This study primarily investigated the learning process of RLR and LLR for a surgeon trained simultaneously in both MILS techniques and compared it to LC of LLR of another young consultant in otherwise similar conditions. The study, which is the first of its kind to our knowledge, found: 1) a substantial similarity of the progression in the robotic and laparoscopic technique for up to intermediate difficulty liver resections, countering the view of a faster learning curve for robotic surgery, and 2) the feasibility and safety of the simultaneous training in both MILS techniques at least for this type of procedures.
To measure the learning process of a surgical procedure is a challenging task.(20) Existing studies in the field of MILS are very heterogeneous as used different types of statistical analysis or design, some included multiple-surgeons or institutional LC (sometimes not specifying the number of surgeons involved) rather than single-surgeon series, included a single standardized procedure (i.e., left lateral sectionectomy, or right hepatectomy, etc.) rather than a cumulative experience of consecutive resections. Moreover, different variables can be used to estimate the LC (operative time, blood loss, conversions to open surgery, postoperative morbidity, progression in the difficulty score of MILS, etc.,).(11, 14, 21, 22) In summary, a consensus among clinicians is lacking regarding the optimal methodology to measure the LC, making difficult to extrapolate solid conclusions. A recent systematic review with meta-regression analysis by Chua, et al. queried the literature about the LC of MILS: out of 15 studies included for the quantitative analysis, 7 measured the LC of RLR. The number of cases needed to overcome the LC ranged from 11 to 30, but mixed anatomically minor and major hepatectomies.(15) Interestingly, the LC was measured with the CUSUM methodology in only 3 studies, was arbitrarily assessed in 3 others, finally one study identified the learning phase as the number of cases needed to obtain a significant increase in the difficulty score index. The variables of interest were also different: operative time was used in two studies, conversion to open surgery in two, blood loss and post-operative morbidity in two others, difficulty score in one study. Moreover, three studies measured the LC of 2 surgeons while the remainders measured the institutional one.
Data on the direct comparison of the LC of RLR vs. LLR are scarce, but concordant in the sense of a shorter learning phase for RLR vs. LLR. O’Connor, et al. in 2017 showed superior outcomes (blood loss, hospital stay and post-operative complications) for RLR vs. LLR (minor hepatectomies) after 25 cases by two surgeons both already proficient in LLR. The cut-off of 25 cases however was arbitrability based on previous published papers on LLR and no CUSUM analysis was performed.(23) Efanov, et al. in 2017 defined the LC in the number of procedures necessary to significantly increase the difficulty index or the rate of resections in the posterior-superior liver segments.(21, 24) The authors reported a two-times longer LC for LLR vs. RLR (29 LLRs vs. 16 RLRs) for two surgeons who initiated RLR shortly before LLR. The mean difficulty index of RLRs was 5.0 in the learning phase vs. 7.3 (p < 0.001) once the LC was overcome (the last 24 cases). Of note, only 40 RLRs and 91 LLRs from two surgeons were included, peri-cystectomies were not excluded, cirrhotic patients were only 8% of RLR group and were all operated in the post-learning phase.(21) More recently, Krenzien, et. al used a “complexity-adjusted” CUSUM analysis of operative time, conversion rate and blood transfusions normalized by the IWATE score to compare the institutional LC of 132 RLRs vs. 514 LLRs. The authors estimated the learning phase in 93 cases for RLR and 117 cases for LLR, about 4 times longer than reported in previous studies, but still shorter for robotics. In this case RLR program was developed following almost 10 years of LLR experience.(16)
Trying to capture a real-life LC of RLR and LLR, we decided to analyze all the consecutive MILSs performed by a single surgeon (single-surgeon learning curve), irrespective of the type of resection. The choice of a single-surgeon (rather than institutional) LC was because the robotic cases were performed by only one surgeon, as in most surgical realities starting robotic surgery programs, and was supported by the literature (14 out of 19 studies focused on single-surgeon LCs in the recent meta-analysis by Chua, et al.).(15) One could arbitrarily argue that our results might not be directly translated to realities where the robotic platform is shared among several liver surgeons. However, this eventuality is rare in newborn robotic programs, and we do not consider this a methodologic limitation of the current study.
