Program Information
Individual Dose Alert Thresholds for Abdominal CT Protocols Due to Statistical Independence of Dose Distributions
BK Stewart*, University of Washington School of Medicine, Seattle, WA
Presentations
SU-G-206-14 (Sunday, July 31, 2016) 4:00 PM - 6:00 PM Room: 206
Purpose: To determine whether a single dose management system alert threshold will suffice for the 34 abdomen protocols on our institution's CT scanners, or if these alert thresholds should be set per protocol. Parametric statistical analysis is ideally valid only for distributions demonstrating: (1) normality, (2) homogeneity of variance, (3) interval scale and (4) independence. Dose distributions are notoriously non-normal and thus non-parametric statistical analysis is typically required.
Methods: Dose distribution data were collected from a HD750 CT scanner (GE Healthcare) at our institution over the third and fourth quarters of 2015. In total there were 1793 abdomen studies using 34 different protocols. Protocols were selected for statistical analysis (SPSS, IBM Corp.) if the count over the period was ≥ 50 to produce statistically meaningful results. Nine protocols with count ranging from 61-398 were selected. Protocol dose distributions were statistically analyzed to determine: (1) normality – Shapiro-Wilk test, (2) homogeneity of variance – Levene Statistic, (3) similarity of medians – Median test, and (4) similarity of distribution – Kruskal-Wallis test.
Results: For the nine abdomen protocols used with exam count ≥ 50, the dose distributions were statistically significantly different from normal for eight protocols (p≤[0.001-0.035]), homogeneity of variance invalidated for five protocols (p≤[0.001-0.037]), median values not the same across all protocols (p<0.001), and all dose distributions different from each other (p<0.001).
Conclusion: That all abdomen protocols with exam count ≥ 50, the distributions were predominantly: (1) non-normal, (2) of unequal variance over the distributions, (3) of unequal means and (4) non-identical distributions indicates that the abdomen protocols in general must be treated independently with non-parametric statistical analysis and thus the alert thresholds should be calculated and set individually, that is, there cannot be a solitary meaningful alert threshold utilized over all abdominal protocols examined in this study.
Contact Email: