Learning curves in laparoscopic distal pancreatectomy: a different experience for each generation

Background: Learning curves of laparoscopic distal pancreatectomy (LDP) are mostly based on ‘self-taught’ surgeons who acquired sufficient proficiency largely through self-teaching. No learning curves have been investigated for ‘trained’ surgeons who received training and built on the experience of the ‘self-taught’ surgeons. This study compared the learning curves and outcome of LDP between ‘self-taught’ and ‘trained’ surgeons in terms of feasibility and proficiency using short-term outcomes. Materials and methods: Data of consecutive patients with benign or malignant disease of the left pancreas who underwent LDP by four ‘self-taught’ and four ‘trained’ surgeons between 1997 and 2019 were collected, starting from the first patient operated by a contributing surgeon. Risk-adjusted cumulative sum (RA-CUSUM) analyses were performed to determine phase-1 feasibility (operative time) and phase-2 proficiency (major complications) learning curves. Outcomes were compared based on the inflection points of the learning curves. Results: The inflection points for the feasibility and proficiency learning curves were 24 and 36 procedures for ‘trained’ surgeons compared to 64 and 85 procedures for ‘self-taught’ surgeons, respectively. In ‘trained’ surgeons, operative time was reduced after completion of the learning curves (230.5–203 min, P=0.028). In ‘self-taught’ surgeons, operative time (240–195 min, P≤0.001), major complications (20.6–7.8%, P=0.008), and length of hospital stay (9–5 days, P≤0.001) reduced after completion of the learning curves. Conclusion: This retrospective international cohort study showed that the feasibility and proficiency learning curves for LDP of ‘trained’ surgeons were at least halved as compared to ‘self-taught’ surgeons.


Introduction
The implementation of laparoscopic distal pancreatectomy (LDP) has developed exponentially since its introduction in 1994 [1,2] . In recent years, numerous observational studies and systematic reviews comparing LDP with open distal pancreatectomy (ODP) demonstrated comparable or favorable outcomes of LDP, making LDP considered a safe alternative to ODP [3][4][5][6][7][8] . Two randomized trials showed the superiority of LDP over ODP in terms of time to functional recovery, length of hospital stay, intraoperative blood loss, and delayed gastric emptying with comparable mortality rates [9,10] . Based on these promising results, the Miami Guidelines on minimally invasive pancreatic resection recommended LDP over ODP as surgical treatment for benign and low-grade malignant tumors in experienced centers [11] .
However, as LDP remains a complex and technically difficult procedure, careful implementation is required to minimize a negative impact of the learning curve on patient outcomes. For that reason, the implementation of LDP has followed the surgical IDEAL framework, a structured model that describes five stages of innovation in surgery to ensure the safe implementation of novel surgical techniques [12][13][14][15] . Besides this, increasing attention has been devoted to the assessment of learning curves since studies have shown that the adoption of minimally invasive pancreatic surgery during the learning curve can lead to an increase in morbidity and mortality rates [16,17] . As a result, many studies assessing the learning curve of laparoscopic distal pancreatectomy have been published [18][19][20][21][22] . However, most of these studies investigated the learning curves in single-center case series or in the experience of the first generation of 'self-taught' surgeons who acquired sufficient proficiency largely through selfteaching, while studies to the learning curves of the next generation of 'trained' surgeons are still lacking.
Therefore, the aim of this study is to compare the learning curves and outcome of LDP in 'self-taught' surgeons and 'trained' surgeons in terms of phase-1 feasibility and phase-2 proficiency using intraoperative and postoperative outcomes such as operative time and major complications. This may eventually contribute to a more nuanced learning curve identification specifically targeted at different generations of surgeons.

Patients and design
Data of all consecutive patients who underwent elective LDP by 'self-taught' surgeons and 'trained' surgeons in the period 1997-2019 were collected from a retrospective database of tertiary referral centers participating in the European Consortium on Minimally Invasive Pancreatic Surgery (E-MIPS) or International Consortium on Minimally Invasive Pancreatic Surgery (I-MIPS). The collected data included the name of the surgeon so inclusion criteria for participating surgeons could be determined. Surgeons who were experienced with a minimum of 40 LDPs in total with at least 10 LDPs per year and who performed their LDPs in the same tertiary referral institution (i.e. no outside institutions) were included. The study was conducted according to the principles of the Declaration of Helsinki (64th Fortaleza Brazil, October 2013) and followed the guidelines of the Strengthening The Reporting Of Cohort Studies in Surgery (STROCSS, Supplemental Digital Content 2, http://links.lww.com/JS9/A436) [23] . The ethical board of Amsterdam UMC waived the need for informed consent due to the retrospective design.

