SU‐E‐I‐06: A Robust and Improved Method of Analysis for Dose Calibrator Linearity Data Based On First Principles

Abstract

Purpose:Identify shortcomings in the common mathematical methods of dose calibrator linearity analysis, then propose an improved and more robust method.Methods:Precision of activity measurements using attenuator sleeves was measured using inter‐user (several users) & intra‐user (single user) measurements. We performed retrospective follow‐up investigation of several reported dose calibrator linearity failures at our institution. Common methods of linearity analysis generate best‐fits without using the known decay constant, no‐weighting of data points, and/or compare scaled values to the “zero” measurement (inherently assumed to be error‐free).Fixing the known slope (‐λ) of our data on a semilog plot, we then use least‐squares methodology to find y‐intercept of data's best fit line, against which linearity is assessed.Results:Two‐sigma precision ranges of attenuator sleeve measurements were ±0.4–1.2% for inter‐user measurements, and ±0.4–2.1% for intra‐user measurements. Retrospective analysis of all reported dose calibrator linearity failures from previous four years showed that robust least squares method of analysis would have prevented all repeat testing. Trending data using different methods shows more consistent data ranges (max. deviation 4.9–6.7% versus 3.2–4.2% for select datasets) i.e. more consistent device performance (as anecdotally expected) is demonstrated when analyzed with least squares method. Lower absolute maximum variation values are also calculated across nearly all linearity data sets when using least squares method of analysis.Conclusion:Precision of attenuator sleeve measurements confirm signal averaging must be used in analysis. Dose calibrator performance specification for accuracy is 1%. Linearity tolerance for heightened scrutiny, and arguably a best practice action limit, is 5%. Thus a considerable fraction of the allowed tolerance for the test can be occupied by measurement error. Proposed method of analysis is improvement over other common methods based on first principles, and is demonstrated to be robust.