The learning curves of major laparoscopic and robotic procedures in urology: a systematic review

Background: Urology has been at the forefront of adopting laparoscopic and robot-assisted techniques to improve patient outcomes. This systematic review aimed to examine the literature relating to the learning curves of major urological robotic and laparoscopic procedures. Methods: In accordance with PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines, a systematic literature search strategy was employed across PubMed, EMBASE, and the Cochrane Library from inception to December 2021, alongside a search of the grey literature. Two independent reviewers completed the article screening and data extraction stages using the Newcastle–Ottawa Scale as a quality assessment tool. The review was reported in accordance with AMSTAR (A MeaSurement Tool to Assess systematic Reviews) guidelines. Results: Of 3702 records identified, 97 eligible studies were included for narrative synthesis. Learning curves are mapped using an array of measurements including operative time (OT), estimated blood loss, complication rates as well as procedure-specific outcomes, with OT being the most commonly used metric by eligible studies. The learning curve for OT was identified as 10–250 cases for robot-assisted laparoscopic prostatectomy and 40–250 for laparoscopic radical prostatectomy. The robot-assisted partial nephrectomy learning curve for warm ischaemia time is 4–150 cases. No high-quality studies evaluating the learning curve for laparoscopic radical cystectomy and for robotic and laparoscopic retroperitoneal lymph node dissection were identified. Conclusion: There was considerable variation in the definitions of outcome measures and performance thresholds, with poor reporting of potential confounders. Future studies should use multiple surgeons and large sample sizes of cases to identify the currently undefined learning curves for robotic and laparoscopic urological procedures.


Introduction
The concept of a 'learning curve' was first described in the aeronautics industry in 1936 to illustrate the correlation between improvements in the production of plane components and the increasing experience of the workforce involved [1] . It has subsequently been adopted in the context of surgical practice, where it has been defined as the number of cases that a surgical trainee needs to undertake in order to reach the point where their inexperience no longer affects the outcomes of the procedure [2] . This particular stage in their acquisition of skills is represented on conventional learning curve graphs by a plateau [3] . Notably, although learning curves map the achievement of technical skills proficiency, surgeons must also achieve proficiency in non-technical skills in order to gain overall surgical competency in the procedure [4] .
Various learning curve metrics are used to plot the surgical learning curve. Operative time (OT) and estimated blood loss (EBL) are two of the most commonly reported learning curve metrics [5] , with reductions in these variables being associated HIGHLIGHTS • The robot-assisted prostatectomy learning curve ranges from 10 to 250 cases. • The robot-assisted partial nephrectomy curve ranges from 4 to 150 cases. • Retroperitoneal lymph node dissection and laparoscopic cystectomy learning curves are undefined. • Standardized reporting of outcomes is required to enable future meta-analysis.
with surgical training progression [6] . Urology-specific patientoutcome variables include measures of potency and continence postoperatively, the impairment of which can have a significant impact on the patient's quality of life [7] . One or more of four principal methods of analysis are then used to assess change in performance as case number increases [8] ; graphical inspection, the split-group method (dividing the data into consecutive groups and comparing group means), logistic regression and cumulative sum (CUSUM) analysis. It is highly important to define the learning curve for surgical procedures because detrimental outcomes associated with surgeon inexperience can have a major impact on patient safety, as exemplified by findings of the Bristol Royal Infirmary enquiry [9] . Knowledge of the learning curve enables the safeguarding of patients by identifying trainees in the learning phase and providing them with adequate supportive measures such as close senior supervision and additional simulation-based skill practice [10] . Mapping the learning curve can also inform the design of surgical training programmes by illustrating the impact of educational interventions on the learning process. This is particularly relevant in the context of the laparoscopic and robotic approaches which have been widely adopted as the standard of care for urological operations such as prostatectomy and radical nephrectomy. Given their relatively recent implementation in contrast to traditional open approaches, knowledge of their learning curves is important in informing and improving training programmes for these modalities so as to effectively achieve competency across all outcome measures and ensure patient safety.
The last systematic review to include an evaluation of the learning curve for laparoscopic prostatectomy was conducted in 2014 [11] . The most recent systematic review evaluating the learning curve for robot-assisted laparoscopic prostatectomy (RALP), robot-assisted radical cystectomy (RARC), robot-assisted partial nephrectomy (RAPN) and robotic pyeloplasty, conducted database searching in February 2018 [12] , although this particular review excluded studies published prior to January 2012 as well as single-surgeon studies which constitute the bulk of the relevant learning curve literature base.
The principal aim of this review is to provide an updated insight into the learning curves for major urological robotic and laparoscopic procedures to act as a reference for expected progress. Other aims include the evaluation of the methods used to analyse the learning curves and to provide recommendations for future studies in the field in terms of their scope and methodology.

