[P187] A crowd-knowledge-based analysis of DVHs in SBRT: First steps towards a national virtual audit

Purpose Currently, most of the multicenter analyses on treatment planning rely on the extraction of selected data from the DVH of each plan. A grouped analysis can be biased due to different algorithms implemented in different TPSs used to generate the DVH. In this work we used a consistent method to present a preliminary analysis of multiple data coming from a national survey on stereotactic body radiotherapy (SBRT) planning. Methods A single spine case was shared among 36 radiation oncology centers. The dose prescription was 30 Gy in 3 fractions with specific constraints on target coverage and dose to nearby organs at risk. Data were collected in DICOM-RT format. A script was developed in R language using the RadOnc R-Package for recalculating the DVHs using the same algorithm. Specific DVH points (V30Gy, D90%, D2%) collected from the centers were compared with those recalculated with RadOnc. A grouped analysis of recalculated DVHs was performed therefore eliminating the bias due to different DVH calculation algorithms. Results Differences between collected and recalculated DVHs were minimal, however in some cases deviations up to 1.5% were observed. The multiple-DVH analysis showed a notable variability on target dose level up to 150% likely related to constraints on target coverage and SBRT technique. This variability was caused mainly by different planning optimization strategies, rather than by the use of a specific treatment technology. Conclusions The observed variability suggests that comparable standards in patient treatment among different centers can be obtained if a consistent high-level data sharing capability is granted. In the strive to harmonize the planning process, this analysis constitutes a first step toward the creation of a platform of crowd-knowledge-based planning guidelines. This platform could represent a high-quality benchmark for those centers that are willing to implement SBRT techniques.