Abstract
Purpose:
Robot-assisted simple prostatectomy (RASP) has excellent outcomes when treating large volume prostates and incorporates the already familiar skills to most robotic surgeons. Our objective was to determine the learning curve for RASP.
Materials and Methods:
A retrospective review of RASP on 120 consecutive cases performed by two experienced robotic surgeons from 2014 to 2017 was conducted. We defined “learning curve” as the point at which operative parameters transition from logarithmic to linear improvement. Scatter plots of operative outcomes were constructed and logarithmic and linear best-fit line were estimated to determine the point of transition from logarithmic to linear improvement.
Results:
Surgeon 1 operated on 76 cases and surgeon 2 on 44 cases. The median age of the 120 patients who underwent RASP was 70.0 years (interquartile range [IQR] 65.0–74.0 years) and median prostate mass was 121.5 g (IQR = 102.0–149.3). Overall, high-grade complication rate was 7.5%; median hematocrit change was 5.4% (IQR = 3.2–7.7) and tissue yield was 61.2 g (IQR = 49.7–76.9). Tissue yield demonstrated logarithmic improvement over the first 12 cases and then transitioned to a linear patter for one surgeon. Operative time in the last 10 cases was statistically different from the first 10 cases (p < 0.01). Drop in hematocrit (ΔHct) for surgeon 2 demonstrated logarithmic improvement for the first 10 cases and then transitioned to a linear pattern.
Conclusion:
The learning curve for RASP varied depending on the variable examined. Blood loss (ΔHct) and tissue yield showed the greatest improvement over time, but neither showed significant improvement beyond 12 cases. We estimated the learning curve for RASP to be ∼10 to 12 cases for experienced robotic surgeons.
Introduction
B
Methods
Surgeon experience
This study coupled data of two experienced robotic surgeons. Surgeon 1 specializes in lower urinary tract interventions and has completed >2500 laparoscopy robot-assisted radical prostatectomies. Surgeon 2 is fellowship trained in robotic operation and has completed >250 laparoscopy robot-assisted radical prostatectomies. Surgeon 1 performed 76 RASPs between December 2014 and September 2017, whereas Surgeon 2 performed 44 RASPs between January 2016 and September 2017 (120 RASPs of the two surgeons). Both surgeons had similar preoperative work-up and postoperative follow-up. Both surgeons obtained prostate sizing by either cross-sectional imaging or transrectal ultrasonography. Both surgeons were responsible for teaching residents and fellows. Portions of the case that were not performed by each surgeon were directly supervised by the surgeon.
Study population
American Urological Association (AUA) symptom scores and noninvasive urodynamic studies were collected prospectively consistent with our clinical practice, whereas all perioperative data were collected retrospectively. Patients without prostate imaging underwent a digital rectal examination and subsequent prostate imaging if it was felt to be larger than 80 g. Patients with prostate volume >80 g were offered RASP as surgical treatment for LUTS. After RASP, each patient was seen in clinic within 2 to 4 weeks with an evaluation of stress urinary incontinence (SUI), flow rate, and postvoid residual (PVR).
Surgical technique
The surgical steps of RASP are outlined in Table 1. Both surgeons utilized a Da Vinci Si or Xi surgical robotic system (Intuitive Surgical, Sunnyvale, CA). The transperitoneal approach was similar to previously described, without entering the retropubic space (dropping the bladder). 11 Continuous bladder irrigation was started immediately after bladder closure and turned off the morning of postoperative day 1 or as clinically appropriate.
Learning curve parameters
After Institutional Review Board approval, retrospective data were extracted from the medical record. Preoperative, intraoperative, and postoperative measures were obtained. Primary surgical outcomes included percentage of prostate tissue yield, preoperative to postoperative hematocrit drop (ΔHct), complications, and operating time. Adenoma mass was obtained through digital scale immediately after extraction by the operating surgeon before being sent to pathology analysis. Percentage prostate tissue yield was calculated by dividing this weight by preoperative imaging estimation of volume. Hematocrit drop was calculated by subtracting postoperative day 0 hematocrit (drawn between 4 and 12 hours postoperatively—whichever lowest) from preoperative hematocrit. Operative time was calculated from start of insufflation to skin closure. Patient cohorts were defined as early, intermediate, and late. The early cohort consisted of the surgeons' first 10 cases. The intermediate cohort consisted of the surgeon cases 11 to 20. The late cohort was the surgeon's most recent 10 patients.