Methods
Design A systematic review was performed utilising the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA), Supplemental Digital Content 1, http://links.lww.com/ JS9/A423 statement [13] and was prospectively registered on the International Prospective Register of Systematic Reviews (PROSPERO) database (ID number: CRD42021251186) [14] . The review was reported in line with the AMSTAR-2 (A MeaSurement Tool to Assess systematic Reviews) checklist [15] , Supplemental Digital Content 2, http://links.lww.com/JS9/A424, achieving a moderate quality score.

Eligibility criteria
The participants for which the learning curve was mapped were surgeons at any stage of training with no restriction placed on their prior surgical experience. Studies conducted in a simulated surgical setting were excluded. The interventions included were RALP, laparoscopic radical prostatectomy (LRP), RAPN, laparoscopic nephrectomy (including both conventional laparoscopic and hand-assisted techniques), RARC, laparoscopic radical cystectomy (LRC), robotic adult and paediatric pyeloplasty, laparoscopic adult and paediatric pyeloplasty, robotic retroperitoneal lymph node dissection (RPLND) and laparoscopic RPLND. Studies that reported outcomes for multiple eligible procedures but failed to provide separate data for each procedure were excluded.
Randomised controlled trials, non-randomised interventional studies, cohort studies, case series and case-control studies evaluating the learning curve were included. No language restriction was placed on studies. Review articles, conference abstracts, editorials and letters were excluded.

Search strategy
A systematic literature search strategy was employed across PubMed, EMBASE and the Cochrane Library from inception to December 2021 for eligible studies. Both free-text terms and medical subject headings (MeSH) were used in the database search strategies as listed in the Appendix, Supplemental Digital Content 3, http://links.lww.com/JS9/A425. To find relevant full-text articles in the 'grey literature', the Google Scholar search engine was used alongside website searching and citation chaining.

Study selection and screening
Two independent reviewers performed the initial article title and abstract screening and then screened the full text of the potentially eligible articles against the inclusion criteria. Where discrepancies were encountered, a third reviewer was consulted to resolve the disagreement.

Data extraction and risk of bias assessment
The two independent reviewers used a pre-defined, standardised form to undertake the data extraction. No specific risk of bias assessment tool exists for learning curve studies, so the two reviewers utilised the Newcastle-Ottawa scale [16] as a quality assessment tool. As before, disagreements between the reviewers over the risk of bias scores for the included studies were resolved by referral to a third reviewer.

Data synthesis
It was not possible to perform a meta-analysis of the data collected due to substantial heterogeneity in the included studies, so the results were narratively synthesised in accordance with PRISMA [13] guidelines, Supplemental Digital Content 1, http:// links.lww.com/JS9/A423, with summary diagrams produced to provide the range of values for which the learning curve associated with a particular procedure and its outcomes were numerically defined. This provides surgeons at any stage of training with a guide of expected progress against which they can compare their own performances. Similar descriptive and diagrammatic approaches have been utilised in previous systematic reviews evaluating urological learning curves [11,17] , where meta-analysis was not possible.

Results
Three thousand eight hundred and seventy-eight records were identified through database searching, with an additional 25 records identified through a search of the grey literature and citation chaining. After deduplication, 2912 studies were eligible for the title and abstract screening. Two thousand two hundred and sixty-one studies were then excluded, so 651 studies underwent full-text screening. Five hundred and fifty-four studies were then excluded, with 97 studies thus included for narrative synthesis in this review. Figure 1 illustrates this study's selection and screening process.

