S2399
Physics - Quality assurance and auditing
ESTRO 2026
Purpose/Objective: More complex RT plans are subject to larger uncertainties in dose calculation and treatment delivery compared to non-modulated plans [1]. Complexity metrics quantify these uncertainties using treatment plan parameters such as beam apertures, leaf travel, dose rate, and gantry speed. However, departmental procedures and clinical practices may influence the degree of modulation even when similar software and equipment are used. Likewise, the correlation between complexity metrics and patient- specific QA (PSQA) has been shown to be institution- dependent [2]. This study aimed to perform a retrospective evaluation to quantify and compare plan complexity and deliverability across two large multicentre databases. Material/Methods: VMAT plans delivered between 2021 and 2024 at two institutions (Inst.1 and Inst.2) were collected. Nineteen complexity metrics were computed for each VMAT arc [3]. Both centres performed PSQA with EPID using 2%/2 mm criteria and a 10% threshold, but with different software. Gamma analyses were extracted using two custom scripts. VMAT arcs were stratified by treatment site and inter-institutional differences in complexity metrics were assessed using Mann- Whitney tests. Correlations between complexity metrics and gamma passing rate (GPR) were evaluated using Spearman’s correlation coefficients. Bonferroni correction was applied to reduce false discovery rates. Results: A total of 13,903 VMAT arcs were analysed. The characteristics of the two databases are summarised in Table 1, while distributions of representative complexity metrics and treatment sites are shown in Figure 1. Distributions were significantly different between institutions, with Inst.2 producing less complex arcs than Inst.1 (e.g., MeanTGI = 0.20 vs 0.35, p<0.01; EdgeMetric = 0.09 mm-1 vs 0.21 mm-1, p<0.01). Inst.2 used TPS-integrated tools (Aperture Shape Controller) to reduce MLC aperture variability, whereas Inst.1 introduced this approach only recently. Gamma analyses were highly satisfactory, with a median GPR of 99.5% (Q1: 98.5%; Q3: 99.9%) for Inst.1 and 99.9% (99.7%; 100%) for Inst.2. Most metrics showed weak or no correlation with GPR (r < 0.2; 11/19 for Inst.1 and 15/19 for Inst.2), while only EdgeMetric exhibited a moderate correlation (r = − 0.46) for Inst.1.
procedure not followed, lack of standardization)software design (incorrect
configuration file, sub-optimal software design)Clinical practice–related issues should be addressed through multidisciplinary discussions to ensure accurate incorporation of clinical information into the automated script. Software design–related issues should be ranked according to their frequency of occurrence and incorporated into the roadmap for subsequent software releases.
Conclusion: Log mining tools provide rapid and actionable insights into script utilization and adoption. When combined with structured root - cause analysis, this approach enables early detection of shifts in clinical practice and facilitates iterative refinement of the script to better align with clinical requirements. Consequently, the trajectory of script development becomes less reliant on incomplete or inconsistent user reporting. Keywords: log mining, automation, plan checks A retrospective analysis of plan complexity and deliverability of two large multicentre databases Nicola Lambri 1,2 , Sarah Osman 3 , Andrew Nisbet 4 , Catharine H. Clark 3 , Pietro Mancosu 1,5 1 Radiotherapy and Radiosurgery Department, IRCCS Humanitas Research Hospital, Rozzano, Italy. 2 Scuola di Specializzazione in Fisica Medica, Università degli Studi di Milano, Milan, Italy. 3 University College London Hospitals NHS Foundation Trust Department of Radiotherapy, University College London, London, United Kingdom. 4 Department of Medical Physics and Biomedical Engineering, University College London, London, United Kingdom. 5 Department of Biomedical Sciences, Humanitas University, Pieve Emanuele, Italy Digital Poster Highlight 4452
Made with FlippingBook - Share PDF online