Multi-TPS institutional surveillance of VMAT plan complexity and PSQA results over three years.

Abstract

BackgroundRunning VMAT plan-complexity surveillance directly inside an oncology information system (OIS) makes it practical to monitor plan quality across a large, multi-linac program. Complexity metrics are increasingly used to anticipate patient-specific QA (PSQA) outcomes, but few studies span multiple years or tie complexity monitoring to the OIS at scale, which leaves published benchmarks hard to generalize and prospective monitoring slow to enter routine practice.PurposeTo describe an automated Python pipeline integrated with the MOSAIQ OIS for computing eight established VMAT plan complexity metrics from DICOM-RT files at scale, and to characterize the resulting distributions across a three-year institutional VMAT program stratified by treatment planning system (TPS), anatomic site, dosimetrist, and calendar year, with correlation to matched PSQA outcomes.MethodsAll clinically approved VMAT plans from March 2023 to March 2026 were identified via MOSAIQ OIS (v2.6, Elekta AB) at a multisite comprehensive cancer center. A purpose-built Python tool queried the MOSAIQ SQL Server database and parsed DICOM-RT plan files to compute MU Factor, mean aperture area (MAA), mean leaf gap (MLG), aperture irregularity (AI), small aperture scores (SASs) at 5 and 10 mm, modulation complexity score for VMAT (MCSv), and Modulation Index Total (MITotal). Plans were stratified by TPS (Pinnacle3 vs. Monaco), anatomic site group (11 groups), dosimetrist, and year. Mann-Whitney U and Kruskal-Wallis tests were applied for group comparisons. PSQA pass rates (3%/2 mm global gamma, ≥95% threshold) were extracted from MOSAIQ and matched to complexity data for 3402 of 3751 plans (90.7%); Spearman rank correlations quantified complexity-pass-rate associations.ResultsThe institutional median (IQR) MU Factor was 2.96 (2.26-4.00) MU/cGy and MCSv was 0.966 (0.955-0.978). Monaco plans showed significantly higher MU Factors (+32%, p ConclusionsOIS-integrated VMAT complexity and PSQA surveillance is feasible at scale and yields clinically useful information. AI and MCSv were the strongest complexity-based predictors of PSQA outcome. The TPS-, site-, and linac-specific complexity profiles we observed argue for risk-stratified PSQA design. The pipeline is readily transferable to other MOSAIQ-based programs for prospective plan-quality monitoring.