ESTRO 2021 Abstract Book

S741

ESTRO 2021

(Spearman’s rank test, p<0.01). ROC analysis to predict a QA failure from the plan complexity evaluation was conducted. Results Mean gamma passing rates were 92.3% (sd 3.1%), 96.6% (sd 2.2%) and 98.8% (sd 0.8%) for Octavius 729, 1500 and 1600SRS, respectively. The differences between systems were statistically significant. CL and AL obtained with SPC were: 84.9% and 80.0% for 729, 91.3% and 88.1% for 1500 and 96.4% and 95.6% for 1600 SRS. The process is in control for all the QA systems but only the 1600SRS provided CL and AL values higher than the universal limits recommended by TG-218 (95% and 90%). Passing rate i-charts for the three systems are shown in Fig.1.

CMs showed moderate correlation with passing rates and, in general, higher complexities were associated with lower passing rates. The strongest statistically significant correlations were found for MCS (0.57 for 729, 0.73 for 1500 and 0.35 for 1600 SRS) Edge Metric (0.40 for 729, 0.37 for 1500 and 0.31 for 1600 SRS) and Q1gap (0.59 for 729, 0.80 for 1500 and 0.23 for 1600 SRS). ROC analysis showed that some CM can be used as classifiers when the exceeding of the CL evaluated with SPC is used for the "truth" parameter of the ROC curve. In Fig.2 MCS is plotted vs the passing rate for each QA device and the corresponding ROC curves are shown along with the AUC (area under the curve) values.