Transcript

Proceedings of the 50th Annual ASTRO Meeting S667

3106 Clinical Results of a Medical Error Reduction/Compliance Software Program in Radiation Oncology

E. B. Kline1, P. M. Stafford2

1Advanced Partners in Radiation, Albuquerque, NM, 2New Mexico Oncology Hematology Consultants, Ltd., Albuquerque, NM

Purpose/Objective(s): Ensuring patient safety and compliance with regulations is a significant goal of radiation oncology centers.The purpose of this work is as follows:

1. Describe the design of two medical error reduction models, one paper-based and one software-based, used for: (1) monitoringkey processes and (2) self-identifying patient safety errors, accreditation failures, and regulatory violations.

2. Present how the software model: (1) analyzes errors, failures, and violations; and (2) implements a risk reduction strategy.3. Show the results of: (1) compiled data describing frequency, type and root causes; and (2) action plan(s).

Materials/Methods: The paper-based model was designed for use at 17 geographically dispersed radiation oncology clinicslocated in 9 states. The limitations and weaknesses of the paper-based model lead to the design of a software-based Medical ErrorReduction/Compliance Program named MERP. Identification and tracking of errors in MERP is accomplished using preset stan-dardized error codes and the classification of pre- and post-treatment errors.

Results: Based upon the total number of treatment fields delivered over 1.75 years as recorded by a record and verify system at 17radiation oncology centers and the total number of unintended deviations self-reported by the paper-based system, excluding the initialtime spent during the ‘‘learning curve’’, the overall error rate for both minor and significant unintended deviations was approximately0.052% (5.2 in 10,000 patient treatments). The significant unintended deviation reporting rate that could lead to a misadministrationwas calculated to be approximately 0.018% (1.8 in 10,000 patient treatments). The software-based MERP installed in 1 radiation on-cology center identified a total of 1,122 treatment-related errors over 1.5 years. The most prevalent pre-treatment errors identified are53% for record and verify, 12% for computer treatment planning, 10% for CT simulations, 9% for scheduling, 5% for portal images,and 3% for in-room treatment setups. The most prevalent post-treatment errors identified are 41% for patient documentation/notes,27% for billing (charge capture), 16% for portal images, and 12% for treatment delivery. Of the total post-treatment errors, 18 errorsaffected patient treatments. Non-patient errors showed the occurrence of 16 quality assurance and 7 radiation safety failures.

Conclusions: Based on the experience gained from the clinical application of the paper-based model at 20 centers, the software-based MERP was developed and deployed. The result of using MERP at 1 center has shown measurable improvement in thereduction of medical errors and regulatory violations.

Author Disclosure: E.B. Kline, President of RadPhysics Services - Provider of MERP software, E. Ownership Interest; P.M. Staf-ford, None.

3107 Towards Real-time Radiation Therapy: Superposition/Convolution at Interactive Rates

R. A. Jacques, R. H. Taylor, J. W. Wong, T. R. McNutt

Johns Hopkins University, Baltimore, MD

Purpose/Objective(s): By providing superposition/convolution dose calculation and intensity modulation objective derivatives atinteractive rates we enable novel Real-Time Radiation Therapy (RT2) workflows.

Materials/Methods: The Superposition/Convolution (S/C) algorithm is the clinical standard of care for dose computation. It is wellsuited to acceleration using consumer graphics processing units (GPU). We parallelized the standard S/C algorithm and implementedit on the GPU. Previously S/C was considered too slow for use in intensity modulation (IM) optimization. We implement a novel2-pass S/C based method for use in IM. Standard S/C transports a beam’s incident fluence, generated by source model, from its sourcethrough a patient representation depositing a TERMA volume. Using a pre-calculated dose deposition kernel, superposition thenspreads the energy released at each voxel out into a dose volume. The TERMA calculation suffers from read-write conflicts, whichwe solve by running groups of spatially separated divergent rays. We use the inverse dose deposition kernel as the forward kernel alsosuffers from read-write conflicts. We reduced memory bandwidth by caching lookup-tables in textures, using a variable step ray-tracer, independently attenuating each spectral energy bin in the TERMA calculation and using volumetric mip-maps to approximatethe ray as a true solid angle. The core component of all IM methods is the calculation of the derivative of the dose objectives withrespect to fluence (dO/dF). This is equal to the sum of the dose objective derivatives (dO/dD) at each voxel weighted by the dosedelivered by a unit of fluence. We compute this in stages. First, S/C computes the dose and dO/dD is calculated. Then the derivativewith respect to TERMA energy is computed by a forward superposition that gathers the weighted dose derivatives. Divergence cor-rection is applied. A fluence ray may then be cast through this volume to gather the weighted sum of dO/dD.

Results: We compared our method to Pinnacle3 (Philips - Madison, WI) for a low (4 mm) and high resolution (2 mm) 25.6 cm cubewater phantom using similar settings. Pinnacle3 was run on a SunFire v250 machine with times of 31.0s and 366.0s respectively.We performed our calculations on a NVIDIA GeForce 8800 GTX with times of 0.22s and 2.20s respectively. Using ‘Fast’ settingsdO/dF was computed in 0.17s and 1.76s respectively. Preliminary experiments using volumetric mip-maps showed performanceimprovements with minor accuracy loss.

Conclusions: We have completed a GPU accelerated superposition/convolution dose and intensity modulation objective deriva-tives engine, providing a substantial performance gain over CPU based implementations - indicating that real time dose compu-tation is feasible with the accuracy levels of the S/C algorithm.

Author Disclosure: R.A. Jacques, None; R.H. Taylor, None; J.W. Wong, None; T.R. McNutt, None.

3108 Preliminary Clinical Application of Adaptive Artificial Intelligence Technique to Inverse Treatment

Planning

H. Yan, S. Zhou, S. Das, F. Yin, C. Willett

Duke University Medical Center, Durham, NC

Purpose/Objective(s): IMRT planning requires extensive human interaction, in the form of parameter modification during plan-ning, to produce an acceptable plan. Despite this effort, the final plan can still be suboptimal, since it is impossible for a human

Top Related