Outcomes and definitions
Surgeons were labeled as 'self-taught' surgeons if they independently started performing LDP during time phase-2 of the IDEAL framework, the Development and Exploration phase [15] , without receiving any specific form of training. The next generation of 'trained' surgeons were defined as fellowship-trained surgeons who received training without prior independent experience in LDP [24] . They received education in LDP and had training possibilities in the forms of fellowships, simulation and proctored programs, and (hands-on) courses. Preoperative variables and operative details included baseline characteristics such as age, sex, and BMI, American Society of Anesthesiologists (ASA) classification [25] , tumor size, pancreatic ductal adenocarcinoma (PDAC), and the type of spleen-preserving procedure. Study endpoints included surgical variables such as operative time, intraoperative blood loss, multivisceral resection, conversion, and failure to preserve the spleen and postoperative variables as clinically relevant postoperative pancreatic fistula (POPF grade B/C), major complications, length of hospital stay, readmission, and mortality. Postoperative outcomes were recorded up to 90 days postoperatively. The definition of clinically relevant pancreatic fistula followed the definitions of the International Study Group on Pancreatic Surgery (ISGPS) [26] . Major complications were classified by the Clavien-Dindo classification of surgical complications, defined as a Clavien-Dindo grade 3a or higher [27] . Conversion refers to any procedure that started as a laparoscopic procedure but required conversion to open surgery [28] . Spleen-preserving procedures were classified according to the Kimura [29] splenic vessels preserving or Warshaw [30] splenic vessels resecting method. Learning curves for feasibility (phase-1) were based on operative time, and learning curves for proficiency (phase-2) on major complications, as described in previous literature [18,31] .

Statistical analysis
Data analysis was performed using IBM SPSS statistics for Windows version 26.0 (IBM). Categorical data were presented as proportions and compared between groups using χ 2 test. Continuous data were reported as median with interquartile range (IQR) due to the skewness of the data. The Mann-Whitney U test was used to compare the continuous data between two groups, and the Kruskal-Wallis test was used for comparison between more than two groups. Univariable and multivariable linear and logistic regression analyses were performed to create risk-adjusted models for variables independently associated with operative time and major complications. Variables with P < 0.1 at univariable analysis were included in the multivariable model. Statistical significance was set at a two-sided P value of <0.05.

Learning curve analysis
Risk-adjusted cumulative sum (RA-CUSUM) analysis was used to establish feasibility and proficiency learning curves. RA-CUSUM analysis is a graphical method that portrays the accumulated sequential difference between the data of individual cases (observed outcome) and the mean value of all data (expected outcome) while adjusting for the risk of a particular case mix [32,33] . To establish learning curves for the groups of either 'self-taught' or 'trained' surgeons rather than for 8 individual surgeons, the methodology of a similar study by Halls et al. [34] was followed, although there are also examples of other recent publications that performed multi-surgeon or multicenter CUSUM analysis [35,36] . First, data from all patients were combined to construct risk-adjusted regression models for operative time and major complications using univariable and multivariable regression analyses with backward selection. The final model of operative time included male sex, ASA I-II classification, and multivisceral resection, and the final model of major complications included only BMI > 30 kg/m 2 (Supplementary  Tables 1 and 2, Supplemental Digital Content 1, http://links.lww. com/JS9/A435). After this, data of 'self-taught' surgeons and 'trained' surgeons were analyzed separately to create separate RA-CUSUM plots. The RA-CUSUM plots of the operative time and major complications were calculated using the following equation: . X i indicates the observed operative time or occurrence of major complications, τ indicates the mean operative time or event rate of major complications. P i indicates the expected mean operative time or event rate of major complications based on the logistic regression model in each case. The cases of each surgeon within the group of 'self-taught' or 'trained' surgeons were chronologically numbered from the first case to the last case based on the date of operation. Hereafter, the corresponding cases (i.e. case 1 with case 1, case 2 with case 2, and so on) of each surgeon within a group were pooled with equal weighting. In this way, a specific case point represented the average performance of a group (consisting of four 'self-taught' or four 'trained' surgeons) instead of an individual surgeon. Since participating surgeons only had to perform a minimum of 40 LDPs to be included in the study, not all surgeons performed the same number of LDPs, which may have meant that in higher case numbers, the learning curve point was based on fewer surgeons than four. In the final RA-CUSUM plot, every case was plotted from left to right, where a rising curve indicated longer operative times or higher occurrence of complications compared to the overall mean, whereas a falling curve indicated shorter operative times or lower complications compared to the overall mean thus indicating surgical success. A plateau in the learning curve indicated a steady performance level. Changes in performances or in learning curve phases were based on the inflection points in the curves. The significance of these points was tested by comparing the clinical outcomes of either 'self-taught' surgeons or 'trained' surgeons before, between, and after the inflection points of the feasibility and proficiency learning curves.

