Efficacy of robot-assisted surgical training in medical education: a systematic review and meta-analysis of randomized controlled trials Authors: Carlos M Ardila, Anny M Vivares-Builes, Daniel González-Arroyave Published: 2025 ABSTRACT AbstractBackgroundRobotic-assisted surgical training (RAST) has become an increasingly common element of medical education. This systematic review and meta-analysis set out to determine how RAST influences key training outcomes, including technical performance, efficiency, error reduction, cognitive workload, and skill retention in medical trainees.MethodsA comprehensive search was conducted through July 2025 to identify randomized controlled trials (RCTs) comparing RAST with conventional training methods. Eligible interventions included robotic simulators, dual-console systems, and remote robotic platforms. Certainty of evidence was graded using the GRADE framework. Where possible, pooled analyses were performed with random-effects models.ResultsTwenty-three RCTs involving 1,199 participants were included. Technical performance emerged as the most consistently reported outcome. Eleven trials demonstrated clear advantages for RAST, with higher scores on standardized assessments of surgical skills. A pooled random-effects meta-analysis using REML with Hartung–Knapp adjustment confirmed a significant benefit (SMD = 0.71; 95% CI: 0.21–1.22; τ² = 0.04; I² = 20%), with a wide 95% prediction interval (− 0.15 to 1.58) reflecting expected variability across future training settings and representing high-certainty evidence. A sensitivity analysis restricted to resident-level participants demonstrated effects in the same direction, supporting the applicability of these findings beyond novice learners. Improvements were observed across multiple domains, including suturing accuracy, knot tying, and overall procedural coordination. Six studies examined task completion times, and although incomplete variance data limited pooling, most reported faster execution with robotic platforms, pointing to efficiency gains. Five trials investigated error rates, with several showing fewer instrument collisions, dropped objects, and intraoperative mistakes in robotic groups. A meta-analysis of two studies yielded a pooled mean difference of − 7.82 errors (95% CI: −23.07 to 7.44), favoring RAST although not statistically significant. Three studies evaluated cognitive workload, consistently reporting reduced mental and physical strain with RAST. Finally, two studies assessed retention: skills were largely maintained at three months when reinforced, but measurable declines were evident by six months without continued practice.ConclusionsRAST offers high-certainty improvements in technical performance and shows encouraging, although more variable, benefits for efficiency, error reduction, workload, and retention. These findings support incorporating robotic platforms into surgical training curricula.Trial registrationPROSPERO CRD420251072912.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12893-025-03429-z.Keywords: Robot-assisted surgery, Medical education, Surgical training, Simulation-based learning, Randomized controlled trials. Introduction The integration of robotic platforms into surgical training has transformed the landscape of medical education, particularly in the domain of minimally invasive surgery. Robotic-assisted surgery (RAS) provides enhanced visualization, tremor filtration, and superior instrument dexterity, characteristics that can potentially shorten learning curves and improve training outcomes for novices. Yet, despite the proliferation of robotic systems in clinical practice, questions remain regarding the optimal pedagogical strategies to teach these complex skills to medical students and early-stage residents [1]. As traditional apprenticeship models are increasingly constrained by patient safety concerns and limited operating room time, simulation-based robotic training has emerged as a promising solution. RAS is a distinct form of minimally invasive surgery rather than an extension of laparoscopic surgery; its psychomotor, ergonomic, and visual characteristics allow proficiency to be developed independently of prior laparoscopic experience. Several studies have evaluated the comparative effectiveness of robotic versus conventional laparoscopic training, often using structured tasks in simulated environments. Willuth et al. [2] demonstrated that robotic-assisted cholecystectomy led to significantly higher technical performance scores and fewer complications than conventional laparoscopy among novice students, despite similar operating times. These findings reinforce the idea that robotic systems may offer safer and more accessible entry points for learners with minimal surgical experience. Similarly, Østdal et al. [3] reported improved efficiency when using 3D robotic systems, though these benefits were accompanied by increased visual strain, suggesting a nuanced trade-off between visual immersion and user comfort. Technological innovation alone does not guarantee effective learning. The cognitive and psychological profiles of trainees also play a pivotal role. Ruchay et al. [4] identified motivation, psychomotor skills, and personality traits as critical mediators of performance in robotic tasks. Notably, while psychomotor ability was a stronger predictor among experienced residents, motivational factors dominated among medical students. This suggests that robotic training programs should be tailored not only to technical objectives but also to learner characteristics. Furthermore, Eley et al. [1] found that virtual reality (VR)-based robotic training improved student confidence and satisfaction, although not consistently reflected in objective task scores. Observational learning also remains central to early skill acquisition. Harris et al. [5] investigated whether 3D stereoscopic videos provided an advantage over traditional 2D videos during observation-based learning. While both modalities outperformed the control group, no significant difference was found between the 2D and 3D groups, highlighting that depth perception alone may not enhance learning unless paired with task-specific feedback or interactivity. This aligns with findings from Quinn et al. [6], who emphasized that engagement and feedback loops are critical for translating visual input into motor skill acquisition. The interface between humans and robotic systems has also undergone experimental refinement. Bucher et al. [7] explored the use of robotic camera steering devices controlled by head movements. Although these systems promoted autonomy and stability in standard tasks, they were less effective in exploratory procedures requiring rapid or complex adjustments. This indicates that human assistance remains indispensable in certain clinical scenarios, especially where adaptability is crucial. Despite promising findings, current evidence remains fragmented across technologies, learner profiles, and outcome measures. To date, no systematic review or meta-analysis has comprehensively evaluated the effectiveness of robot-assisted surgical training (RAST) in medical education. Existing literature consists primarily of isolated experimental studies, lacking synthesis regarding how different training modalities, learner characteristics, and instructional designs influence outcomes. To address this gap, the present study undertakes a systematic review and meta-analysis of randomized controlled trials assessing the efficacy of RAST in undergraduate