Microsurgical Training and Skill Acquisition: A Systematic Review and Meta-analysis

Abstract

Background:

Simulation in the context of microsurgical training has been explored and examined in the literature with a difference of opinion regarding the optimal model to use for the training and acquisition of microsurgical skills. In some reports, the fidelity of the live animal models offers the opportunity learn surgical as well as surgical-scenario contextualized skills that a less realistic model is unable to offer. In other publications, the skills that can be taught on lower fidelity models are equivalent and offer a more viable and feasible approach to surgical training.

Methods:

A systematic review of the available microsurgical simulation models was performed with a focus on the acquisition of microsurgical skills measured with validated microsurgical instruments. A meta-analysis of the subgroups was then performed to identify the optimal microsurgical training tool.

Results:

All training with simulation models resulted in large and statistically significant improvements in microsurgical skills across the range of groupings of microsurgical simulations. This effect was consistently identified across the different groups.

Conclusions:

Choosing a model to train microsurgical skills may be less important than initially thought. Across the range of living and nonliving biological models and synthetic models, the potential to improve microsurgical skills significantly is available regardless of which simulation model is chosen.

Takeaways.

Question: Which simulation model when used for microsurgical training resulted in the most improvement in microsurgical skill acquisition?

Findings: Using a systematic review and meta-analysis, the authors demonstrate that training with microsurgical simulation models of varying degrees of fidelity resulted in uniformly large improvements in microsurgical skills across the variety of models assessed.

Meaning: The choice of microsurgical training model may not be significant, making it viable to choose more available and affordable options to practice and improve microsurgical skills.

INTRODUCTION

The Halstedian maxim “see one, do one, teach one” has long characterized surgical training. However, contemporary education has moved toward a new paradigm of “simulation before escalation.” In experimental biology, the principle of the 3Rs has frequently been applied to guide the humane use of test subjects and has also guided the use of living models in simulation. Competency-based microsurgical education was further advanced by Kobayashi et al in 2015, who introduced the 3Cs framework as an alternative to the traditional “learn by doing” approach.

• Curriculum: carefully structured with a range of training models.

• Competence: learner-centered progression; individuals advance at their own pace and achieve defined objectives.

• Clinical performance: the translation of acquired technical skills into effective clinical practice.¹

Grober et al assessed the robustness of laboratory-based technical competence by comparing outcomes from simulation with bench models versus didactic instruction alone. Their results demonstrated that participants trained with bench models achieved substantially higher retention checklist scores, superior global rating scores, and improved patency of anastomoses when compared with those who participated in didactic training only.²

Lascar et al highlighted the significance of the learning curve within experimental surgical training. Using anastomotic patency as the primary endpoint measured, they compared novice trainees with an advanced group and demonstrated that, after 52 anastomoses, no significant differences in patency rates remained between the groups. Analysis of 2 intermediate groups revealed that increasing prior experience was linked with a progressively less steep learning curve.³ A separate study examined only 2 cohorts—novice and advanced trainees. The investigators found that after 25 anastomoses, patency rates between the groups converged, with no statistically significant differences observed thereafter.⁴

Two independent studies have demonstrated that ordered microvascular training significantly augments clinical outcomes among surgical trainees.^5,6 Both of these studies showed participation and application in microvascular simulation and training facilitates the acquisition of operative independence, mitigates cognitive workload, and augments technical proficiency.^5,6

A previous systematic review evaluated the utility of synthetic models in microsurgery training. They concluded that the availability of synthetic models proved important for the development of microsurgical skills before the application of surgical skills to in vivo models.⁷

The above shows that training in microsurgical techniques translates into improved skills and surgical proficiency beyond physical measurable skills but also in decision-making and effort required.