Statistical analysis
All statistical analyses were performed with IBS SPSS Statistics version 22. Statistical analysis was performed with chi-square analysis, nonparametric Kruskal–Wallis test. A p-value of <0.05 was considered to be statistically significant. Initial analysis was performed by comparing early, intermediate, and late patient cohorts. Preoperative, perioperative, and postoperative parameters were collected for each cohort and compared.
In addition, the learning curve was also calculated as described by Brunckhorst and colleagues for evaluating HoLEP. 7 “Learning curve” was defined as the point at which operative parameters transition from logarithmic to linear improvement. Perioperative outcome measures were plotted vs case number and evaluated in sets of 20 cases. This number of cases was selected because it allows a sufficient number to determine a best-fit line and not so many that the proficiency of the surgeon dominates the cohort. Both linear and logarithmic lines of best fit were determined over this 20-case window. This analysis was performed incrementally across case experience (cases 1–20 were analyzed, then 2–21, then 3–22, etc). The R 2 value for both the linear and logarithmic curves were compared to determine if the plot better fit a liner or logarithmic model over each set of cases. The point at which the linear model became a better fit than the logarithmic model was determined as the estimated learning curve. This process is given in Figure 1.

In addition, we sought to evaluate the degree of change of the operative parameters after the initial learning curve. For each surgeon, operative parameters were plotted excluding the first 20 cases. A linear line of best fit was applied and the slope of that line was calculated. The slope indicates rate of change per case of the particular operative parameter. This gives an indication of how the parameters continue to change as the surgeon becomes proficient after the initial learning curve.
Results
The median age of the 120 patients who underwent RASP was 70.0 years (interquartile range [IQR] = 65.0–74.0 years) and median prostate mass was 121.5 g (IQR = 102.0–149.3). Median preoperative Prostate specific antigen (PSA) was 6.4 ng/mL (IQR = 4.0–10.0). Median follow-up was 3.2 months (IQR = 1.0–6.5). Twenty-nine of 120 (24%) patients presented with urinary retention (Table 2). Thirty-three cases were performed in the first 15 months of this study. In the later 15 months, 87 cases were performed, indicating that RASP was utilized more, as the surgeons became more comfortable with this approach.
BMI = body mass index; IQR = interquartile range; Hct = hematocrit; PSA = prostate specific antigen.
Among all RASP procedures, the median hematocrit drop was 5.4% (IQR = 3.2–7.7) and 3.3% patients necessitating a blood transfusion. Median operative time was 157 minutes (IQR = 136–180). Malignant pathology report was found in 11% pathologic specimens. Only grade group 1 prostate cancer was identified. Median length of stay was 1.0 day (IQR = 1.0–2.0). The incidence of transient stress urinary incontinence was 5.0%. No patient experienced urinary incontinence beyond 1 month. High-grade complications (Clavien III or greater) occurred in 7.5% patients. The overall complication rate was 18%. Median prostate tissue yield was 63 ± 20%. Median catheterization time was 4.0 days (IQR = 4.0–6.0). Median postoperative PSA dropped by 90.1% (IQR = 79.8–93.9). Functionally, RASP had a median peak flow rate improvement (ΔQmax) of 9.9 mL/second (IQR = 2.2–15.9) and median AUA symptom score improvement of 13.0 points (IQR = 6.0–17.3). Overall, median postoperative peak flow rate (Qmax) was 18.8 mL/second (IQR = 13.7–26.4).