Study population
In total, 639 patients undergoing LDP by four 'trained' surgeons and four 'self-taught' surgeons from five centers were included. Overall, 215 patients were operated by 'trained' surgeons and 424 by 'self-taught' surgeons. Patient characteristics and operative details are shown in Table 1. Overall, 'trained' surgeons performed fewer Kimura spleen-preserving procedures (21.3 vs. 83.6%, P ≤ 0.001) compared to 'self-taught' surgeons. The median operative times were 210 and 215 min (P = 0.769), median blood loss was 100 and 200 ml (P ≤ 0.001), and the rate of conversion was 4.2 and 4.2% (P = 0.972) for 'trained' surgeons and 'self-taught' surgeons, respectively.

Learning curve and inflection points
Analysis of the RA-CUSUM learning curves of the 'trained' surgeons revealed inflection points of the feasibility learning curve for an operative time at 24 procedures (Fig. 1A) and the proficiency learning curve for major complications at 36 procedures ( Fig. 2A). Learning curves of the 'self-taught' surgeons showed inflection points at 64 and 85 procedures, respectively (Figs 1B, 2B). The learning curves are more detailed and displayed in Supplementary Figures 1 and 2, Supplemental Digital Content 1, http://links.lww.com/JS9/A435.

Outcomes within 'trained' surgeons
Analysis of the outcomes and operative trends in 'trained' surgeons based on the inflection points of 24 (feasibility) and 36 (proficiency) procedures are shown in Table 2. A significant increase in ASA III-IV patients was found in the proficiency phase compared to the feasibility phase (9.5 vs. 25%, P = 0.018). Operative time significantly reduced after the feasibility learning curve (230.5-187.5 min, P = 0.028). Intraoperative blood loss decreased as well after the feasibility learning curve but increased after the proficiency learning curve, so no significance was reached (120-71.5-100 ml, P = 0.532). The rate of major complications increased during the proficiency phase but decreased after its completion (25.5-17.4%, P = 0.480). Failure rates of spleen preservation significantly increased during the learning curve phases (2-21.6%, P = 0.011). Outcomes within 'self-taught' surgeons Analysis of the outcomes and operative trends of the 'self-taught' surgeons based on the inflection points of 64 (feasibility) and 85 (proficiency) cases are shown in Table 3. Through the learning curve phases, 'self-taught' surgeons performed significantly more resections for PDACs (6.1-25.9%, P ≤ 0.001) and treated older and more obese patients (54.5-68 years, P ≤ 0.001 and 23.4-25.6 kg/m 2 , P ≤ 0.001, respectively). Operative time and intraoperative blood loss significantly reduced after the feasibility learning curve (240-195 min, P ≤ 0.001 and 200-150 ml, P = 0.019). The rate of major complications remained unchanged during the feasibility and proficiency phases but significantly decreased after surpassing them (20.6-7.8%, P = 0.008). Length of hospital stay significantly decreased during and after both learning curve phases (9-6-5 days, P ≤ 0.001). These changes were associated with a slightly decreased readmission rate during the feasibility and proficiency phases (8.2-6.3%) but a significantly higher readmission rate after their completion (6.3-19.8%, P = 0.002).