Ramachandran et al conducted a comprehensive review of validated microsurgical assessment tools designed to evaluate microsurgical performance. Their analysis encompassed numerous versions and modifications of global rating scales, developed under various names and acronyms by multiple authors. In their review, the authors concluded that no single “gold standard” currently exists for the objective assessment of microsurgical skills. They further expanded on the qualities and characteristics of what would comprise an ideal assessment instrument. The author’s recommendation was that it should reliably detect surgical competence across multiple domains. To be considered robust, a validated assessment tool must demonstrate feasibility, reliability, and validity.⁸

The optimal model for microsurgical training has long been a subject of debate. Gasteratos and colleagues concluded that the rat model continues to represent the gold standard for microsurgical education. This conclusion was attributed to the model’s capacity to facilitate the acquisition of both technical and nontechnical competencies, including leadership, situational awareness, decision-making, and teamwork.⁹ The fidelity of the live rat model then in theory provides the opportunity to develop these skills in the way that other models may not.

In a separate experimental investigation, the rat model was found to demonstrate a high degree of tolerance to technical errors, with clinically unacceptable mistakes nevertheless resulting in patent anastomoses. The authors cautioned that this inherent tolerance should be considered when using the rat model for educational purposes. They recommended that training should focus on suture quality and operative precision rather than vessel patency alone.¹⁰

Given the conflicting evidence regarding the most effective model for microsurgical training, an important question remains unanswered. To address this, a systematic review was conducted.

The rationale behind the study is to try to compare similar validated models in groups to attempt to assess which model, when its effect is measured with validated tools, produces the most improvement in microsurgical skills.

METHODS

Study Design

This study was designed as a systematic review conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.¹¹

The objective was to systematically review and compare different models used for simulation and teaching of microsurgical skills, with a specific focus on evaluating the ability of these models to improve microsurgical performance using validated assessment tools.

Search Strategy

A search across multiple electronic databases was performed to attain the literature which was used for this study, including Academic Search Ultimate, Africa-Wide Information, Applied Science and Technology Source Ultimate, ScienceDirect, MEDLINE, and Embase. The search encompassed all available studies with no lower date limit, up to June 2025. Free-text keywords in combination and Medical Subject Headings terms were used. Additionally, reference lists of articles were also screened, when applicable eligible studies where included as part of a “snowball search” (Table 1).

Table 1.

Medical Subject Headings Terms

1	microsurgeon* or “plastic surgeon*” or surgeon* or registrar* or residen*
2 AND	“micro surg*” or micro-surg* or microsurg* or anastomo* or micro-anastomo* or microanastomo* or “micro anastomo*” or “microvascular anastomos*” or “microvascular sutur*” or microsutur* or “free flap*” or “microvascular sutur*”
3 AND	“bench model*” or “high fidelit*” or simulation* or “low fidelit*” or animal* or rat or rats or mice or mouse or murine or porcine or chicken* or poultry or gelatin* or pva or “Polyvinyl acetate” or silicone or hydrogel* or noodle* or petal* or cadaver*
4 AND	train* or educat* or “skill* acquisition” or “clinical* competen*” or “clinical skill*” or “task perform*” or “microsurgical skill*” or validat* or effectiv* or compare or comparison* or improve* or improving

Eligibility Criteria

The following are the inclusion criteria:

1. Primary research studies evaluating microsurgical training models.

2. The simulation training model was clearly described.

3. Studies provided sufficient data to analyze the association between the training model used and improvement in microsurgical skills.

4. Microsurgical skill acquisition was measured using a validated assessment instrument.

The following are the exclusion criteria:

1. Not written in or translated into English.

2. Conference abstracts.

3. Studies that used simulators but did not use validated instruments to assess skill improvement.

Validity of assessment instruments was determined by whether there was published evidence to support their ability to measure differences in surgical skill and whether they had been shown to reliably measure the difference between novice and experienced surgeons.

Study Selection

Covidence (Cochrane-approved) was used to perform the systematic review (available at https://www.covidence.org).

Abstracts and titles retrieved from the searches of the databases were screened independently by 2 reviewers (A.E.E. and T.E.P.) according to the eligibility criteria. Full texts were subsequently reviewed against the inclusion and exclusion criteria.

In cases were disagreements occurred, these were mediated and resolved through discussion with a third reviewer (C.P.G.N.) (Fig. 1).

Fig. 1.

The study selection process PRISMA flow diagram.

Data Extraction

Data were extracted from Covidence for each included study. The following variables were collected.