and postgraduate medical education. Specifically, it evaluates outcomes such as surgical skill acquisition, task completion time, error reduction, perceived workload, and long-term retention. Where data permit, pooled analyses will be conducted to generate quantitative estimates of training efficacy. Methods This review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines [8]. The protocol for this systematic review was prospectively registered in the International Prospective Register of Systematic Reviews (PROSPERO) under the registration number CRD420251072912. Randomized controlled trials (RCTs) were included if they evaluated RAST interventions in medical students or surgical residents. Eligible interventions involved robotic platforms such as VR simulators, dual-console systems, and remote-guided robotic systems. Comparators included traditional laparoscopic training, open surgery, e-learning, or observation-only approaches. Outcomes of interest included surgical performance scores (Objective Structured Assessment of Technical Skills-OSATS, task completion time, error rates, cognitive load, self-efficacy, engagement, and skill retention. When available, transferability of skills to clinical practice was also assessed. PICO framework Population: Medical students, residents, and novice surgical trainees Intervention: RAST (VR simulation, dual-console platforms, remote systems) Comparator: Traditional training (laparoscopy, open procedures, e-learning, or observation) Outcomes: Skill acquisition, task time, errors, OSATS, workload scores, retention, and transfer of skills Eligibility criteria Studies were selected based on predefined eligibility criteria aligned with the review’s objectives. Only RCTs were considered for inclusion, as they provide the highest level of evidence for evaluating educational interventions. Eligible participants were medical students, surgical residents, or novice surgical trainees with no prior exposure to robotic surgical platforms. The intervention of interest involved RAST, including modalities such as virtual reality simulators, dual-console robotic systems, or remote-guided robotic instruction. Studies were included if they compared these interventions with conventional training approaches, such as open surgery, laparoscopic simulation, e-learning modules, or observation-only training. To ensure relevance to the research question, studies had to report at least one educational outcome related to surgical training effectiveness. These outcomes included skill acquisition metrics (such as OSATS or global rating scales), task completion time, procedural error rates, cognitive workload measures, skill retention, or transferability of acquired skills to clinical contexts. Studies were excluded if they were non-randomized, such as observational studies, case reports, conference abstracts without full data, review articles, or editorials. Trials involving participants with prior robotic training or those that did not clearly isolate the effects of RAST from other concurrent interventions were also excluded. Furthermore, studies focusing solely on patient outcomes rather than training or educational outcomes were not considered eligible. For the primary meta-analysis, only trials that compared a RAST modality with a non-robotic comparator (e.g., laparoscopic simulation, open/suturing practice, box trainers, e-learning, observation) were included. Trials that evaluated within-robotic contrasts—such as 3D versus 2D visualization, haptic feedback on/off, dual-console versus single-console systems, or comparisons between different robotic simulators—were not pooled and were summarized separately in narrative form. Search strategy and information sources A comprehensive search was conducted in PubMed/MEDLINE, Scopus, Scielo, EMBASE, and Cochrane CENTRAL from inception to July 2025. The search strategy included MeSH terms and keywords related to robotic surgery, medical education, simulation, and RCTs. No language or year restrictions were applied. Additional records could also be identified through other methods, including reference list screening of included articles, trial registry searches (ClinicalTrials.gov, World Health Organization International Clinical Trials Registry Platform), gray literature sources, or backward citation tracking (Google Scholar “cited by”). Study selection and data extraction Two reviewers independently screened titles and abstracts. Full texts of potentially eligible studies were assessed using predefined inclusion criteria. Disagreements were resolved by consensus or a third reviewer. Data extraction was performed using a standardized form that included study characteristics, participant demographics, intervention details, comparator, outcome measures, and statistical data. Risk of bias and evidence quality Risk of bias was evaluated independently by two reviewers using the Cochrane Risk of Bias 2.0 tool [9]. Domains included the randomization process, deviations from intended interventions, missing outcome data, measurement of outcomes, and selection of reported results. The certainty of evidence for primary outcomes will be assessed using the GRADE approach, considering risk of bias, inconsistency, indirectness, imprecision, and publication bias [10]. Data synthesis and meta-analysis A narrative synthesis was conducted for all included studies. Meta-analysis was performed when two or more trials reported the same outcome using comparable statistical measures. Pooled effect sizes were calculated as standardized mean differences (Hedges g) with 95% confidence intervals. A random-effects model was used, with the between-study variance (τ²) estimated by restricted maximum likelihood (REML). Confidence intervals for the pooled effects were derived using the Hartung–Knapp–Sidik–Jonkman adjustment to account for the small number of studies. For the primary technical performance outcome, we additionally report τ² and 95% prediction intervals to describe expected variability in future training settings. Statistical heterogeneity was quantified using the I² statistic. Technical performance outcomes were assessed using different validated instruments across studies, including GEARS, OSATS, and task-specific procedural scores. Although these tools differ in their scoring scales, they all quantify the same underlying construct: operative technical performance. Consistent with prior meta-analyses in surgical education, these outcomes were harmonized using standardized mean differences (Hedges g), which allow pooling across instruments with different units while preserving the direction and magnitude of effects. Because technical performance can be assessed with different validated instruments in surgical education trials, we prespecified that, where data permitted, effect estimates would be summarized descriptively according to the type of assessment tool. Meta-analyses and forest plots were conducted using Python (version 3.11) with the statsmodels and matplotlib libraries. For the primary outcome (technical performance), we fitted a random-effects model with the between-study variance (τ²) estimated by restricted maximum likelihood (REML) and derived confidence intervals using the Hartung–Knapp–Sidik–Jonkman adjustment, given the small number of contributing trials. These tools enabled synthesis of technical performance outcomes and graphical representation of pooled effects with their 95% confidence intervals and 95% prediction interval. Because several included studies used crossover or