Characteristics of included studies
Fifty-one studies were retrospective in design whilst 46 were prospective studies. Fifty-five studies mapped the learning curve of single surgeons, 39 involved multiple surgeons (with one study [18] including as many as 744 surgeons) and 3 did not report the exact number of surgeons involved. The number of procedures ranged from 20 [19] to 27 348 [18] with study durations ranging from 10 months [20] to 259 months [21] . The earliest publication date was 2000 [22] , with the most recent included study published in April 2021 [23] . Table 1 displays the full study characteristics of the included studies.

Risk of bias assessment
Included studies scored either 5 or 6 on the Newcastle-Ottawa scale, scoring lowest in the 'Selection' domain due to the potential for selection bias and the lack of control groups.

Findings
The main findings of the included studies are listed in Table 2.
Where studies defined the learning curve for 'overall performance', this referred to the number of cases required to achieve competency across the range of outcomes they measured.

Chahal et al. International Journal of Surgery (2023)
International Journal of Surgery     [25] was unique in defining the learning curve for potency, demonstrating that surgeon experience correlated with improved sexual function at 5 months (P = 0.007) and 12 months (P = 0.061) up to a plateau phase of 250-300 nerve-sparing RALP cases. Fossatti et al. [35] reported urinary continence recovery increasing from 60% initially to 90% after 400 procedures, with surgeon experience being a significant predictor of continence recovery (P < 0.001).
Samadi et al. [51] evaluated a single surgeon's first 70 RALPs, noting a sustained downward trend in OT (P < 0.0001), length of stay (P = 0.003) and EBL (P < 0.00001). Gumus et al. [38] also observed a continued decrease in OT, with it decreasing from 182 min in the first 40 patients to 168 min in the second 40 patients and then down to 139 min in the third 40 patients.
Bravi et al. [28] found that the risk of PSMs decreased from 15.3% for a surgeon with 10 prior RALPs experience to 6.7% for a surgeon with experience of 250 RALPs. Williams et al. [59] used the CUSUM method to define the PSM learning curve in RALP, setting acceptable and unacceptable positive margin rates at 10 and 15%, respectively. They concluded that around 110 cases are required to overcome the learning curve for pT2 PSMs.
Van der Poel et al. [58] reported an increase in lymph node yield and in the node positivity rate, which significantly increased from 4 to 23.1% from the first 50 cases to their 351st-400th cases. A decrease in Clavien-Dindo grade I and II complications were also noted as surgeon experience increased, but this downward trend was not observed for grade III and IV complications.
The four studies [20,48,49,60] reporting the lowest number of cases to overcome the learning curve notably used carefully selected patients so as to ease the transition for open surgeons to the robotic interface. For example, Pardalidis et al. [48] initially only included patients with a prostate volume less than 50 cm 3 , Gleason score ≤ 7, BMI <30 and with no previous major pelvic surgery.
Mitre et al. [69] noted a significant decrease in intraoperative complications after the first 51 cases (P < 0.05) alongside a significant decrease in the PSM rates from 29.1 to 21.8 to 5.5% for a single surgeon's first, second and third groups of 55 patients, respectively. Vickers et al. [74] reported that surgeons who experienced an open radical prostatectomy achieved significantly poorer results for biochemical recurrence than those naïve to the open procedure (risk difference of 12.3%; 95% CI: 8.8-15.7%). A similar trend was observed with regard to complications by Hruza et al. [67] , who concluded that 700 cases were required for surgeons experienced in the open approach to overcome the Calvert et al.
learning curve compared to 250 cases for surgeons with minimal open surgical experience. Figure 2 summarises the RALP and LRP learning curves.

