A quantitative QT hysteresis model

We report on a model of the QT/RR relationship using the preceding 255 RR intervals to predict the current QT interval. The parameters of the model are sufficiently stable across two study populations to suggest the possibility of a broadly applicable quantitative model. All data is from 12-lead 24-hour Holter recordings with a sampling rate of 1000 s/s. The study populations are 33 Mortara company employees and 383 recordings from the GISSI-HF study. Each hour of RR and QT interval data, measured by an automatic algorithm, was fitted to a linear QT/RR model. The model is the simple linear one, but with the preceding RR of each normal beat QT replaced by a weighted average (called the “corrected” RR, or RR_c) of the preceding 255 RR intervals. The parameters of the model are the slope, intercept, and weights applied to the prior RR intervals. Equal weights are used for successively doubled numbers of RR intervals, going backward in time from the current beat, resulting in 8 weights (with a sum of unity) for 255 beats. Hours with insufficient normal beats or rate variation were omitted. Of 9741 hours of data, 8114 were acceptable. Illustrating the importance of the RR history, the weighted RR improved QT prediction by reducing rms error to 2.94ms from 5.74ms using only the prior RR. A composite set of weights, taken from the average of all results, and applied to each hour resulted in an rms prediction error for QT of 3.12ms, suggesting that one universal set of RR weights may apply to a diverse population. Further, the largest mean difference (between universal and hourly optimized weights) of rms errors for any single recording was 1.0ms.