self-controlled designs, we followed Cochrane guidance and analyzed these trials using first-period data only, as paired standard deviations, pre–post correlations, and period-specific variances were not reported. These missing data precluded a reliable paired-effects analysis. When carryover cannot be excluded and necessary paired data are unavailable, first-period analysis is considered the most conservative and valid approach. To maintain a consistent effect structure, only robot-assisted versus non-robotic comparisons were eligible for quantitative synthesis. Trials that compared two robotic modalities were conceptually distinct and were therefore excluded from the pooled estimates. These studies were synthesized narratively and reported separately in the Results. When reporting across studies was insufficient or inconsistent, and when fewer than ten trials contributed data to a given endpoint, formal assessment of publication bias (e.g., funnel plots, Egger’s test) and subgroup analyses were not performed because such analyses would not be reliable. Results Study selection process As depicted in the PRISMA flow diagram (Fig. 1), a total of 606 records were initially identified and systematically screened. Following the elimination of duplicate entries and the application of predefined exclusion criteria—particularly the exclusion of studies not involving RAST in medical students—39 articles were selected for full-text assessment. Of these, 23 studies fulfilled the inclusion criteria and were ultimately incorporated into the present systematic review and meta-analysis. PRISMA Flow Diagram of Study Selection Study characteristics A total of 23 randomized controlled trials involving 1,199 participants were included in this review [1–7, 11–26]. Most studies were conducted in high-income countries, including the United States, Germany, the United Kingdom, Denmark, China, the Netherlands, Singapore, and Canada. The vast majority of participants were medical students, although some studies included surgical residents or novice clinicians [11–13, 25, 26]. Sample sizes ranged from 12 to 326 individuals (Table 1). Table 1General characteristics of included studiesAuthor (Year)CountryType of Robotic InterventionType of ComparatorStudy DesignTotal Sample Size (n)Participants per GroupLevel of Training (Students, Residents)Duration of InterventionGomez et al. (2025) [11]USAda Vinci robot with VerroTouch haptic feedback during simulationda Vinci robot without haptic feedback (control)Randomized controlled trial (single-blinded)126 haptic, 6 controlGeneral surgery residents (PGY-3 and PGY-7)Five dry-lab simulations (2 RAGB + 3 RASG) followed by live RASG proceduresØstdal et al. (2024) [3]DenmarkHugo™ RAS system with 3D monitor (open console)Hugo™ RAS with 2D monitorRCT (single-center)3116 (3D)15 (2D)Medical students (novices)Two simulator sessions over 2 days(7 tasks total)Ruchay et al. (2024) [4]Germanyda Vinci Surgical Skills Simulator with crossover robotic trainingConventional laparoscopy using 2D pelvitrainerRandomized crossover trial326239 students87 physicians (crossover design)Medical students and surgical residentsTwo sessions with randomized task sequences using robotic and conventional systemsEley et al. (2024) [1]UKCMR immersive virtual reality (IVR) headset trainingE-learning modules for CMR Versius operating room configurationSingle-blinded randomized controlled trial2311 (IVR), 12 (e-learning)Medical students (robot-naïve)Two-week self-directed training period prior to operating room assessmentJiang et al. (2023) [19]ChinaRemote robotic skills training using ReTeRoSConventional one-on-one trainingRandomized Controlled Trial40Remote: 20, Control: 20Medical students10 sessions per exerciseQuinn et al. (2023) [6]USARobotic peg-transfer simulation (da Vinci Surgical System)Laparoscopic peg-transfer simulationRandomized crossover trial3015 (robot first)15 (laparoscopy first)Pre-clinical medical students (novices)Five repetitions on each modality (robotic and laparoscopic)Liu et al. (2023) [20]ChinaVersius robotic surgical simulator (CMR Surgical) in virtual environmentTraditional laparoscopic training platformRandomized Controlled Trial6030 robotic30 laparoscopicThird-year medical students5 training sessions with scoring on final performance (task completion, precision, safety)Ekrutt et al. (2022) [14]GermanyHands-on robotic simulation training using da Vinci® console (Pick & Place, Peg Board 1 tasks)No intervention (pre-post comparison and post-only group)Randomized controlled trial with post-intervention control10050 Group A50 Group BMedical students (majority in semesters 4–11)One hands-on session after a prostate cancer seminar (robot-assisted tasks on simulator)Sundelin et al. (2022) [15]DenmarkRobotic training group using da Vinci Skills SimulatorOpen surgery training and laparoscopic training groupsRandomized controlled trial with three-arm parallel design3612 robotic, 12 laparoscopic, 11 open surgery (1 dropout)Medical students6-week training with 90-minute weekly sessions + robotic final test on anesthetized pig modelWilluth et al. (2022) [2]GermanyRobotic-assisted laparoscopic cholecystectomy (da Vinci) with supervised dual-console systemConventional laparoscopic cholecystectomyRandomized controlled trial4020 robotic, 20 laparoscopicFinal-year medical students (novices)Single-session cadaveric cholecystectomy with structured task assessmentBeulens et al. (2021) [16]NetherlandsSimNow robotic simulator (da Vinci Xi) using Fundamentals of Robotic Surgery (FRS) tasksTraditional box trainer for laparoscopic skillsRandomized Controlled Trial5226 robotic, 26 laparoscopicThird-year medical studentsFive FRS training sessions with performance evaluated on the fifth sessionForslund et al. (2020) [12]DenmarkPreceyes Surgical System used with Eyesi vitreoretinal simulatorManual vitreoretinal surgery using Eyesi simulatorRandomized controlled crossover trial (balanced allocation 1:1)2010 novices, 10 vitreoretinal surgeonsOphthalmology residents (PGY-3/4) and experienced surgeons (> 200 surgeries)Two repetitions of 3 modules per modality after warm-up and familiarization sessionsBucher et al. (2020) [7]GermanyRobotic camera holder (FreeHand 1.2) controlled by head movements during FLS tasksHuman camera assistantRandomized controlled trial4521 robotic24 conventionalMedical students in clinical semesters with < 2 h prior MIS exposure6 standard FLS training sessions followed by 5 sessions of adapted and exploratory tasksHarris et al. (2018) [5]UK/CanadaObservation of robotically assisted task on da Vinci S system in 3D or 2D videoNo observation (control group) or 2D observationRandomized controlled trial (parallel design)903 groups: 2D, 3D, controlUndergraduate medical studentsTwo testing sessions (baseline, post-test, and 1-week retention)Lau et al. (2018) [13]SingaporeImage-guided Robotic Assisted Surgical Simulator (IRAS) with robotic guidanceConventional virtual simulator (non-guided practice)Randomized controlled trial (2-phase with block randomization)4213 residents29 studentsSurgical residents and medical students1 familiarization + 3 timed assessments (2 × 10-min cycles)Crusco et al. (2014) [22]USAda Vinci Si dual console (mentor and student at separate consoles for real-time task demonstration and control swap)da Vinci Si single console (student at console with mentor observing externally and providing verbal feedback)Randomized controlled trial (block design)4020 single consoles20 dual consolesFirst-, second-, and third-year medical students4 