Robot-assisted partial nephrectomy (RAPN)
Fourteen studies [75][76][77][78][79][80][81][82][83][84][85][86][87][88] analysed the learning curve for RAPN. Mottrie et al. [83] reported a short RAPN learning curve for an experienced robotic surgeon (prior experience of 100 RALPs) with only 20 cases required to achieve a console time of under 100 min. Motoyama et al. [82] reported similar findings with 6 and 4 cases required for a surgeon with prior experience of over 300 RALPs to achieve a console time of under 150 min and a warm ischaemia time (WIT) of under 20 min, respectively. Bajalia et al. [75] evaluated 418 consecutive RAPNs, concluding that OT decreased by 2.5% per 50% increase in the case number (P < 0.001) up to the plateau phase at 77 cases. Larcher et al. [80] adjusted for case mix through the use of multivariable regression analyses, reporting that surgeon experience was significantly associated with decreased WIT (P < 0.0001) and increased probability of a postoperative course without a Clavien-Dindo grade II or higher complication (P = 0.001). Haseebuddin et al. [79] concluded that the RAPN learning curve for WIT is short at just 26 cases for a surgeon with substantial experience in laparoscopic partial nephrectomy.

Laparoscopic partial nephrectomy
Five studies [78,85,90,95,97] evaluated the learning curve for laparoscopic partial nephrectomy. Bhayani et al. [90] evaluated the first 50 laparoscopic partial nephrectomy cases of a single surgeon, noting that only the length of hospital stay significantly decreased from 3.1 days in the first 25 patients down to 2.5 days in the last 25 patients (P = 0.01). Porpiglia et al. [97] assessed the laparoscopic partial nephrectomy learning curve using the 'margin, ischaemia and complication' (MIC) scoring system, reporting that MIC rates increased from 29.4% in the first 51 patients up to 84.9% in the 150th-206th cases.

Other forms of laparoscopic nephrectomy
Three studies [21,91,92] evaluated the learning curve for laparoscopic radical nephrectomy, while two studies [93,95] grouped multiple Figure 2. The number of cases required to overcome the learning curve for RALP and LRP. Values are given as a range, with 'n' denoting the number of studies which numerically defined the learning curve for the outcome. LRP, laparoscopic radical prostatectomy; PLND, pelvic lymph node dissection; PSM, positive surgical margins; RALP, robot-assisted laparoscopic prostatectomy. Where studies defined learning curve for 'overall performance', this referred to the number of cases required to achieve competency across the range of outcomes they measured. techniques together to map an overall learning curve under the umbrella term of 'laparoscopic nephrectomy'. Gill et al. [91] analysed a single surgeon's initial 100 laparoscopic radical nephrectomies, with the only significant decrease between the first 50 and second 50 cases being shorter OT (P = 0.02).
Three studies [19,89,94] assessed the learning curve for handassisted laparoscopic nephrectomy. Azawi et al. [89] reported achievement of 5 min or less WIT after 40 procedures due to the safe and easily learned procedural step of early arterial clamp removal. Figure 3 provides a summary of the RAPN and laparoscopic nephrectomy learning curves.

Laparoscopic radical cystectomy
One study [106] evaluated the learning curve for LRC. Aboumarzouk et al. [106] split a single surgeon's first 65 LRCs into halves, reporting a significant decrease only in OT (303 28 vs. 285 22.93 min, P = 0.002), with a nonsignificant decrease in EBL. The learning curve for LRC and the number of cases required to achieve the plateau phase still remain numerically undefined.
Sorensen et al. [111] concluded that only 15-20 paediatric robotic pyeloplasty cases were required to attain an OT with no significant difference to that of open pyeloplasty (P = 0.23). Cundy et al. [108] identified a classically shaped learning curve for the console time CUSUM chart, but the CUSUM charts for setup and docking times unexpectedly displayed second transition points, which reflected a relocation of the surgical service to a different institution and a change of staff. The learning curve for robotic pyeloplasty is summarised in Figure 4.

