Reducing the Potential for Catastrophic Radiation Error by Using Historical Treatment Data

Abstract

Listen

Translate article

The increasing complexity of IMRT treatment plans requires carefully checking the correctness and consistency of an increasingly large number of treatment parameters between planning and each delivery session. A new level of safety can be added in the planning and delivery process by exploiting the similarities that exist among treatment parameters of a large pool of plans for the same disease using the same technique. The goal of this study is to develop the proof of concept for a software program that analyzes machine parameter distributions from historical treatment data for a given disease site and alerts the user of any deviation. The distribution of treatment parameters for a population of 60 patients treated with whole brain irradiation using a forward planning technique with parallel opposed beams was analyzed. A total of 15 treatment parameters were considered, including the number of beams, beam energy, gantry, collimator and couch angles, SSD, field size, number of monitor units, number of monitor units (MU) per Gy at isocenter and beam weight. For each parameter, a range of acceptable values was extracted from the population distribution. A new plan was considered consistent with historical data if each of its parameter values was compatible with 60% of the population. In order to test the software, errors such as a wrong number of beams, beam energy, CT dataset or absence of heterogeneity correction were manually introduced in new plans. As expected, the population of whole brain plans is very homogeneous. The SSD and number of MU per Gy exhibited a narrow Gaussian distribution. The field size, beam weight and number of MU show 2 Gaussian peaks corresponding to the open field and field-in-field, respectively. All other parameters had a single value. These narrow distributions made deviations easy to detect. The plans with the wrong number of beams or beam energy were obvious. The plan using the wrong CT dataset led to an SSD that was outside the historical range and was detected. Interestingly, the plan calculated without homogeneity correction was not detected. However, since such corrections are not of great importance for whole brain radiation, the machine parameters were not affected and such a treatment would not have led to a radiation error. We developed a software program to compare each new treatment plan to historical treatment parameters for the same disease. We successfully tested it on a population of whole brain treatment plans. The use of such a program can detect potential errors that were accidentally introduced during data transfer and recording, such as those reported in the New York Times. In addition, it will ultimately demonstrate that by using more standardized treatment approaches for a given disease, the risk for errors can be reduced.