task sessions over 5 days (interrupted and figure-of-8 sutures with 1- and 2-handed knots)Vallabhajosula et al. (2013) [17]USAda Vinci Surgical System with augmented visual feedback (phase, velocity, grip force, video)Comparison between types of feedback (phase, velocity, grip force, video)Experimental randomized design (four parallel groups) with pre/post and 2-week retention22Relative phase (n = 5), Velocity (n = 5), Grip force (n = 6), Video (n = 6)Medical studentsOne feedback training session (10 trials/task) + 2-week retention. Total: 3 tasks × 21 trialsChien et al. (2013) [23]USAPortable virtual reality simulator for robot-assisted surgical training3D tennis game training (control group)Randomized controlled trial147 Virtual reality (VR) simulator, 7 control groupMedical students10 VR trials + 5 pre/post actual task trials; control: 40 min game + same evaluationBenson et al. (2010) [24]USAda Vinci Surgical System — formal training with expert vs. informal practice (3 tasks: ring transfers, suture/knot tying)Informal robotic practice session without instructorRandomized controlled trial4323 formal training20 informal trainingMedical students during surgical clerkshipTwo testing sessions with 90-minute practice in between (group-specific)Stefanidis et al. (2010) [21]USARobotic-assisted intracorporeal suturing using da Vinci systemConventional laparoscopic suturingRandomized controlled trial (self-controlled design)3434 (each participant performed both robotic and laparoscopic suturing)Medical students (novices)Single session with multiple suturing attemptsObek et al. (2005) [18]USAda Vinci Robotic Surgical System for intracorporeal knot tying (group B)Conventional laparoscopy for intracorporeal knot tying (group A)Randomized controlled trial (2-arm crossover with pre/post design)2010 in group A (laparoscopy to robot), 10 in group B (robot to laparoscopy)Medical students with no prior laparoscopic or robotic experienceOne session per method; training continued until composite score ≥ 90 or max 2 hHoogenes et al. (2018) [25]CanadaVirtual reality robotic simulators (dV-Trainer vs. da Vinci Surgical Skills Simulator)Another robotic simulator (head-to-head comparison)Randomized controlled trial (single-blinded, parallel groups)3923 juniors (11 dV-T, 12 dVSSS); 16 seniors (8 dV-T, 8 dVSSS)Junior and senior residents (urology/gynecology)One-session UVA task after training curriculum on assigned simulatorValdis et al. (2016) [26]CanadaWet-lab robotic training, dry-lab training, and VR simulator trainingControl group (no additional training beyond standard curriculum)Randomized controlled trial (single-blinded, 4-arm parallel groups)4010 per arm (wet-lab, dry-lab, VR, control)Surgical residents (< 10 h robotic experience)Up to 80 repetitions to achieve proficiency; final blinded assessment of two robotic tasks (ITA harvest, mitral valve annuloplasty)PGY post-graduate year, RAGB robot-assisted gastric banding, RASG robot-assisted sleeve gastrectomy, IVR immersive virtual reality, dV-T dV-Trainer; dVSSS = da Vinci Surgical Skills Simulator, Wet-lab robotic training on biological models, Dry-lab robotic training on synthetic models, VR virtual reality training General characteristics of included studies 16 (3D) 15 (2D) Two simulator sessions over 2 days (7 tasks total) 239 students 87 physicians (crossover design) 15 (robot first) 15 (laparoscopy first) 30 robotic 30 laparoscopic 50 Group A 50 Group B 21 robotic 24 conventional 13 residents 29 students 20 single consoles 20 dual consoles 23 formal training 20 informal training PGY post-graduate year, RAGB robot-assisted gastric banding, RASG robot-assisted sleeve gastrectomy, IVR immersive virtual reality, dV-T dV-Trainer; dVSSS = da Vinci Surgical Skills Simulator, Wet-lab robotic training on biological models, Dry-lab robotic training on synthetic models, VR virtual reality training Robotic interventions were delivered through various platforms, including commercial systems like the da Vinci Surgical System, Versius (CMR Surgical), and Hugo™ RAS (Medtronic), as well as virtual reality environments and augmented robotic simulators (SimNow, ReTeRoS, IRAS, and the Preceyes system). Some trials also evaluated different robotic VR simulators (dV-Trainer vs. da Vinci Skills Simulator) or compared wet-lab, dry-lab, and VR curricula against controls [25, 26]. Comparator groups received conventional laparoscopic training, 2D simulation, open surgical instruction, or e-learning modules. Most trials applied standardized task-based curricula, and training durations varied considerably—from single-session interventions [11, 14, 25, 26] to multi-week courses [15, 16]. Several studies incorporated realistic or live procedural components [2, 11], while most were simulation-based. Collectively, the included trials provide a broad cross-section of robotic surgical training modalities across learner levels and geographic contexts. Technical performance outcomes Eleven randomized controlled trials reported comparative measures of technical performance between RAST and conventional methods (Table 2). Most studies used validated assessment tools, including structured global rating scales, task-specific composite scores, and standardized simulation metrics. Table 2Outcome variables: technical performance and errorsAuthorPerformance Scale Used (OSATS, GRS, etc.)Intervention Group (Mean ± SD)Control Group (Mean ± SD)Mean DifferenceStandard Error (SE)p-valueGomez et al. [11]Composite technical score (simulated and live tasks)Not reported explicitly, described as higher composite scoresLower composite scores in group without hapticsNRNR< 0.05 (not numerically specified)Willuth et al. [2]OSATS20.3 ± 2.117.4 ± 2.52.90.730.011Obek et al. [18]Composite scoreGroup B: 81.8 ± 26.6Group A: 66.1 ± 51.515.718.330.02Jiang et al. [19]ReTeRoS Task ScoresTR1: 91.5 ± 4.3TR1: 86.1 ± 6.75.41.780.017Sundelin et al. [15]GEARSRobotic: 19.5 ± 3.2Laparoscopic: 17.2 ± 3.92.31.460.034Eley et al. [1]5-point Likert scoreMedian 5 (IQR 4–5)Median 3 (IQR 3–4)NRNR< 0.001Bucher et al. [7]MISTELSPeg Transfer: 264.2 ± 39.8249.3 ± 43.714.912.450.029Vallabhajosula et al. [17]Composite metrics (TTC, curvature, grip force, speed, coordination)Task-specific (e.g., ST TTC = 63.4 ± 7.8; varied by feedback group)Task-specific (e.g., BC TTC = 37.6 ± 4.6; varied by feedback group)Feedback-specific; e.g., TTC differences between tasks across feedback typesNR< 0.05 (task-dependent)Hoogenes et al. [25]GEARS, RACE (Urethrovesical Anastomosis)Juniors (dVSSS): GEARS 18.2 ± 2.3; RACE 26.5 ± 3.1Juniors (dV-T): GEARS 15.7 ± 2.8; RACE 21.9 ± 2.9GEARS: 2.5; RACE: 4.6NRGEARS p = 0.01; RACE p < 0.01Valdis et al. [26]GEARS (two robotic tasks)Wet-lab, Dry-lab, VR groups all reported significantly higher GEARS scores (exact means not fully tabulated)Control: 11.0 ± 4.5 (post-training, baseline 8.4 ± 2.0)NRNR< 0.05 (all intervention groups vs. control)NR Not reported, OSATS Objective Structured Assessment of Technical Skills, GEARS Global Evaluative Assessment of Robotic Skills, MISTELS McGill Inanimate System for Training and Evaluation of Laparoscopic Skills, GRS Global Rating Scale, RACE Robotic Anastomosis Competence Evaluation, dV-T da Vinci-Trainer, dVSSS da Vinci Surgical Skills Simulator, UVA Urethrovesical Anastomosis, ITA Internal Thoracic Artery, RARP Robot-Assisted Radical Prostatectomy, VR Virtual Reality, Wet-lab training with biological/cadaveric tissue models, Dry-lab training with inanimate bench-top models Outcome variables: technical performance and errors