• Author and year of publication.

• Country and study design.

• Participant characteristics.

• Type of microsurgical model.

• Duration, structure of training, assessment methods.

• Outcomes; numerical scores from assessment.

• Key results and conclusions.

When data were ambiguous or incomplete, corresponding authors were contacted via e-mail for clarification and additional data or raw data. Studies for which sufficient information could not be obtained were excluded.

Risk-of-bias Assessment

The methodological quality and risk of bias of across the included studies was assessed primarily using the ROBINS-I V2 tool for nonrandomized studies. Two studies were evaluated using the revised Cochrane risk-of-bias tool for randomized trials (RoB 2) (Figs. 2, 3).

Fig. 2.

ROBINS-I V2 tool for risk of bias nonrandomized studies.

Fig. 3.

Risk of bias using ROB 2 for randomized trials. ROB, risk of bias.

Data Synthesis

Where the mean and/or the SD were not available, they were estimated as follows:

• Mean and SD were estimated from the available median, minimum and maximum, or from the available median, first and third quartiles, using the methods of Wan et al.¹²

• SD was estimated from the available mean and P values for the test of significant difference between baseline and endline scores, using the method of Higgins et al.¹³

• Where only mean and SD of component (not overall) scores were available, the mean and SD of the overall score were estimated by adding the means of the component scores and applying the mean relative SD of the component scores.

• For the Soto-Miranda and Ver Halen study,¹⁴ the SD was calculated from the raw data presented in the graphs demonstrated in the publication.

For each study, the difference in mean score between baseline and endline was converted to a standardized mean difference using the Hedges g value, adjusted for small sample sizes (Fig. 4), where M1 − M2 is the difference in means and SD* pooled is the pooled and weighted SD (Fig. 5), where n represents the sample sizes and SD represents the standard deviations for the baseline and endline groups.

Fig. 4.

Calculation for the Hedges g value.

Fig. 5.

Calculation for pooled SD.

where n are the sample sizes, and SD is the standard deviation for the baseline and endline groups. Random-effects meta-analyses on standardised mean differences were conducted for each group of studies. The DerSimonian and Laird (inverse variance) model was used to calculate the pooled effect estimate with a 95% confidence interval (CI) across studies. The following heterogeneity statistics were reported:

• Tau-squared (τ²): This is the most direct measure of heterogeneity in a random-effects model. It estimates the variance of the true treatment effects across different studies. A larger τ²i ndicates more heterogeneity.

• I-squared (I²): This statistic represents the percentage of total variation across studies that is due to heterogeneity, rather than random sampling error. An I² of 0% indicates no heterogeneity. An I² of 75% suggests that 75% of the variation is due to true differences between studies. It should be interpreted with caution, especially in meta-analyses with few studies, as it can be biased.

• Q statistic and P value: The Q statistic is a test to determine if the observed effect sizes are more variable than expected by chance alone. A small P value (eg, P < 0.10) suggests there is statistically significant heterogeneity, rejecting the null hypothesis of homogeneity. However, Q is influenced by the number of studies, so its P value can lack statistical power in meta-analyses with few studies.

It is important to consider both statistical heterogeneity (the variability of results beyond chance) and clinical heterogeneity (differences in populations, interventions, and outcomes). Data analysis was carried out in Stata (Reference: StataCorp. 2017. Stata Statistical Software: Release 15, StataCorp LLC, College Station, TX).

Ethical Considerations

All data were taken from information that had previously been published. The University of the Free State Health Sciences Research and Ethics Committee was consulted, a study protocol was provided, and ethical approval was obtained: UFS-HSD2023/1404/2507-0003.

RESULTS

Study Selection and Characteristics

A total of 29 studies (some with 2 subsets of data) fulfilled the eligibility criteria and were included in the quantitative analysis. Data were extracted for baseline and endline performance scores, including participant numbers, means, and SDs. In cases where descriptive statistics were incomplete, missing means or SDs were estimated according to established methods.^12,13 Standardized mean differences (Hedges g) were calculated for each study to allow comparison across diverse scoring systems. Random-effects models were applied for all meta-analyses.