Comparing early and late cohorts, there was a statistically significant improvement in median operative time (162 and 134 minutes, respectively; p = 0.01). The intermediate cohort demonstrated a similar median operative time to the early cohort (178 minutes). Both ΔHct and percentage tissue yield did not vary among the cohorts. Median tissue yields were consistent with 61.5, 53.7, and 60.4 g for early, intermediate, and late cohorts, respectively (p = 0.7). Median ΔHcts were 6.0%, 4.7%, and 4.1% for early, intermediate, and late cohorts, respectively (p = 0.4). A higher rate of high-grade complications (Clavien III or greater) were not seen in the early cohort (0%, 5%, and 0% for early, intermediate, and late, respectively).
The median length of catheterization was 4 days (IQR = 4–6), and this did not vary significantly between the early, intermediate, and late cohorts. The incidence of postoperative transient stress urinary incontinence improved from the early cohort to the late cohort. In early group 10% patients developed transient stress urinary incontinence and this decreased to 5% in the intermediate group. No one from the late group had SUI (p = 0.16). No patient experienced urinary incontinence beyond 1 month. Functional outcomes (postoperative Qmax and PVR) were similar among these cohorts. Likewise, change in International Prostate Symptom Score (IPSS) and PSA were not significantly different between cohorts (Table 3).
ΔHct = drop in hematocrit; IPSS = international prostate symptom score; PVR = postvoid residual.
The learning curve was determined by analyzing scatter plots of operative parameters vs case number as described in the “Methods” section. The point at which the linear best-fit line outperformed the logarithmic line was determined. For surgeon 1, hematocrit and operative time maintained a negative linear relationship throughout the case experience. Tissue yield percentage, however, transitioned from logarithmic to linear improvement in case number 12. For surgeon 2, operative time maintained a negative linear relationship throughout the case experience, and tissue volume did not trend either direction. Reviewing hematocrit drop, surgeon 2 demonstrated logarithmic improvement over the first 10 cases, and linear improvement thereafter (Fig. 1).
In an attempt to determine the rate of change in operative parameters once the surgeon was out of the learning curve, case parameters were replotted excluding the first 20 cases. Best-fit lines were drawn and the slopes of these lines were measured. A linear best-fit model was chosen here as the surgeons were out of their learning curves. For both surgeons, there was minimal improvement in tissue yield after the initial learning curve (0.08% and 0.9% prostate tissue yield per case, for surgeon 1 and surgeon 2, respectively). Likewise, hematocrit drop changed little after the first 20 cases (0.1% and −0.09% change in hematocrit per case for surgeon 1 and surgeon 2, respectively). With respect to operative time, surgeon 1 continued to decrease the operative time by 0.9 minutes per case, whereas surgeon 2 improved at a rate of 1.8 minutes per case (Table 4).
First 20 cases excluded.
In summary, RASP is a safe and effective treatment modality for BPH with enlarged prostates. Learning curve analysis demonstrates that operative parameters stabilize after about 10 cases for experienced surgeons. The greatest improvement is operative time as the surgeon performs more cases. As surgeons performed more RASPs, there was a trend toward fewer complications; however, this was not significant.
Discussion
RASP has been demonstrated to be safe and effective in a number of studies. 12 –14 Indeed, we have previously shown that RASP, when compared with open simple prostatectomy, has equivalent functional outcomes with the advantage of reduced length of stay and decreased blood loss. 9 Other minimally invasive procedures for large volume BPH, such as HoLEP, have been demonstrated to be efficacious as well, although technically challenging. 15 Thulium laser vaporization is likewise an emerging treatment modality for high-volume prostates with promising retrospective data. 4 From our experience RASP is a relatively easy to technique to adopt, and we thus set out to better define the learning curve for urologists already comfortable with robotic procedure.
The urology literature tends to vary in how learning curve is defined and calculated. Some studies have evaluated temporal parameters such as operative time, enucleation time, and morcellation time, 16 whereas others focused on postoperative outcomes such as Qmax, PVR, quality of life, and IPSS score. 6 Given the significant variability in the literature, learning curves have been difficult to define accurately.
