Sensitivity of Insurance Claims Codes in Identifying Robotic Assisted Surgery

To determine the sensitivity of insurance claims codes in identifying robotic-assisted surgery (RAS), assess bias from misclassification, and evaluate the generalizability of findings across data sources.

Insurer-generated databases are widely used to study RAS outcomes, but inconsistent use of claims codes may lead to misclassification and biased estimates.

This retrospective cohort study compared a test definition (claims only) to a reference definition (claims plus free-text hospital billing data) for identifying RAS from 2018-2023. Two U.S. datasets were used: the Premier Healthcare Database (PHD), a large hospital discharge database, and Merative™, a major claims database for insured employees and dependents. Seven procedures-inguinal hernia repair, cholecystectomy, sleeve gastrectomy, Roux-en-Y gastric bypass, lobectomy, right colectomy, and hysterectomy-were evaluated in inpatient and outpatient settings. Misclassification bias was assessed for operative time, length of stay, conversion to open surgery, and surgical site infection. Generalizability was examined by comparing RAS rates across datasets.

Among 2,978,390 procedures in PHD, the sensitivity of claims-only identification was 0.578. Sensitivity exceeded 0.8 for all inpatient procedures across years but was very low for outpatient procedures, falling below 0.5 by 2021. For procedures commonly performed outpatient, effect estimates based on the claims-only definition were frequently biased. RAS rates using the test definition in PHD were generally higher than those observed in the claims-only Merative™ dataset.

Sensitivity of claims data to identify RAS varies by procedure, setting, and time. Low sensitivity causes substantial misclassification bias, impacting analyses of surgical modality and outcomes.