Author: Zachary Carr, Katie W. Hulme, Zaiyang Long, Nathan A. Quails, Ashley Tao 👨🔬
Affiliation: Gundersen Health System, Ohio State University Wexner Medical Center, Mayo Clinic, Ohio State Wexner Medical Center, The Cleveland Clinic 🌍
Purpose: Meaningful interpretation of deviation index (DI) in clinical practice relies on appropriately set target exposure indices (EIT). EIT values for a given exam-view can be derived from the median DI if practice is relatively consistent. This work aims to establish guidance criteria for assessment of practice consistency to appropriately use DI statistics to set EITs.
Methods: Three institutions under IRB approval extracted EI, EIT, and DI data for statistical analysis. Radiographic units from 5 manufacturers were verified for EI accuracy and yielded 107,226 data points. The standard deviation, skew, and excess kurtosis (EK) were calculated for the DI distribution for twelve exam-views for each unit. Calculated metrics were averaged across units, weighted by number of data points, to quantify nominal practice variation in DI for each exam-view (σDI,w, skewDI,w, EKDI,w). Outliers were then identified either through known errors or statistical analysis (|Z-score|>3) and removed from each data set. Analysis was repeated on the filtered data. Proposed benchmark values (σ<2.5, |skew|≤1.5, and EK>0) were applied, and flagged distributions were evaluated for appropriateness for establishing EIT.
Results: σDI,w ranged from 1.23-4.70 and 1.00-3.96, skewDI,w ranged from 0.02-7.84 and 0.07-0.71, and EKDI,w ranged from 0.54-240.24 and 0.41-4.72, with and without outliers. Results demonstrate sensitivity to outliers and indicate removal is a necessary step before using these metrics to assess practice consistency. For filtered data, average σDI,w was 2.45 and 1.10, skewDI,w was 0.22 and 0.47, and EKDI,w was 1.21 and 2.86, for manual versus photo-timed exams, respectively. Using proposed benchmarks, 29% of distributions were flagged for inconsistency. Visual assessment of flagged distributions revealed that 56% could be used to adjust EIT which informed recommended benchmark values (σ<3.0, |skew|≤1.5, and EK>-1).
Conclusion: Benchmark values for a first pass check of practice consistency have been established.