Decoding surgical proficiency and complexity: a machine learning framework for robotic herniorrhaphy

TL;DR

This paper introduces a machine learning framework to assess surgical complexity and skill development in robotic hernia repair using performance indicators.

Contribution

A novel machine learning approach is proposed to decode surgical proficiency and complexity using objective performance indicators in robotic herniorrhaphy.

Findings

Gradient boosting models combining clinical and OPI data achieved an F1 score of 0.87 for predicting case complexity.

OPIs alone had a lower predictive accuracy (F1 score of 0.58) for case complexity.

Skill acquisition is reflected in the stabilization of OPI variability within 10 months despite increasing case complexity.

Abstract

To evaluate the predictive value of objective performance indicators (OPIs) for case complexity assessment and explore their role in quantifying skill acquisition during robotic ventral herniorrhaphy.

Despite advances in herniorrhaphy techniques, unclear metrics of case complexity have significant implications for operative planning, resource allocation, and patient outcomes. While existing complexity definitions rely primarily on clinical factors external to operator behavior, the expanding adoption of robotic platforms in ventral hernia repair provides unprecedented access to quantifiable surgical performance metrics. However, the relationship between these objective performance indicators and both case complexity and skill development remains incompletely understood, representing a gap that machine learning approaches may help address.

OPI and clinical data from 561 consecutive robotic ventral hernia repairs over eight years were analyzed using iterative ensemble machine learning models to predict case complexity. Dimensional reduction analyses using t-distributed stochastic neighbor embedding tracked skill evolution, with Euclidean distances calculated between successive cases to quantify skill acquisition over time.

Gradient boosting models integrating clinical and OPI variables achieved F1 score of 0.87, while OPIs alone scored 0.58. Longitudinal analysis revealed high OPI variability during early cases, stabilizing within 10 months despite increasing case complexity, indicating skill acquisition may compensate for procedural difficulty. Dimensional reduction analyses captured this evolution through weighted Euclidean distances.

Objective performance indicators poorly predict case complexity independently, yet their temporal evolution reveals surgical skill acquisition. The concurrent stabilization of OPI stochasticity and progression to more complex cases demonstrates that surgical proficiency and complexity assessment are interdependent phenomena, establishing digital metrics as tools for understanding the dynamic relationship between surgeon learning and case difficulty.

The online version contains supplementary material available at 10.1007/s00464-025-12412-x.

Figures