The studies were categorized into five meta-analytic groups (G1–G5) based on the training model type. Effect sizes represented the magnitude of improvement in microsurgical skill from baseline to endline following training intervention.

Grouping

The grouping used for this meta-analysis was based on previous publications where models were divided into their characteristics (Table 2).

Table 2.

Grouping of Models Used for Training

Groups
G1	Synthetic
G2	Biological: Plant
G3	Biological: Animal living
G4	Biological: Animal nonliving
G5	Combined

In a systematic review of simulation and training models in microsurgery they categorized the training models into 6 groups:

1. Bench.

2. Cadaveric animal tissue.

3. Cadaveric human tissue.

4. Live animal.

5. Virtual reality.

6. Training curricula.¹⁵

In another publication, Evgeniou et al categorized models into:

1. Synthetic.

2. Cadaveric animal.

3. Cadaveric human.

4. Live animal.¹⁶

Kania et al grouped models under 3 headings:

1. Prosthetic.

2. Animal.

3. Human cadveric.¹⁷

Penn et al divided microsurgical simulation models first into physical and virtual. The physical models were subdivided into:

1. Nonanimal.

2. Nonliving, animal.

3. Nonliving, human.¹⁸

The grouping is largely similar to that used by Penn; the nonanimal models were subdivided into synthetic and biological, plants-based.

A fifth category of a combination of training models was included to be representative of many studies that used a combination of models in their training.

Group G1

The pooled analysis for G1 demonstrated a very large and statistically significant improvement in microsurgical performance following training (Hedges g = 2.0; 95% CI: 1.0–3.0; P < 0.001). There was some evidence of heterogeneity, driven primarily by a lower-effect study (Fann 2008). However, this study was retained to provide a conservative pooled estimate (Fig. 6).

Fig. 6.

Forest plot for group 1.

Interpretation

Training interventions in G1 were consistently effective, producing substantial improvement across studies.

Group G2

The G2 meta-analysis similarly demonstrated a very large and statistically significant pooled effect size (Hedges g = 2.0; 95% CI: 1.4–2.7; P < 0.001). Minimal heterogeneity was observed, indicating high consistency across included studies (Fig. 7).

Fig. 7.

Forest plot for group 2.

Interpretation

The training models within G2 were methodologically similar and produced uniformly strong improvements in microsurgical skill.

Group G3

Only a single study was available within G3, preventing meta-analysis. Although the individual effect size could be described narratively, the group provided insufficient data for pooled estimation (Fig. 8).

Fig. 8.

Forest plot for group 3.

Group G4

The initial analysis revealed substantial heterogeneity, with the Hong et al¹⁹ study showing an unusually large effect compared to all other studies. As the study was only accessible through its abstract, data verification was limited, and the study was excluded for sensitivity analysis

Sensitivity Analyses

• Excluding Hong et al¹⁹: Heterogeneity remained elevated, primarily due to the Esanu et al²⁰ study.

• Excluding both Hong et al¹⁹ and Esanu el al²⁰: The recalculated model demonstrated a large and statistically significant pooled effect (Hedges g = 1.2; 95% CI: 0.9–1.5; P < 0.001) with substantially reduced heterogeneity (Fig. 9).

Fig. 9.

Forest plot for group 4.

Interpretation

The results for G4 were robust, with large improvements observed even after removal of influential studies. Heterogeneity was explained by study outliers rather than inconsistency in core evidence.

Group G5

G5 exhibited substantial heterogeneity, driven by several studies with exceptionally large effect sizes. Mattar et al²¹ demonstrated problematic reporting, including undefined variability measures and ambiguous error bars. Other high-effect studies^22–24 appeared valid but contributed disproportionately to variability.

Sensitivity Analysis

After exclusion of these high-effect studies:

• The pooled effect remained very large and statistically significant (Hedges g = 1.5; 95% CI: 1.1–1.9; P < 0.001).

• Some heterogeneity persisted, largely skewed toward a low-effect study,²⁵ which was retained for conservative interpretation (Fig. 10).

Fig. 10.

Forest plot for group 5.