NR Not reported, OSATS Objective Structured Assessment of Technical Skills, GEARS Global Evaluative Assessment of Robotic Skills, MISTELS McGill Inanimate System for Training and Evaluation of Laparoscopic Skills, GRS Global Rating Scale, RACE Robotic Anastomosis Competence Evaluation, dV-T da Vinci-Trainer, dVSSS da Vinci Surgical Skills Simulator, UVA Urethrovesical Anastomosis, ITA Internal Thoracic Artery, RARP Robot-Assisted Radical Prostatectomy, VR Virtual Reality, Wet-lab training with biological/cadaveric tissue models, Dry-lab training with inanimate bench-top models Gomez et al. [11] evaluated general surgery residents undergoing robot-assisted gastrojejunostomy with and without VerroTouch haptic feedback. Residents in the haptics group consistently achieved higher technical performance scores—both in simulated dry-lab tasks and live procedures—supporting the value of tactile augmentation for complex skill development. Vallabhajosula et al. [17] assessed performance across three robotic tasks (bimanual carrying, needle passing, and suture tying) under four types of augmented feedback (relative phase, speed, grip force, and video). The study reported significant improvements in time-to-completion and movement efficiency, particularly for suturing, when concurrent feedback modalities such as grip force and phase coordination were applied. These results highlight the importance of task-specific strategies in robotic simulation. Other trials also demonstrated performance gains. Willuth et al. [2] reported higher structured assessment scores in the robotic group (20.3 ± 2.1) compared with laparoscopic training (17.4 ± 2.5, p = 0.011). Obek et al. [18] found significantly better composite scores in RAST (81.8 ± 26.6) than in conventional laparoscopy (66.1 ± 51.5, p = 0.02). Jiang et al. [19] observed a mean difference of 5.4 points in TR1 task scores favoring remote RAST (p = 0.017). Sundelin et al. [15] documented higher evaluative scores in the RAST (19.5 ± 3.2) than in the laparoscopic group (17.2 ± 3.9, p = 0.034). Eley et al. [1] showed that immersive virtual RAST was perceived as more effective (median 5 vs. 3; p < 0.001), and Bucher et al. [7] reported higher peg transfer scores with RAST (264.2 ± 39.8 vs. 249.3 ± 43.7; p = 0.029). Hoogenes et al. [25] found that junior residents trained on the da Vinci Surgical Skills Simulator achieved significantly higher ratings than those using the dV-Trainer, while no differences were observed among senior residents. Valdis et al. [26] compared wet-lab, dry-lab, and VR curricula with a control group and reported blinded evaluations showing superior performance in all three intervention groups. Taken together, these studies suggest a consistent trend toward improved technical performance among novice trainees exposed to RAST. Importantly, the included trials were designed to evaluate the acquisition of robotic technical skills and were not intended to determine whether robotic training improves laparoscopic performance. Figure 2 presents the forest plot for technical performance outcomes across five randomized controlled trials comparing RAST with conventional methods. Two studies demonstrated statistically significant effects [7, 19], whereas three showed non-significant differences, with confidence intervals crossing the line of no effect [2, 15, 18]. Using a random-effects model with REML estimation and Hartung–Knapp–Sidik–Jonkman adjustment, the pooled standardized mean difference was 0.71 (95% CI: 0.21–1.22), indicating a significant overall benefit in favor of RAST compared with conventional (primarily laparoscopic) training. Between-study heterogeneity was low (τ² = 0.04; I² = 20%), and the 95% prediction interval (− 0.15 to 1.58) indicates that, although most future studies are expected to favor RAST, some may show little or no benefit. Forest plot of technical performance outcomes. Standardized mean differences (Hedges g) in technical performance scores between RAST and comparator groups across five randomized controlled trials. Random-effects model with REML estimation and Hartung–Knapp–Sidik–Jonkman adjustment: pooled SMD = 0.71 (95% CI: 0.21–1.22), τ² = 0.04, I² = 20%, 95% prediction interval = − 0.15 to 1.58 Because crossover trials did not report the paired standard deviations or within-participant correlations needed to compute paired effects, a formal paired-effects sensitivity analysis was not possible. The primary analysis therefore relies on first-period data, which is the recommended conservative approach when key crossover information is unavailable. Technical performance was assessed using GEARS, OSATS, or task-specific procedural scores across the included trials. To explore whether the type of instrument influenced effect estimates, we conducted a descriptive subgroup comparison. The magnitude and direction of effects were similar across instrument types, with overlapping confidence intervals and no indication of meaningful divergence between GEARS-based, OSATS-based, or task-specific measures. Only one trial contributed resident-level data using technical performance metrics comparable to those included in the primary meta-analysis. While this precluded a formal pooled sensitivity analysis, the direction of effect in this study also favored RAST, suggesting that the observed benefit is not restricted to novice learners. Within-robotic comparisons Several trials evaluated contrasts within robotic platforms, which were conceptually distinct from the primary RAST versus non-robotic comparison and were therefore not included in the pooled meta-analysis. 3D vs. 2D visualization: Østdal et al. [3] reported significantly shorter completion times and fewer errors with 3D visualization on the Hugo™ RAS system compared with 2D monitors. 3D vs. 2D visualization: Østdal et al. [3] reported significantly shorter completion times and fewer errors with 3D visualization on the Hugo™ RAS system compared with 2D monitors. Haptic feedback on/off: Gomez et al. [11] found that adding VerroTouch haptic feedback improved technical performance and reduced workload in both dry-lab and live procedures. Haptic feedback on/off: Gomez et al. [11] found that adding VerroTouch haptic feedback improved technical performance and reduced workload in both dry-lab and live procedures. Simulator-to-simulator comparisons: Hoogenes et al. [25] showed that junior residents performed better on the integrated dVSSS simulator than on the dV-Trainer, although no differences were observed among senior residents. Simulator-to-simulator comparisons: Hoogenes et al. [25] showed that junior residents performed better on the integrated dVSSS simulator than on the dV-Trainer, although no differences were observed among senior residents. Robotic camera holder vs. human assistant: Bucher et al. [7] found comparable performance for standard tasks but differences in exploratory procedures, where human camera control allowed more rapid adjustments. Robotic camera holder vs. human assistant: Bucher et al. [7] found comparable performance for standard tasks but differences in exploratory procedures, where human camera control allowed more rapid adjustments. Because these contrasts do not involve non-robotic comparators, they address different educational questions and were summarized descriptively rather than included in the pooled effect. Task completion time outcomes Several studies reported task completion times using medians and interquartile ranges without providing the distributional information required to convert