Development of Institutional Consistency Metrics for the Prevention of Injurious and Reportable Radiation Delivery Errors
J Molloy1*, U Langner2, S Gerring3, D Lewis4, P Otageri5, J Yagelski6, M Gillie7, T Greist8, (1) Univ Kentucky, Lexington, KY, (2) University of Kentucky, Lexington, Kentucky, (3) ,Lexington, KY, (4) ,Lexington, KY, (5) ,Lexington, KY, (6) ,Michigan City, IN, (7) ,Bristol, VA, (8) ,Columbus, OHSU-E-T-205 Sunday 3:00PM - 6:00PM Room: Exhibit Hall
Radiotherapy treatment techniques are too complex for humans to assess the reasonableness of machine settings. Contrary to conventional wisdom, generalized logical tests, such as comparison to institutional norms, can be applied to patient-specific machine settings. Such tests can be automatically applied to machine settings immediately prior to the point of treatment delivery and have a powerful potential to reveal and prevent potentially injurious errors.
We tested two consistency metrics, one each for conventional (Mc,conv) and IMRT (Mc, IMRT) treatments. The metrics assess the weighted, total number of monitor units in a proposed treatment. Refinement of the IMRT metric was tested for the larynx category and incorporated an optimized exponential. The performance of Mc,conv was tested using Receiver Operator Characteristic (ROC) analysis. A deviation of > 2σ was flagged as an error by the metric, and ground truth was determined by full dose calculation, with either 20% or 5% dose thresholds considered indications of significant errors. The ROC analysis was performed for purely dosimetric errors, as well as for a cadre of dosimetric and geometric errors.
The standard deviations in the metrics used for conventional and step-and-shoot IMRT treatments are less than 13%, and 17%, respectively. The optimized exponential in the IMRT metric reduced the standard deviation of the larynx population from 14 % to 6 %, thus improving the predictive power of the metric. ROC analysis indicates that the metrics sensitivity to dosimetric errors is strong at both the 20% and 5% dose threshold levels. Even when geometric errors are included, the metric retains good predictive power for maximum dose.
Comparison of patient-specific treatment delivery parameters to institutional norms can reveal dosimetric errors of less than 20%. These methods have the potential to prevent potentially injurious radiation therapy errors from propagating to treatment.
Funding Support, Disclosures, and Conflict of Interest: Varian Corporation research grant