Assessing the validity of simulation-based sensitivity analysis for patient-specific quality assurance devices

Patient-Specific Quality Assurance (PSQA) is a crucial component of modern radiotherapy, ensuring treatment accuracy and patient safety; however assessing PSQA system sensitivity to errors remains a challenge, with no standardized methodology currently available. This study compares two sensitivity analysis methodologies for EPID-based PSQA: the Direct Error (DE) method, where errors are introduced in the delivery phase before measurement, and the Inverse Error (IE) method, where errors are incorporated into the Treatment Planning System (TPS). The aim is to determine whether the IE method can reliably replace the more time-intensive DE method for error detection and classification.

Nine treatment plans (six 3DCRT, three VMAT) were created for a thoracic phantom. Three types of errors-delivery inaccuracies, setup misalignments, and anatomical variations-were introduced at different magnitudes using both DE and IE methods. Dosimetric differences (ΔPTVD50%, ΔPTVD98%, ΔPTVD2%) were evaluated. Sensitivity and specificity were analyzed across different error thresholds.

Agreement between DE and IE methods varied by error type and threshold. While DE and IE approaches were nearly consistent for delivery inaccuracies, patient errors showed differences between these approaches. Higher thresholds (≥10%) improved agreement between DE and IE for anatomical and delivery errors but remained inconsistent for setup errors. Sensitivity was overestimated in the IE method at lower thresholds, particularly for patient positioning errors.

The IE method offers a practical alternative for PSQA sensitivity analysis but lacks commutativity with the DE method for certain error types. Validation on a subset of cases is recommended before large-scale application.