Accuracy of Robot-Assisted Pedicle Screw Placement: Two-Center Experience with Learning Curve Analysis

Background: Accurate pedicle screw placement remains essential in spinal instrumentation, and robotic navigation has been introduced to improve safety, reproducibility, and workflow standardization. This study evaluated the accuracy of robot-assisted pedicle screw placement using the Excelsius GPS platform during the first year of implementation at two centers and analyzed the associated learning curve. Methods: Consecutive patients undergoing robot-assisted spinal instrumentation between April 2024 and April 2025 were retrospectively reviewed. Screw accuracy was assessed on intraoperative three-dimensional imaging using the Gertzbein-Robbins Scale (GRS). Grades A and B were considered clinically acceptable, whereas grades C-E were considered clinically non-acceptable. Sacral S1 screws and oncological cases requiring carbon fiber-reinforced PEEK instrumentation were excluded from the primary analysis and evaluated separately when appropriate. Robotic workflow time was defined as the interval between the first intraoperative three-dimensional acquisition used for planning and the second acquisition used for screw verification. Results: The primary standard non-oncological cohort included 102 patients and 455 non-S1 screws. Overall, 411 screws were classified as GRS A, yielding a perfect intrapedicular placement rate of 90.3%. Clinically acceptable accuracy was achieved in 449 of 455 screws, corresponding to a GRS A + B rate of 98.7% (95% CI, 97.2-99.4%). Only six screws were classified as GRS C-E, with no GRS D screws observed. Clinically acceptable accuracy was comparable between centers. In Center 1, all clinically non-acceptable screws occurred within the first nine cases, and GRS A + B accuracy increased from 92.9% in the first trimester to 100% thereafter. Median robotic workflow time was 64.4 min per case and 13.9 min per screw. Conclusions: This two-center early experience supports the accuracy and reproducibility of ExcelsiusGPS-assisted spinal instrumentation. Chronological analysis showed that clinically non-acceptable breaches were concentrated in the early implementation phase; however, this observation should be considered exploratory because of the low event count. The study supports high clinically acceptable accuracy and broadly comparable robotic workflow metrics across centers. Chronological patterns observed during early implementation should be interpreted as exploratory rather than as proof of a formal learning curve.