Interpretation

Even after the removal of extreme values, G5 demonstrated strong, clinically meaningful improvement in microsurgical performance.

Overall Meta-analytic Findings

Across groups G1, G2, G4, and G5, the pooled effect sizes consistently reflected large to very large improvements in microsurgical skill after training, with effect sizes ranging from a Hedges g value of 1.2 to 2.0. All pooled analyses were statistically significant at a P value less than 0.001.

Although heterogeneity was present in several groups, it was explainable through identifiable outlier studies. Sensitivity analyses confirmed that the observed effects were stable, robust, and not dependent on anomalous datasets.

Summary Interpretation

• All training models evaluated led to substantial improvements in microsurgical skill performance.

• Effect sizes were uniformly large, indicating strong educational impact across diverse training modalities.

• Heterogeneity was anticipated due to variation in training models, assessment tools, and participant backgrounds but did not weaken overall conclusions.

• Sensitivity analyses strengthened confidence in the findings, demonstrating that results remained significant even after removal of influential studies.

• The effect sizes, with their 95% CIs, are summarized in the plot below. We concluded that we should ignore the result for G3 because there was only one study in this group. The CIs for the other 4 groups overlap, so there is not one group with a better or poorer performance than other groups. All groups had an effect size significantly greater than 0, and all were in the “large” range, showing that all the techniques represented by the groups were very successful in improving surgical technique.

DISCUSSION

Evidence from published studies has indicated that the use of simulation-assisted surgical training can significantly reduce the number of live rats used without adversely affecting the quality of microsurgical skill acquisition. Despite this finding, it has been recommended that live rat models continue to serve as the final stage of microsurgical training before operative experience.²⁶

Lahiri et al further demonstrated, in a comparative study involving two groups of trainees, that additional time dedicated to ex vivo and synthetic model training improved the acquisition and development of technical surgical skills. This improvement translated into enhanced performance and more effective utilization of rat models. Nonetheless, the authors maintained that live rats remain an essential component of comprehensive microsurgical training and should not be entirely excluded from educational programs.²⁷

Similarly, Grober et al, in studies published in Annals of Surgery, compared skill acquisition between silicone tubing models and live rat vas deferens models. Their findings revealed that low-fidelity bench simulations were equally as effective as high-fidelity live animal training, suggesting that the educational value of simpler simulation models should not be understated.²

In a randomized comparative study utilizing the Objective Structured Assessment of Technical Skills framework, nonliving and living animal models for microvascular surgical training were evaluated. Trainees who began their training on nonliving models demonstrated better scores compared to those who began with living models. On this basis, the authors proposed that validated nonliving simulation platforms represent suitable alternatives to live animal models for microsurgical education.²⁸

CONCLUSIONS

Some weaknesses and limitations were identified in this study. There were many studies which did not publish enough data for inclusion. Some authors were not available to answer requests for additional information. In some publications, available information was used to calculate data used for the analysis.

An additional consideration which may have had an impact on the results of this study is the lack of consensus with regards to the ideal validated microsurgical skill assessment tool, and as a result, the measurement of surgical skill acquisition was measured in different studies by different tools. The length of time spent training was also not evaluated or compared as a factor in the improvement of microsurgical skills. Future studies may benefit from inclusion of wider model definitions including amnion/placenta, some of which were included in the review with the current set of Mesh terms but may not be exhaustive. Group 3 was notably small, reflecting the already generally accepted practice of using models of low fidelity to introduce microsurgical skills rather than to start training with live models.

The current evidence is insufficient to prove that any model or combination thereof is superior (Fig. 11). What has been clearly demonstrated is that training with any simulation model has a significant positive impact on the acquisition of skills. Even the simpler and lower fidelity models facilitate enhancement of microsurgical skills. As the literature review revealed, these skills are translated into intraoperative benefits when applied to clinical practice.

Fig. 11.

Effect size of each group of models as calculated by the meta-analysis.

Multiple models are available and have been shown to be effective in microsurgical training. The choice of training model can be determined by training levels, resources, and other institutional factors.

DISCLOSURE

The authors have no financial interest to declare in relation to the content of this article.