Our objective was to determine the learning curve for RASP by examining the point at which the surgeon transitioned to a state of proficiency. The distinction between learning curve and surgeon proficiency is subtle. By evaluating the operative parameters in both a linear and logarithmic context, we sought to make a more objective definition of the learning curve. Not every parameter for both surgeons followed this transition from logarithmic improvement to linear improvement which is not surprising given the inherent variability in surgeon, patient, and case.
Overall, RASP can be completed safely and effectively, especially for experienced robotic surgeons. After 10 to 12 cases, the period of logarithmic improvement was no longer seen for any variable, and the surgeon's improvement progressed in a linear manner. Of note, none of the functional outcomes or complication rates improved after the first 10 cases, which indicates that although surgeon efficiency did progress, there is little substantive improvement for the benefit of the patient. The operative parameters after the first 20 cases were plotted against case number to verify that the surgeons were out of the learning curve. After the first 20 cases, tissue yield and hematocrit drop changed little for the both surgeons. This reinforces the conclusion that by case 20, one is outside the learning curve. Operative time improved at a rate of 1 to 2 minutes per case, which represents the gaining proficiency of the surgeon and becoming more efficient with the steps.
Many other studies have sought to evaluate learning curve for technically challenging procedures. 15,17,18 A prospective study found that the learning curve for ThLEP was ∼8 to 16 cases if an experienced mentor surgeon was available. 19 Brunckhorst et al. did a similar analysis of HoLEP and found the learning curve to be ∼50 cases. 7 Within the first 40 cases, the complication rate was ∼20% and three of these patients developed transient incontinence. Of interest, despite RASP being a transabdominal procedure, HoLEP is associated with a similar complication rates. 4,20 These rates improved after 60 to 80 cases. Similar to our study, case data were reviewed retrospectively and studied through scatter plots. The learning curve parameters used were enucleation efficiency, morcellation efficiency, and complication rate. Logarithmic and linear best-fit curves were applied and R 2 values analyzed. In their analyses, determination of the length of the learning curve was assessed by visually inspecting the plateauing of the best-fit line. 7 In our study, we performed a similar analysis, but used a more quantitative measure of finding the plateauing of the operative parameters. By comparing the R 2 values of the logarithmic and linear best-fit lines, the transition and thus learning curve were better defined. Compared with the steep learning curve for HoLEP, we found the RASP learning curve to be 10 to 12 cases. One reason for the steeper learning curve seen with HoLEP may be the unfamiliar techniques used. RASP however incorporates skills already familiar to a surgeon experienced in robotic radical prostatectomy.
This study has several strengths. Both surgeons had no experience with RASP before commencement of the study; however, they each had extensive experience in robotic procedure. All bedside assistants were experienced advanced practice providers minimizing the variation associated with manipulation of the robot instruments, passing sutures, and suctioning. This is a single-center study that minimizes many of the operating room variables. However, there are also limitations to this study. It evaluates two experienced robotic surgeons; therefore, the results may not apply to less experienced surgeons. Given the retrospective nature of this study there are certain aspects of the learning curve that we were unable to analyze. Being able to break down the operative time analysis by portions of the procedure (port placement, adenoma resection, cystography, etc.) would allow for a better understanding of what improves during the learning curve. However, this is not possible as these data were not collected prospectively. In addition, this study consists only of 120 patients, and therefore changes in these curves occurring after 120 patients may not be recognized, but seems statistically unlikely. Since the adoption of RASP by each surgeon, fewer than five open simple prostatectomies have been undertaken at our institution. Likewise, follow-up is somewhat limited. With longer follow-up, better assessments of functional outcomes may be made as to the durability of our findings, but would not likely influence the calculation of the learning curve.
Conclusions
In the hands of experienced robotic surgeons, RASP is a safe and effective surgical intervention for large volume, symptomatic BPH. The learning curve for RASP among experienced robotic surgeons appears to be ∼10 to 12 cases based on variables with the most pronounced learning curve (blood loss and tissue yield). Further studies are needed to better elucidate the learning curve for trainees and surgeons with less robotic experience.
Footnotes
Author Disclosure Statement
No competing financial interests exist.
