Author: Siyong Kim, Ava Paek, Joseph B. Schulz, James J. Sohn, Eric D. Stolen, Ethan D. Stolen 👨🔬
Affiliation: Department of Radiation & Cellular Oncology, University of Chicago, University of Central Florida, Virginia Commonwealth University, Department of Radiation and Cellular Oncology, University of Chicago, Department of Radiation Oncology, Stanford University School of Medicine 🌍
Purpose: The American Association of Physicists in Medicine Task Group 142 (AAPM TG-142) guidelines recommend three measurements during monthly quality assurance to keep results within two standard deviations of the mean. However, acquiring multiple measurements is resource-intensive. This study utilizes a Bayesian statistical framework to assess whether reducing the measurement count from three to two can preserve QA accuracy.
Methods: A retrospective analysis was performed on three years of monthly QA data from four linear accelerators at a single institution, spanning nine beam energies (6 MeV, 9 MeV, 12 MeV, 16 MeV, 6 MV, 6 MV FFF, 10 MV, 15 MV). Each QA session included three measurements per beam energy, following TG-142 guidelines. Posterior distributions were computed under two- and three-measurement conditions. Bayesian decision theory then estimated false-positive/negative rates at ±2% (action) and ±0.5% (tolerance) thresholds.
Results: Photon measurements showed strong agreement (98–100%) between two- and three-measurement protocols, as indicated by near-zero posterior differences in pass/fail classifications at ±2%. Electron beams exhibited slightly higher variability, notably at 6 MeV where false-positive rates rose by 0.86% with two readings. Nonetheless, this increase was clinically negligible. Bayesian model comparison yielded Bayes factors endorsing protocol equivalence for most beams, with parameter estimates lying comfortably within tolerance thresholds.
Conclusion: Although some frequentist analyses identified statistically significant differences, Bayesian posterior estimates consistently fell below clinically relevant thresholds. Electron beams, particularly at lower energies, were more sensitive to the two-measurement protocol but still displayed minimal safety risk. Overall, the data suggest that in many clinical scenarios, two readings suffice to maintain QA fidelity while reducing measurement time, labor, and cost. By emphasizing posterior distributions and decision thresholds, a Bayesian approach provides robust evidence that resource optimization need not compromise patient safety or treatment accuracy or quality.