Limitations of procedural learning curves: reassessment of the clinical relevance of learning curve analysis in robotic right hemicolectomy

Learning curves in robotic right hemicolectomy (RRH) are assessed using cumulative sum (CUSUM) analysis. CUSUM was originally designed for industrial processes using external benchmarks. Medical applications rely on internal benchmarks, typically the cohort-specific median. This introduces circularity, obscures early deviations and distorts inflection points. In addition, phase definitions are based on procedural metrics with limited linkage to clinical outcomes. Current models may not reflect surgical proficiency. Data were drawn from a prospectively maintained institutional database covering RRH procedures performed between 2010 and 2025; after exclusions, 229 cases remained for analysis. Cases were split into a training group (TG; n = 164) and a control group (CG; n = 65), the latter comprising procedures by two surgeons already in their mastery phase and used as an external benchmark (EB). Learning-curve analysis of RRH referenced against this EB formed the primary endpoint, while RA-CUSUM analysis and perioperative outcome parameters were assessed as secondary endpoints. Application of the external benchmark approach revealed a prolonged initial learning phase (approximate transition zone 35–55 cases). Major complications, conversion rate, blood loss, and length of stay showed the steepest changes initially, with stabilization thereafter, while lymph node yield reached an early plateau. Operative time demonstrated a continuous increase throughout. Compared with previously reported analyses, learning phases were extended and showed different patterns when clinical outcomes were considered. The prolonged first learning phase for RRH indicates delayed perioperative proficiency. Learning curve analysis should shift to a clinically driven approach with external benchmark methodology. Benchmark values are institution-specific; the methodology may serve as a conceptual framework for future analyses.