When including different surgical procedures, it is important to take into account all the confounding factors (i.e., patient-, liver- or surgery-related) as well as the difficulty of the liver resection itself.(17) We used a multivariate analysis (ANCOVA test) to adjust for confounders and we stratified the procedures by difficulty score (IWATE criteria). Then the CUSUM methodology was used to assess the learning phase primarily based on the adjusted operative time, secondarily on conversions to open surgery, post-operative complications, difficulty of MILS according to the IWATE criteria. In particular, the estimation of the LC based on the adjusted operative time has been recently adopted in other fields of robotic surgery like upper gastrointestinal and hernia surgery.(25, 26)
To further control the quality of the simultaneous training in RLR and LLR, we included a control group of laparoscopic procedures performed by a second surgeon only trained in LLR (LapSurg2) from another referral institution for HPB surgery. According to the CUSUM analysis of the adjusted operative time, the learning phase for LapSurg2 consisted in 48 cases and was indeed similar to that of the surgeon 1 in RLR and LLR (both 40 cases for RobSurg1 and LapSurg1). This taking also into account that in the LapSurg2 group there were more difficult cases (mean IWATE score of 4.0, 4.3 and 5.8 for RobSurg1, LapSurg1, LapSurg2 respectively) and the surgical background of the surgeon 2 was slightly different from that of the surgeon 1 (ongoing learning curve in OLR, no liver transplant experience).
The finding of the similar LC with both techniques counters the common knowledge of a faster LC for RLR, reported in previous studies and for other surgical domains. In fact, the groups who reported a shorter LC for RLR vs. LLR included surgeons with previous experience in LLR, or did not clearly state whether the surgeons had a background of LLRs.(15, 16, 23). Our findings may be explained by the lack of a bias due to previous experience in former studies. One other possible reason is the absence of tools specific to liver resection by robotic approach which instead are available for the laparoscopic setting (contrary to other types of robotic surgery in which these are not needed).
The question of simultaneous training is now relevant, as the new generation of trainees in HPB surgery is acceding to MILS. Major hepatobiliary centers usually started robotic programs after a solid background of laparoscopic liver surgery, and the point of the safety of simultaneous training could not be addressed. For example, D’Hondt and his team from Belgium performed LLRs-only for 8 years before the transition to RLR; similarly, the Berlin experience consisted of at least 7 years of LLRs before starting with RLRs.(11, 27) The features of a purely robotic LC would be interesting, but could not be addressed in the present study; however, it could be investigated following the same scheme.
Some limitations of this study must be addressed and deserve some comments. First, this is a retrospective study of a single-surgeon LC including only anatomically minor or low-intermediate difficulty (IWATE) RLRs: our conclusions can’t be extrapolated to the training in robotics for more complex cases. Similarly, we were not able to comment on any differences between a concomitant (RLR and LLR) versus a sequential training (LLR followed by RLR) as the surgeon from the control group is only now starting the robotic training. Second, the control group (LapSurg2) consisted of more difficult resections compared to the procedures from the Surgeon1 (RobSurg1, LapSurg1, concomitant training) resulting in la slightly longer learning phase. Conversely, the incidence of underlying chronic liver disease (notably cirrhosis) and portal hypertension rate were significantly lower in the control group. We believe this did not bias our conclusions as our primary focus was on the simultaneous training cohorts, which consisted of similar procedures (as per difficulty index) in similar populations. Third, despite MILS cases in the cohorts of simultaneous training were selected in principle to be similar (patients characteristics, IWATE score) (Tables 1 and 2), selection bias could not be completely excluded because of the retrospective nature of the analysis and the fact that the robotic platform was available only once a week or once every two weeks, while laparoscopy was freely available. The latter could have led to a more cautious selection of cases for RLRs. The ANCOVA test was adopted to account for all these confounding factors above mentioned. Last, we could not find a “learning curve effect” in the LapSurg1 and LapSurg2 groups as per postoperative complications and conversions. We hypothesized that this was because of the clinically appropriate selection bias, chosing easier cases for the beginning, and progressing carefully to more difficult resections. Accordingly, we were not able to comment on the LC based on these variables.
In conclusion, considering the limitations abovementioned, our experience showed a similar learning phase of approximately 40 cases for simultaneous training up to intermediate difficulty RLRs and LLRs. The concomitant training in both techniques was safe in the development of a MILS program. Our findings tend to counter the view of a faster learning curve for robotic liver resections (at least for low and intermediate difficulty procedures), possibly because of the lack of a bias due to previous experience in former studies, and because of the absence of dedicated robotic tools for parenchymal dissection.