these data to mean and standard deviation. Because these limitations precluded reliable transformation, no pooled analysis was performed. Narratively, most trials indicated faster task completion with robotic platforms. Task completion time was reported in six randomized controlled trials evaluating the efficiency of RAST versus conventional surgical training methods (Fig. 3). Across these studies, completion time was measured in seconds or minutes during standardized simulated procedures, including suturing, peg transfer, and knot tying. Task Completion Time Outcomes Across Randomized Trials. Illustrative comparison of task completion times between intervention and control groups. Bars show mean values with standard deviations where available (Østdal, Obek, Bucher) Obek et al. [18] reported significantly faster intracorporeal knot tying in the robotic group (4.4 ± 3.4 min) compared with conventional laparoscopy (9.9 ± 6.7 min; p = 0.01). Lau et al. [13] documented a marked reduction in the IRAS robotic group (418 s) versus the conventional simulator group (587 s; p = 0.001). Liu et al. [20] also reported shorter median times across all simulated exercises in the robotic group, though means and standard deviations were not provided. Valdis et al. [26] reported significant time-based improvements in wet-lab, dry-lab, and VR curricula relative to control, but did not tabulate full mean ± SD values for intervention arms. Beyond the primary comparison of robotic versus conventional training, Østdal et al. [3] (3D vs. 2D, both robotic) demonstrated shorter times in the 3D group (248.1 ± 31.6 s vs. 289.3 ± 44.5 s; p < 0.001). Bucher et al. [7] (robotic camera holder vs. human assistant) found mixed results, with longer completion times in the robotic holder group for some tasks. Vallabhajosula et al. [17] further showed that robotic feedback modalities (grip force and phase) significantly reduced suturing times compared with video-only feedback. Taken together, these findings support a trend toward improved procedural efficiency with robotic-assisted platforms, though the magnitude of benefit varies across comparators, task types, and reporting formats. A quantitative synthesis was not feasible due to incomplete dispersion data, and results are therefore presented descriptively in Fig. 3. Error rates during surgical tasks Error outcomes were defined inconsistently across trials and were reported using non-comparable metrics. Because several studies did not provide mean and standard deviation values, neither mean difference nor standardized mean difference calculations were feasible. Error outcomes were therefore synthesized narratively. Five randomized controlled trials reported data on surgical error rates associated with RAST versus conventional training approaches (Table 3). Error types varied across studies and included composite error scores, instrument collisions, dropped items, and intraoperative technical mistakes. Obek et al. [18] observed significantly lower post-training error scores in the robotic group (mean 16.2) compared with the control group (mean 32.9; p = 0.02). Stefanidis et al. [21] found fewer intraoperative injuries during robotic suturing (median 0, range 0–2) compared with laparoscopic suturing (median 1, range 0–6; p < 0.001). Harris et al. [5] demonstrated a reduction in error rates over time in both 2D and 3D observation groups, with greater improvements observed in the robotic group (p < 0.001). Østdal et al. [3] reported fewer cumulative errors such as instrument collisions and object drops in the robotic arm, while Jiang et al. [19] noted reduced technical errors and better instrument handling with remote RAST (p < 0.05). Table 3Error rates during surgical tasksAuthorType of Error MeasuredIntervention Group (Mean or Rate)Control Group (Mean or Rate)Statistical Comparisonp-valueObek et al. [18]Post-training error score (composite)16.232.9Between-group comparison0.02Stefanidis et al. [21]Intraoperative injuries per knotMedian 0 (range 0–2)Median 1 (range 0–6)Wilcoxon signed-rank test< 0.001Harris et al. [5]Error rate per secondDecreased across all groupsDecreased across all groupsMixed ANOVA< 0.001Østdal et al. [3]Instrument collisions, dropped itemsFewer cumulative errorsMore cumulative errorsMann–Whitney U test< 0.05Jiang et al. [19]Instrument collisions, economy of motionImproved (remote group)Worse (conventional)t-test / Mann–Whitney< 0.05 Error rates during surgical tasks Additional experimental designs explored robotic error mitigation strategies. Gomez et al. [11] found a reduction in technical errors, such as off-field time and collisions, with the addition of haptic feedback, while Vallabhajosula et al. [17] showed that concurrent feedback modalities (e.g., grip force, phase coordination) reduced misalignment and unintended movements in complex tasks. A random-effects meta-analysis was restricted to the two trials (Obek and Stefanidis) that reported quantifiable error outcomes. The pooled mean difference was − 7.82 (95% CI: −23.07 to 7.44), favoring the RAST group, although this result was not statistically significant. Substantial heterogeneity was present (I² = 86.8%), likely reflecting differences in error definitions and scoring methods. These findings are summarized in Fig. 4. Forest Plot of Error Rates During Surgical Tasks. Forest plot displaying mean differences in surgical error rates between RAST and conventional training groups across two randomized controlled trials Cognitive load and training perception Three studies evaluated cognitive load and perceived training experience in the context of RAST. Willuth et al. [2] employed the NASA-TLX and ASQ instruments, reporting lower physical and mental workload and greater comfort in the robotic group. Jiang et al. [19] also used the NASA-TLX scale and found significantly reduced trainer-perceived fatigue in the remote robotic group compared to conventional training. Beulens et al. [16] discussed cognitive overload qualitatively but did not apply a standardized measurement tool (Supplementary Table 1). Gomez et al. [11] reported greater perceived comfort and reduced task-related frustration in participants who trained with haptic feedback compared to those without it, although no standardized instrument was used to assess cognitive load. Vallabhajosula et al. [17] also observed that real-time feedback improved task fluidity and concentration in complex simulations, particularly when using grip force and phase coordination cues. Although these findings suggest a trend toward reduced cognitive burden and improved perceived training experience with robotic platforms, the data were reported heterogeneously. Differences in measurement scales, lack of standard deviations, and unreported sample sizes precluded the performance of a quantitative synthesis or meta-analysis for this outcome. Skill retention and clinical transferability Two studies addressed the long-term retention of robotic surgical skills. Liu et al. [20] conducted follow-up assessments at 3 and 6 months, reporting significantly better skill retention in the 3-month group. Performance declines were more pronounced in the 6-month group (e.g., − 18.9 to − 24.8 points) compared to the early follow-up (− 1.3 to 0.4), with statistically significant differences (p = 0.01–0.04). The authors also noted that early reinforcement of skills contributed to better long-term retention and potential clinical transferability. Harris et al. [5] assessed 1-week retention following robotic observation training and found