11,12,13].

Many factors can contribute to the overall operative time including tumor size, location, T-stage, multi-organ resection, experience of the operating room team, and resident involvement. We elected to utilize total operative time as it truly reflects all components of the overall resection such as specimen extraction, anastomotic technique, and endoscopic evaluation of the anastomosis, but also includes the docking, which is a crucial part of the learning experience for robotic surgery. Additionally, most of our cases involved a resident on a teaching console which certainly can influence the operative time.

Patients in the two groups received a variety of neoadjuvant therapy regimens. In total, 79.7% (learning) and 88.0% (proficiency) received some form of neoadjuvant therapy, with a p < 0.0001. 65.8% of those in the learning phase received chemoradiation, compared to 33.3% in the proficiency group (p < 0.05). Total neoadjuvant therapy (TNT), or chemotherapy plus chemoradiation, became the standard treatment over time, and 47.1% of patients in the proficiency group had this therapy vs only 8.9% in the learning phase (p < 0.0001). The standard neoadjuvant therapy drastically changed somewhere around the end of the learning phase and beginning of the proficiency phase. The notable difference reflects the ever-evolving management of cancer in general, but especially the rapid advances in management of rectal cancer.

As far as type of operation performed, there was no significant difference between the two groups. APRs were performed in patients in whom a ~ 1 cm distal margin could not be obtained and low Hartmann’s procedures were performed selectively in patients with low tumors and poor sphincter function who we believed would have better functional outcomes with an end colostomy rather than an ultralow anastomosis. There was no difference in the rate of stoma creation between the two groups, the majority in both being loop ileostomies. The colostomies were all end colostomies that were created in patients who underwent APR or low Hartmann’s, with the intention that they would all be permanent. DLI rate in the learning group was comparable to the ROLARR trial, in which 60.2% of patients had a DLI created. The conversion to open rate was 6.3% and 3.9% in the learning and proficiency phases, respectively (p = 0.434). The conversion rate in the learning group was on par with and the proficiency group improved compared with those reported by two meta-analyses evaluating the outcomes with robotic rectal resections, which were 5.72% [2] and 7.3% [1].

Oncologic outcomes (Table 4) were not significantly different between the two phases. Three patients in the learning phase did not have CRM reported on pathology. Lymphovascular invasion was significantly higher in the learning group on final pathology (p = 0.049); however, 7 of these pathology reports did not comment on lymphovascular invasion. The clinical significance of this difference is uncertain. The difference in TME quality was difficult to assess due to nearly 25% of pathology reports in the learning phase not reporting on this. The pathology reports have since become standardized throughout the institution, and all pathologic data was complete by the proficiency phase. We therefore calculated the rates of complete, near complete and incomplete TME by accordingly using the total number of reports that included TME in the learning group as the denominator. With this, there was an increase in the rate of complete TME to 92.2% vs 79.7% and a decrease in the rate of near complete (5.9% vs 16.9%) and incomplete TME (2.0% vs 3.4%) in the proficiency phase vs learning. Though it cannot be truly assessed for statistical significance, we believe this is clinically significant and is likely attributable to better operative technique as we progressed through the learning curve. Additionally, given the favorable clinical outcome associated with complete and near complete TME, we do not believe oncologic outcomes are compromised during the learning phase. When comparing our outcomes with those of large RCTs analyzing robotic rectal resections, they are all comparable or improved. The rate of complete TME in this analysis was much higher in the proficiency group (92.2%) and comparable in the learning group (79.2%) to the ROLARR (75.4%) and Kim studies (80.3%).

Complete pathologic response, or T0N0 on final pathology, was recorded in 26.6% and 15.7% of the patients in the learning and proficiency groups respectively. The “watch and wait” approach to patients with clinical complete response to neoadjuvant therapy has continued to have promising results for organ preservation [18]. We have adopted a similar approach for highly selected patients which could potentially explain this decrease in pathologic complete response rate. In our patient population, lymph node harvest was very similar to the ROLARR and Kim trials in both of our groups, median (IQR) 18 (13, 260) and 19 (14, 26) (learning, proficiency, respectively) vs 23.2 and 18 [17, 18].

No significant differences were noted between the study groups regarding post-operative complications or 30-day outcomes (Table 5). The rate of AKI in the learning group was higher (13.9% vs 3.9%) and approached significance (p = 0.055). This difference could be related to changes in the pre-operative optimization and the entire perioperative pathway that has been established over the study time frame. The cardiac complications recorded were all for atrial fibrillation except one hypertensive urgency in the learning group. The 30-day re-operation rate was 13 0.7% in the proficiency phase. Review of all operative notes revealed that all 7 cases were exams under anesthesia with drain placement if indicated for concern for anastomotic leaks. None in this group required any major abdominal re-operation.

Our leak rate was 5.1% and 7.8% in the learning and proficiency groups, respectively, compared to 14.8% (ROLARR) and 12.1% (Kim). SSI rates were also lower, 1.3% and 0%, compared with 8.9% (ROLARR) and 1.5% (Kim). Average length of stay (LOS) for the RCTs were 8.0 (ROLARR) and 10.3 days (Kim), vs median (IQR) 4 (3, 6) and 4 (3, 5) days in this study. Nearly all patients in the Kim study had a DLI (98.5%), whereas 60.2% of the patients in the ROLARR trial underwent DLI [17, 18].

A small study by a single surgeon in a community hospital showed similar outcomes after minimally invasive rectal resection compared with larger institutions, pointing out that perhaps the support and availability of a multidisciplinary team matters more than the hospital size/classification itself [19]. We believe this is also reflected in our data with outcomes at a community teaching center that are very consistent with previously published large series. In addition, the learning curve, specifically the learning phase, did not influence the outcomes negatively, showing the safety of performing rectal cancer care in the community and teaching setting. We must note we do have an established multi-disciplinary team of specialized providers in rectal cancer management and have set standards for pathologic and radiologic synoptic reporting, which have certainly helped in ensuring delivery of care along nationally established guidelines.

A review of nearly 24,000 patients who underwent rectal resection (open, laparoscopic, or robotic) was performed using the National Cancer Database to assess for predictors of robotic utilization for rectal cancer resection. They found that younger patients at academic institutions were significantly more likely to have a robotic rectal resection compared to open or laparoscopic [20]. We believe that based on our data this can be extended to community hospitals with well-trained robotic colorectal surgeons.

Limitations of this study include its retrospective nature and the evaluation of only a single surgeon’s outcomes. The latter was unavoidable since our institution only has one colorectal surgeon who routinely performs colorectal resections using the robotic platform. Also, there were many incomplete/non-standardized pathology reports in the learning phase prior to establishing the synoptic reporting, which can certainly influence the data reporting.