Long-term desktop-microscope training for STA–MCA bypass: A three-year learning-curve analysis

Background Superficial temporal artery-middle cerebral artery bypass demands high-level microsurgical suturing, yet opportunities for repetitive clinical anastomosis practice are limited. Although simulation-based training is widely used, the magnitude, plateau, and durability of long-term training effects remain unclear. Methods We retrospectively reviewed a single trainee's desktop-microscope, gauze-based anastomosis training performed intermittently during neurosurgical residency from August 2013 to March 2016 (Phase 1) and compared these data with the identical protocol restarted in January 2026 (Phase 2) after 10-year interval without simulation-based training. For each session, stitch counts and total session time were recorded and converted to time per stitch, which served as the primary objective performance metric. Learning dynamics were modeled within each phase using a power-law function y = ax -b fitted by nonlinear least-squares regression. Results In Phase 1, 3678 stitches were completed over 32 months; time per stitch improved from 10 min initially to approximately 1 min by study end (mean session duration, 24.56 min). The Phase 1 learning curve fit well (pseudo- R 2 = 0.78) and was described by y = 14.20 × -0.38 , with ~80% of the total modeled improvement occurring within the first ~200 cumulative stitches. In Phase 2, 198 stitches were completed over six sessions, with time per stitch improving from 1.17 to 0.94 min; the fitted model was y = 1.69 × -0.099 (pseudo- R 2 = 0.62), indicating a flatter learning profile and faster initial performance than early Phase 1. Conclusion Desktop-microscope gauze practice showed rapid early gains followed by gradual improvement. Performance remained near the late Phase 1 level after a 10-year interval, although intervening clinical bypass experience may have contributed. This study provides long-term quantitative data on microsurgical skill acquisition and reacquisition using a simple training model.