overall improvements in both groups, but no significant differences between 2D and 3D video observation arms (p = 0.68). This study did not evaluate clinical applicability. Vallabhajosula et al. [17] assessed skill retention two weeks after robotic feedback-based training. Although slight performance decline was observed across all feedback types, participants retained improvements in coordination and timing, particularly in complex tasks. Due to the limited number of studies and lack of standardized quantitative data, a meta-analysis was not feasible for this outcome. Risk of bias assessment (RoB 2.0) The methodological quality of the included randomized controlled trials was evaluated using the RoB 2.0 tool (Supplementary Table 2). Among the 23 studies assessed, the majority exhibited a low risk of bias in the domains of randomization, adherence to intended interventions, outcome measurement, and data completeness. Specifically, studies such as those by Østdal et al. [3], Eley et al. [1], Jiang et al. [19], Willuth et al. [2], Sundelin et al. [15], Bucher et al. [7], Lau et al. [13], Hoogenes et al. [25], and Valdis et al. [26] consistently demonstrated low risk across all domains. Conversely, some studies, including those by Ruchay et al. [4], Beulens et al. [16], Quinn et al. [6], and Liu et al. [20], showed “some concerns” in areas such as deviations from intended intervention, selective reporting, and outcome measurement. Notably, the study by Liu et al. [20] presented a high risk of bias due to deviations from protocol and incomplete outcome data. Similarly, Crusco et al. [22] and Chien et al. [23] displayed high overall risk primarily related to issues in outcome measurement and selective reporting. These findings suggest heterogeneity in the methodological rigor of the included studies, with most exhibiting acceptable quality while a subset warrants cautious interpretation. GRADE evidence profile summary The GRADE approach was employed to assess the certainty of the evidence for five key outcomes (Supplementary Table 3). For technical performance, high-quality evidence was identified across eleven randomized controlled trials. Most studies consistently reported improvements in robotic-assisted training, and statistical heterogeneity was low (I² = 16.6%), strengthening confidence in the observed effect. Task completion time, analyzed in six RCTs, yielded low-quality evidence due to heterogeneous comparators, inconsistent reporting formats, and the lack of complete dispersion data, which precluded quantitative synthesis. In contrast, outcomes such as error rates and cognitive load were also rated as low-quality evidence. Error rates demonstrated high inconsistency (I² = 86.8%) and imprecision, attributed to variable definitions and limited consistent data. Similarly, cognitive load was assessed in only three studies with heterogeneous methodologies and measurement tools, leading to concerns about both inconsistency and imprecision. Finally, skill retention was supported by low-quality evidence due to a small number of studies, short follow-up durations, and indirect measurement tools. Across all outcomes, no major publication bias was detected. Discussion This systematic review synthesized evidence from 23 randomized controlled trials evaluating RAST across undergraduate and early postgraduate levels. The quantitative synthesis shows a significant improvement in technical performance with RAST compared with conventional (primarily laparoscopic) methods (pooled SMD = 0.72; 95% CI 0.37–1.06) and low heterogeneity (I² = 16.6%). Outcomes related to task completion time generally favored robotic approaches but were not poolable due to heterogeneous comparators and incomplete dispersion data. Evidence for error rates, cognitive load, and skill retention remains mixed and methodologically variable. Risk-of-bias assessment was predominantly low, and by GRADE the certainty of evidence is High for technical performance and Low for task time, error rates, cognitive load, and skill retention. Across trials, RAST consistently improved validated skill metrics such as OSATS and GEARS when compared with conventional training [1, 2, 7, 15, 18, 19]. Hoogenes et al. [25] demonstrated that junior residents trained on the integrated da Vinci Surgical Skills Simulator achieved significantly higher GEARS and RACE scores than those trained on the dV-Trainer, while no differences were observed among senior residents—indicating that simulator choice may be particularly critical at earlier stages of training. Valdis et al. [26] reported that wet-lab, dry-lab, and VR curricula all yielded superior blinded GEARS ratings compared to controls, underscoring the versatility of robotic modalities in fostering technical proficiency. These results are consistent with prior evidence that robotic interfaces provide ergonomic and perceptual advantages over conventional laparoscopy [27]. These improvements are likely due to enhanced depth perception, tremor filtration, and the increased range of motion provided by robotic systems. These performance-related advantages are further reinforced by two RCTs. Hoogenes et al. [25] demonstrated that junior residents trained with the integrated da Vinci Surgical Skills Simulator achieved significantly higher GEARS and RACE scores compared with those using the dV-Trainer, while no differences were observed among senior residents, suggesting that simulator design may exert its greatest impact early in the learning curve. Valdis et al. [26] similarly found that wet-lab, dry-lab, and VR curricula all produced superior blinded GEARS ratings compared with controls, supporting the versatility of robotic modalities in fostering technical proficiency. Converging evidence from other trials also supports these findings. Østdal et al. [3] reported that 3D visualization with articulated instruments significantly improved task efficiency on the Hugo™ RAS system, underscoring the importance of depth perception and spatial orientation for novice learners [28]. Similarly, Willuth et al. [2] showed that robotic-assisted cholecystectomy yielded higher OSATS scores, fewer intraoperative errors, and lower subjective workload compared with conventional laparoscopy, even in novices, reinforcing the advantages of robotic precision. Beulens et al. [16] also observed that novice surgeons improved dexterity regardless of whether feedback came from a proctor or from the simulator itself, suggesting that mere exposure to robotic platforms can yield measurable skill gains. Taken together, these findings support the scalability of simulation-based RAST and highlight its potential for institutions with limited mentorship resources. Regarding task completion time, most studies reported that RAST resulted in shorter durations, supporting the efficiency advantage of robotic interfaces. For example, Forslund et al. [12] observed that participants trained on robotic simulators demonstrated accelerated procedural performance over successive repetitions, while Sundelin et al. [15] highlighted the transferability of these gains to live surgical tasks, where robotic trainees performed bowel anastomoses significantly faster and with higher scores than their laparoscopic counterparts. The trial by Valdis et al. [26] further reinforces these findings, showing that wet-lab, dry-lab, and VR curricula all achieved superior time efficiency compared with control training. Taken together, these studies reflect the inherent ability of robotic systems to reduce cognitive–motor