Laparoscopic pyeloplasty
Two studies [113,114] evaluated the learning curve for laparoscopic pyeloplasty. Calvert et al. [113] retrospectively evaluated the adult laparoscopic pyeloplasty learning curve over 49 procedures. A nonsignificant decrease in conversion rate to open was observed from 18% in the first third of cases to 6% in the last third of cases (P = 0.53). The authors reported around 30 cases are required to overcome the learning curves associated with complications and conversion rate.
Panek et al. [114] examined the learning curve of paediatric laparoscopic pyeloplasty by splitting a single surgeon's 95 cases into a group of the first 37 and a group of the remaining 58. A statistically nonsignificant decrease in failure rate (rate of surgical reintervention at the ureteropelvic junction) between the two groups was observed from 16.2 to 5.1% (P = 0.147).

Retroperitoneal lymph node dissection
One study [22] assessed the learning curve for laparoscopic RPLND, while another [23] studied the learning curve for robotic RPLND. Janetschek et al. [22] studied 64 laparoscopic RPLND procedures, noting a trend for decreased mean OT from 480 min in the first 14 patients to 222 min in the last 19, thus suggesting the presence of a learning curve. Schermerhorn et al. [23] used linear and logistic regression models to define the learning curve for robotic RPLND. As case number increased, OT and overall complications decreased (P = 0.001 and P = 0.001, respectively), with OT predicted to decrease by 1 h after 44 cases. However, an insufficient number of cases were studied for a plateau phase to be reached. Therefore, the learning curves for both laparoscopic and robotic RPLND remain undefined.

Comparative studies
The four studies [37,54,78,85] examining the learning curves of both laparoscopic and robotic procedures also undertook comparative analyses of these. Good et al. [37] identified similar, lengthy learning curves for RALP and LRP but noted that RALP carries the advantages of comparatively lower PSM rates and improved early continence rates. With regard to PSM and biochemical recurrence rates, Sivaraman et al. [54] identified a shorter learning curve of 100 cases for RALP compared to 350 cases for LRP.
Hanzly et al. [78] reported that RAPN has a shorter learning curve for OT than LPN and more effectively preserves the estimated glomerular filtration rate (eGFR) postoperatively. Furthermore, the authors identified that the learning curve for WIT was reached between cases 29 and 58 for RAPN and between cases 58 and 87 for LPN. Pierororazio et al. [85] similarly reported superior outcomes in OT, WIT and EBL for the RAPN cohort compared to the LPN cohort.

Summary
This systematic review evaluated the existing evidence base for the learning curves of major robotic and laparoscopic urological procedures. For all procedures, the learning curve values varied substantially depending on the outcome measure used to define it, differing learning curve definitions and also on the surgeons' prior surgical experience. Prior surgical experience was not consistently reported and was poorly quantified with general labels of Figure 4. The number of cases required to overcome the learning curve for robot-assisted robotic cystectomy and robotic pyeloplasty. RARC, robot-assisted radical cystectomy. Where studies defined learning curve for 'overall performance', this referred to the number of cases required to achieve competency across the range of outcomes they measured.
'experienced in open and/or laparoscopic surgery' [20,27,34] , providing no precise indication of the number of procedures performed. Multiple studies report that previous laparoscopic experience does reduce the robotic surgery learning curve in the clinical setting, particularly with regard to OT [115,116] , so it is important for authors to report it in order to contextualise the learning curve value.
OT was the metric most commonly used by the included studies to define the procedural learning curves. However, no uniform definition of OT was used, with one study defining it as the time between the carbon dioxide gas going on and off [73,] whereas another described it as the time between 'knife-to-skin' and wound closure [108] . pT3-specific PSM learning curves were defined by several studies [34,37,57] ; this is a much more tumourdependent outcome than surgeon-dependent (whereas the reverse is true for pT2 PSM), so it has limited validity in assessing surgeon performance on the basis of oncological parameters [68] .
Complications and LOS were amongst the patient-outcome variables used to define learning curves, but these also have limitations. LOS is not necessarily representative of the patient's condition leaving the hospital [64] , given it can be affected by patient wishes and even the day of the week that the procedure is performed due to the varied distribution of hospital resources and ancillary support across the week [117] .
Complications also are not always completely reflective of the surgeon's performance as they correlate with the quality of complication reporting which thus introduces potential bias to the results [118] . Urinary incontinence is one of the most important patient-outcome variables for RALP and LRP, given its nature as the most disruptive side-effect post-prostatectomy [119] , but its learning curve was only defined by 6 of the 53 included prostatectomy studies.
Learning curve metrics are also subject to the effects of other confounders. In the context of urological robotic surgery, the format of training undergone by the surgeon influences the learning process and hence the rate at which safe surgical outcomes are achieved [120] . The skills and expertise of the whole operative team have also been reported to affect surgical outcomes in urological procedures. Using an expert bedside assistant has been shown to decrease EBL and PSM early in the learning curve for RALP [121] , with the benefits of a skilled bedside assistant including their ability to handle potential emergencies and thereby decrease the risk of complications, while the console surgeon is seated unscrubbed away from the patient [81] . Gumus et al. [38] proposed a possible learning curve for anaesthetists on the basis that transfusion rates were disproportionately high relative to the EBL for the initial 80 RALP cases and as the decision to transfuse was made by an anaesthetist naïve to robotics and laparoscopy, their inexperience explained this discrepancy. Furthermore, the presence of mentors and the extent of mentorship received by trainees were poorly reported despite it being demonstrated to affect urological learning curves [122] .
The split-group method of analysing the learning curve lacks the sensitivity to define the exact number of cases for which learning curve transitions occur [5] , yet it was the most commonly employed method. Regression techniques were also used by studies, but these may also obscure the identification of key learning curve characteristics such as rate and plateau, given the forced match of the best-fit lines on the data collected [8] .
The conventional view of the surgical learning curve having just one ascent and one plateau phase was challenged in this review by reports of a multiphasic learning curve arising from the tendency of surgeons to take on increasingly complex cases as their confidence improved with experience [94,108] . The case mix effect is well-documented in the literature as a key factor in shaping the learning curve [3,123] , particularly in the post-plateau phase [124] .