Discussion
This retrospective international cohort study found that the feasibility (24 vs. 64) and proficiency (36 vs. 85) learning curves for LDP were at least halved among 'trained' surgeons as compared to 'self-taught' surgeons. Intraoperative parameters such as operative time and blood loss, as well as postoperative outcomes such as major complications and length of hospital stay, improved for both groups after the learning curves.  Previous studies have identified learning curves for LDP ranging from 10 to 50 procedures [18][19][20][21][22]31,37] . In those studies, intraoperative parameters such as operative time and blood loss were used to assess the learning curves. These parameters are assumed to illustrate the surgeons' ability to complete the procedure and are therefore used to reflect the feasibility and first phase of the learning curve [18,31] . The feasibility learning curves of 24 procedures for 'trained' surgeons versus 64 procedures for 'self-taught' surgeons in the present study were based on operative time and did therefore align with the previously published learning curves. On the contrary, the obtained proficiency learning curve of 85 procedures of 'self-taught' surgeons in this study is substantially longer. However, proficiency learning curves reflect more than just a surgeon's ability to complete the procedure; they also reflect surgical proficiency and standards, thus the surgeon's ability to achieve excellence in surgical outcomes. This should be measured by postoperative outcomes such as major complications, readmissions, or length of hospital stay, which indicate advanced experience and thus reflect the second phase of the learning curve [31] . Logically, it takes more procedures to reach this phase. A previous study has suggested that this can be reached between cases 72 and 100, most probably around 80 LDP procedures, based on readmissions and length of hospital stay [18] . The proficiency learning curves of this study were based on major complications and confirmed the previous findings, suggesting that it can require 85 procedures to overcome the proficiency learning curve phase.
For both 'self-taught' and 'trained' surgeons, operative trends and clinical outcomes improved after the inflection points of the feasibility and proficiency learning curves. With increasing experience, 'self-taught' surgeons operated on older and more complex patients while achieving a significant decrease in intraoperative blood loss, operative time, major complications, and length of hospital stay. Although the latter (shorter hospital stay) may have been at the cost of an increase in readmissions. With respect to the 'trained' surgeons, only a significant improvement in operative time was found during the feasibility and proficiency phase, but many operative trends were observed. With increasing experience, 'trained' surgeons operated on older patients with higher ASA scores, a higher proportion of PDAC, and more multivisceral resections. However, this may also have contributed to the slight increase in operative time after the feasibility and proficiency phases. Postoperative outcomes such as POPF grade B/C and major complications substantially decreased after the learning curve phases but did not reach statistical significance.
An important finding, and consistent with the findings of a comparative study in laparoscopic liver surgery [34] , is that the learning curves of 'trained' surgeons of the present study were shorter than those of 'self-taught' surgeons. Those findings suggest that 'trained' surgeons may overcome their learning curves much faster than 'self-taught' surgeons may. Since the introduction of LDP in 1994 [1] , its implementation has followed the steps of the IDEAL framework for surgical innovation. 'Self-taught' surgeons of the present study started performing LDP in 1997 when LDP was still in its development and exploration phase. In this era, they had no prior knowledge of minimally invasive pancreatic surgery, consensus guidelines, innovation, or support for a specific technique or instrument. Therefore, 'self-taught' surgeons had to find their way with the complexity and novelty of the laparoscopic technique, which was mainly based on the concept of 'trial and error'. 'Trained' surgeons, on the other hand, started performing LDP in 2012. Meanwhile, laparoscopic surgery had become more widely applied, studies on surgical techniques had been published, guidelines had been introduced, and instruments had been improved. Moreover, 'trained' surgeons presumably thrived on the experiences gained by the 'selftaught' surgeons and benefited from proctoring, fellowships and training possibilities. The shortened learning curves of 'trained' surgeons compared to 'self-taught' surgeons of the present study do, therefore, not only reflect a different generation but also the effect of a different era centered on technical innovation, improved knowledge, evidence-based practice, and training possibilities. It illustrates that learning curves should not be considered as defined aspects but as flexible concepts that are subject to many external circumstances. Several limitations apply to this study. First and most importantly, its retrospective design. Due to the main objective to examine learning curves from different time stages in the surgical innovation of LDP, the time aspect of the study has a potential impact in which changes in preoperative and postoperative policies may have influenced the outcomes. Interestingly, in the risk-adjusted regression model, ASA 1-2 was associated with increased operative time, which is a finding we are unable to scientifically explain but may be due to the retrospective data collection. In addition to this, as the application of laparoscopic surgery increased over time with widened indications for LDP, 'trained' surgeons, as compared to 'self-taught' surgeons, probably performed more cases in the same time period. This also applies to other laparoscopic procedures and might have contributed to the reduction of the learning curve of 'trained' surgeons. Second, no data was available on the surgeons' level of experience in open distal pancreatectomy. This could be of interest as it may have positively affected their LDP performance. Third, as LDP evolves through time, just like any other surgical procedure, the obtained learning curves of this study are not static and must be revised periodically in relation to the future adoption of LDP.
The main strength of this study is that it is the first multicenter study to identify and compare learning curves in LDP for different generations of surgeons as well as different surgical learning curve phases. Many studies on the learning curves in LDP date back to earlier years and investigated the learning curves of surgeons of the first generation, while this study aimed to examine the effect of further implementation and training in LDP on the second generation of surgeons. The obtained differences in learning curves of this study imply that it should be recommended that surgeons of the current generation must first complete acknowledged training programs in terms of fellowship, proctored programs, or courses before they start performing LDP. Additionally, the obtained learning curves may be adopted in the design of future surgical trials.

Conclusion
This multicenter study shows the impact of the implementation and standardization of LDP over the last two decades and illustrates that learning curves must be considered flexible curves that are dependent on many factors, including time, innovation, evidence-based practice, and training. The reduction in the learning curve of 'trained' surgeons compared to 'self-taught' surgeons show the importance of education and training in the implementation of novel techniques and that these elements can contribute to early improvements of intra and postoperative outcomes.

Ethical approval
The ethical board of Amsterdam UMC approved this study and waived the need for informed consent due to the retrospective design and the use of pseudonymized data.

Sources of funding
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Conflicts of interest disclosure
There are no conflicts of interest.
Research registration unique identifying number (UIN)

Data availability statement
The data are confidential and only available upon reasonable request.