Reader variability in QT measurement due to measurement error and variability in leads selection: a simulation study comparing 2-way vs. 3-way interaction ANOVA model.

Reader variability (RV) results from measurement differences or variability in lead used for QT measurements; the latter is not reflected in conventional methods for estimating RV. Mean and SD of QT intervals in 12 leads of 100 ECGs measured twice were used to simulate data sets with inter-RV of 5, 10, 15, 20, and 25 ms and intra-RV of 3, 6, 9, 12, and 15 ms. Six hundred twenty-five data sets were simulated such that different leads were used in Read1 and Read2 in 0, 10%, 20%, 30%, 40% of ECGs by 25 readers. RV was estimated using ANOVA interaction models: three-way model using Reader, ECG and lead as factors, and 2-way model using reader and ECG as factors. Estimates from three-way model accurately matched inter- and intra-RV that were introduced during simulation regardless of percent of ECGs with lead selection variability. The two-way model provides identical estimates when both reads are in same leads, but higher, more realistically estimates when measurements are made in different leads.