Strengths and limitations
All of the included studies were observational in design, which can introduce confounding and selection bias to results [125] , but measurement of learning curves necessitates such a design as they are based on the observation of changes in variables as surgeon experience increases. Fifty-five studies were single-surgeon in design which limits the generalisability and external validity of their reported learning curves given the vast interpersonal variation between surgeons in terms of technical attributes and prior experience [126] . The lack of adjusting for confounders and variation in outcome measures across the included studies in this review is consistent with the findings of reviews assessing the heterogeneity of the surgical learning curve literature bases [127,128] . Another limitation at the study level was the lack of cost-effectiveness analyses to investigate the association between the length of a surgeon's learning curve and the economic impact on their institution in terms of training costs.
At review-level, the exclusion of conference abstracts has been identified as a source of publication bias [129] , but the justification for doing so is that the data presented in such abstracts frequently involves preliminary results which do not necessarily provide an accurate representation of the eventual findings upon study completion [130] . Another limitation is that no sensitivity analysis was performed to investigate the effect of including studies at high risk of bias on the conclusions formed.
No date restriction was imposed on studies, so the results of older studies may be confounded by the procedural 'discovery' curve in which the surgeon is standardising novel techniques as opposed to learning a standard procedure [12] . However, date restriction would not have been an effective way of eliminating bias from technical changes, given that individual institutions appear to undergo their own procedural discovery curves according to their own local experiences of procedures and outcomes, irrespective of study date. One such example is Sharma et al.'s [53] RALP learning curve study in which visual port placement was altered from the paraumbilical space to the umbilical space in order to reduce an initially high rate of port-site hernias, thereby decreasing complications independent of increasing surgeon experience.
Key strengths of this review include its comprehensive search strategy of multiple databases and the grey literature in order to elicit eligible full-text articles and adherence to PRISMA guidelines), Supplemental Digital Content 1, http://links.lww.com/JS9/ A423. This review updates the urological learning curve literature base and is the first to examine the literature pertaining to the learning curves of laparoscopic nephrectomy, LRC, laparoscopic pyeloplasty and both robotic and laparoscopic RPLND.