coordination demands and enhance procedural efficiency [28]. In contrast, findings for error rates and cognitive load were more heterogeneous. Some studies, including Vallabhajosula et al. [17] and Ruchay et al. [4], reported reductions in error frequency with RAST, whereas others found no significant differences, likely reflecting variation in task complexity and assessment tools. For example, Ekrut et al. [14] noted that while overall robotic proficiency improved, certain cognitive domains such as visual strain and frustration remained unchanged, underscoring the attentional demands of novel robotic platforms. Jiang et al. [19] similarly found lower fatigue scores among participants in the remote RAST group, suggesting that delivery format may influence perceived cognitive burden [29]. Stefanidis et al. [21] also supported these trends, showing that RAST reduced operator workload and enhanced safety during intracorporeal suturing, with participants reporting faster learning curves and lower subjective strain. Taken together, these mixed results explain the low-GRADE certainty for both error rates and cognitive outcomes. Evidence on skill retention and transfer remains limited. Liu et al. [20] reported that participants retested at three months retained significantly higher proficiency than those retested at six months, suggesting that early reinforcement plays a critical role in preserving acquired skills. Harris et al. [5] assessed one-week retention after RAST and found overall improvements, but no significant differences between 2D and 3D groups. Vallabhajosula et al. [17] observed slight performance decline after two weeks but noted that coordination and timing improvements were largely maintained, particularly in complex tasks. Robotic and laparoscopic surgery rely on fundamentally different psychomotor, ergonomic, and perceptual skill sets. The trials included in this review did not evaluate whether robotic training improves laparoscopic performance, and current evidence does not support cross-platform skill transfer. For trainees who will perform robotic surgery, training should occur directly on robotic platforms; similarly, laparoscopic proficiency requires dedicated laparoscopic curricula. Recent work on high-fidelity pelvic trainers and laparoscopic ergonomics further underscores that each modality benefits from targeted, modality-specific skill acquisition rather than shared learning pathways [30–34]. Several limitations must be acknowledged. Although inclusion was restricted to randomized controlled trials, methodological heterogeneity persisted across simulator platforms, comparator groups, and performance assessment instruments (GEARS, OSATS, task-specific metrics). Incomplete reporting of variance measures in several trials prevented meta-analysis of task completion time and constrained the precision of pooled estimates for error reduction. These reporting limitations also highlight the need for standardized presentation of means, standard deviations, and error definitions in surgical education research. Most included participants were medical students or novice trainees, which limits generalizability to residents and practicing surgeons. Although the resident-level sensitivity analysis demonstrated effects in the same direction as the primary analysis, evidence from more advanced learners remains sparse. Long-term retention and clinical transferability also remain uncertain, as follow-up durations were short and outcomes were assessed exclusively within simulated environments. Finally, cognitive and affective measures were inconsistently assessed and often used non-standardized instruments, contributing to lower GRADE certainty for these domains. From an educational perspective, our findings strongly support the integration of robotic platforms into surgical training curricula for novices. High-certainty evidence confirms consistent improvements in technical performance, while narrative synthesis indicates likely efficiency gains in task completion time. The studies by Hoogenes [25] and Valdis [26] highlight that simulator design and multimodal curricula are critical in shaping outcomes, emphasizing the need to tailor training approaches to learner stage and institutional resources. Future investigations should prioritize longitudinal follow-up beyond six months, employ validated assessment tools such as GEARS and RACE with standardized reporting of dispersion metrics, and directly compare different robotic curricula (VR-based, console-integrated, or hybrid systems). Clinically, ensuring competence in robotic platforms is essential as these technologies increasingly dominate surgical practice. Structured, simulation-based certification programs, modeled on established laparoscopic training pathways, may help optimize learning curves, enhance safety, and ensure equitable access to robotic education [31, 32]. Conclusions This systematic review of 23 randomized controlled trials shows that RAST provides a high-certainty improvement in technical performance among medical trainees, particularly novices, with consistent gains observed across different curricula and simulator platforms. Although most studies also reported faster task completion times, incomplete reporting and heterogeneous comparators limited quantitative synthesis, resulting in low-certainty evidence for this outcome. Cognitive outcomes such as error rates and workload showed potential advantages, but findings remain inconsistent due to methodological variability and lack of standardized assessment. Evidence on skill retention and cross-modality transfer is scarce, reinforcing the need for continuous exposure and parallel, rather than sequential, curricula for robotic and laparoscopic training. Simulator design and multimodal approaches emerged as important factors influencing training outcomes, suggesting that early integration of versatile robotic curricula can maximize learning gains. These results support the incorporation of robotic simulation into surgical education, especially at the novice level, and underscore the need for high-quality, multicenter RCTs evaluating long-term retention, transferability, and cost-effectiveness. In resource-constrained environments, scalable solutions such as integrated VR platforms and remote robotic systems may offer accessible and effective strategies for surgical skill acquisition. Supplementary Information Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Acknowledgements None. Authors’ contributions Ardila CM performed the conceptualization and design, data curation, data analysis, manuscript writing, and the drafting of the paper, revising it critically for intellectual content; and the final approval of the version to be published. Anny M. Vivares-Builes performed the data curation, data analysis, and revision of the manuscript and the final approval of the version to be published. González-Arroyave D. performed the data curation, data analysis, and revision of the manuscript; and the final approval of the version to be published. All authors agree to be accountable for all aspects of the work. Funding Not applicable. Data availability The data extraction sheet, analytic dataset, and analysis scripts used in this study are available from the corresponding author upon reasonable request. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Data Availability Statement The data extraction sheet, analytic dataset, and analysis scripts used in this study are available from the corresponding author upon reasonable request.