Research Letter
May 22, 2024
Learning Curve for Robotic-Assisted Cholecystectomy
Kyle H. Sheetz, Jyothi R. Thumma, Stanley Kalata, et al
JAMA Surg. Published online May 22, 2024. doi:10.1001/jamasurg.2024.1221
During the past decade, robotic-assisted cholecystectomy use has increased more than 30-fold across the United States.1 Surgeons gravitated toward this approach for numerous reasons, including promise that the new technology will make the operation safer and to refine their robotic surgical skills with a familiar operation. Recent evidence suggests bile duct injuries, an uncommon yet morbid technical complication, are more common with robotic-assisted than traditional laparoscopic cholecystectomy.2 Critics suggest that bile duct injury rates may only be higher for surgeons still progressing along their learning and therefore may not accurately reflect bile duct injury rates in real-world practice. We evaluated the learning curve for robotic-assisted cholecystectomy among a cohort of US surgeons from 2010 through 2019, when use of this approach moved out of specialty centers and was rapidly adopted into general practice across the country.
Methods
We used 100% nationwide Medicare claims to identify 637 765 patients undergoing cholecystectomy from 2010 through 2019. We identified 4443 individual surgeons by unique National Provider Identifier in Part B claims. Outcomes, data definitions, coding for each approach, and analytic strategy have been previously described.2 We assumed that because robotic-assisted cholecystectomy was uncommon prior to 2010, surgeons’ first cases in the Medicare claims approximate their true first operation. Using logistic regression controlling for patient and hospital characteristics, we estimated bile duct injury rates for surgeons’ first to nth cases sequentially. Case volumes in the Medicare population were inflated to reflect true volumes by dividing hospitals’ total number of Medicare discharges by total admissions from the American Hospital Association survey. We used the adjusted national mean bile duct injury rate for laparoscopic cholecystectomy as a benchmark for the robotic-assisted learning curve since evidence suggests surgeons overcome their learning curve for this operation by the end of formal training.3 We also performed sensitivity analyses restricted to high-volume surgeons and using linear splines to account for different slopes across the learning curve (eAppendix in Supplement 1). Two-tailed P < .05 was considered statistically significant.
This study was deemed exempt by the institutional review board of the University of Michigan. This study used deidentified secondary data for which informed consent was not applicable.
Results
Patient characteristics were similar between those undergoing laparoscopic vs robotic-assisted cholecystectomy (Table). While more patients undergoing laparoscopic cholecystectomy had acute cholecystitis (88.0% vs 77.9%; P < .001), the opposite was observed for symptomatic cholelithiasis (6.8% vs 11.6%; P < .001). Most surgeons (3867 [87.0%]) performed fewer than 10 robotic-assisted cholecystectomies (Figure, A). Bile duct injury rates following robotic-assisted cholecystectomy decreased with increasing experience (Figure, B). To reach equivalent bile duct injury rates with traditional laparoscopic cholecystectomy, surgeons would need to perform between 300 and 450 robotic-assisted cholecystectomies.
Discussion
This study reflecting a cohort of US surgeons suggests that while increasing experience is associated with a decrease in bile duct injury rates following robotic-assisted cholecystectomy, most surgeons do not perform enough to reach equivalence with traditional laparoscopic cholecystectomy. This study may be limited by the assumption that surgeons’ first cases in Medicare claims reflect true first case(s); however, this would bias the results toward shorter learning curves as these early cases would already reflect some experience with the technology. While unmeasured confounding may also limit the conclusions (eg, extent of trainee involvement, prior cholecystostomy tube), similar patient characteristics (higher-severity diagnoses among robotic-assisted cholecystectomy, eg, cholecystitis) and similar rates of overall complications do not suggest that the robotic-assisted cohort has higher risk overall.
These data further call to question 2 contemporary practices in surgery across the United States: overall growth in robotic-assisted cholecystectomy and use of this procedure as a familiar learning case for building robotic skills or achieving local credentialing standards. Proponents of robotic-assisted cholecystectomy contend that surgeons beyond their learning curve provide safe and effective care. Even though surgical training has shifted toward greater exposure to robotics, the prolonged learning curve demonstrated in this study suggests it is worth questioning whether further uptake of robotic-assisted cholecystectomy justifies the added morbidity to reach it.