Implications for research and clinical practice
As is consistent with the findings of previous reviews [5,11,12,17,126] , standardised reporting of outcomes and performance measures is needed in order to reduce heterogeneity and thereby enable a meta-analysis to be performed to combine learning curve values across studies. Given that surgeons' background expertise affects their learning curve [131] , surgeons' baseline characteristics must be reported so as to categorise learning curves by prior experience.
There are ethical concerns raised over surgeons learning on real patients given that more experienced surgeons often attain better outcomes [28] , so increasing surgical experience through simulation-based training is a safer method for reducing the learning curve associated with procedures [132] . Mentorship is another educational intervention that has been demonstrated to reduce the surgical learning curve [133] , with the 'Leipzig Model' of supervision in LRP being one such example [134] . In this example, the trainee performs all the steps they are competent at with a mentor acting as a first assistant, then the student becomes the first assistant and observes for the remaining steps to enhance the learning process. Indeed, surgical outcomes in RALP have been shown to improve if the console surgeon has first gained experience as a bedside assistant because this role improves troubleshooting ability and confidence in dealing with more challenging cases [135] .
Although underused in the included studies, CUSUM analysis is a well-described method for not only precisely plotting an individual's learning curve [8] but also for auditing purposes in the continuous monitoring of a trainee's competency as has been demonstrated in the context of cataract surgery [136] and gynaecology [137] . CUSUM charts enable assessment of a trainee's performance relative to target values and can alert trainers to outof-control processes, at which point trainees can be instructed to either stop and undergo further education or conduct their next procedures under observation to ensure that patient safety is not compromised while the learning process restabilises [46] .
The results of the comparative studies [37,54,78,85] support the view that the robot-assisted approaches to radical prostatectomy and partial nephrectomy have a shorter learning curve than the conventional laparoscopic approaches. Across all outcomes, the laparoscopic modality was not found to confer any advantage over the robot-assisted approach with regard to the learning curve. Thus RALP would be recommended over LRP on the basis of accelerated attainment of competency for the outcomes of PSMs, biochemical recurrence and early continence. In a similar fashion, RAPN would be recommended over LPN owing to its shorter learning curves for OT, WIT and EBL.
Finally, the learning curves for LRC and robotic and laparoscopic RPLND remain undefined, so future studies should evaluate hundreds of consecutive cases to enable the identification of the plateau phase for these procedures and assess the learning curves of multiple surgeons to increase the external validity of their findings.

Conclusion
This systematic review outlines the range of values for the learning curves of major robotic and laparoscopic urological procedures. As has been described in previous reviews, there was substantial variation in the definitions of outcome measures and performance thresholds, with poor reporting of confounders such as prior surgical experience. Although the majority of studies used the split-group method of analysing the learning curve, the CUSUM method is recommended for more precise characterisation of the key learning curve phases as well as for monitoring the competency of surgeons over time as a form of an appraisal. The use of simulation training, mentorship, and gaining experience as a bedside assistant is recommended in order to reduce the learning curve prior to taking full control as the lead surgeon. Lastly, the results of the comparative studies demonstrate that RALP and RAPN have shorter learning curves for a range of metrics compared to their laparoscopic counterparts.

Ethical approval
Ethical approval was not required.

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

Author contribution
B.C.: planned the review, performed the search strategies, extracted the data and wrote the manuscript; M.S.A.A.: acted as the second reviewer and contributed to writing the manuscript; A. A.: acted as the third reviewer, aided in planning the review, and edited and reviewed the manuscript; A.K., M.S.K., K.A. and P.D.: contributed to writing the manuscript and providing revisions.

Conflicts of interest disclosure
Baldev Chahal, Abdullatif Aydin, Mohammad S.A. Amin, Azhar Khan, Muhammad S. Khan and Kamran Ahmed have no conflicts of interest or financial ties to disclose. Prokar Dasgupta declares financial ties as Chief Medical Officer for Proximie Ltd. and Chief Scientific Officer for MysteryVibe Ltd. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
Research registration unique identifying number (UIN)

Data availability statement
Search results and extracted data are available on request.

Provenance and peer review
Not commissioned, externally peer-reviewed.