The heart rate-corrected QT interval of conscious beagle dogs: a formula based on analysis of covariance

Three frequently used and cited formulas used to rate correct the QT interval (Bazett's, Fridericia's, and Van de Water's) were compared and ranked using a large population-based cohort of beagle dogs (99 males and 99 females). In addition, analysis of covariance was used to derive a flexible method to rate correct the QT.interval for heart rate. The method is flexible in that it utilizes pretest or control data to determine the degree of correction. In addition, it can also be used to evaluate whether treatment alters the association between heart rate and QT. Specifically, pretest QT (unadjusted) and heart rate data were used to estimate coefficients in the linear regression log(QT) = alpha + beta log(HR). The estimated slope (beta) from the pretest data was used to heart rate correct the QT interval in the formula log(QT)ca = log(QT) - beta *[log(HR - log(HRm)]. The term "log(HRm)" is included to standardize QTca to a reference value, either a fixed value or an average heart rate for the data set being analyzed. These formulas were retrospectively compared under a typical toxicity study paradigm with a class III antiarrhythmic agent (L-768,673) that selectively prolongs the QT interval by blocking the slow activating component of the delayed rectifying potassium channel (lks). Based on their ability to dissociate the effects of heart rate on the QT interval, the formulas received the following ranking: Covariate Adjustment (preferred) = Van De Water's > Fridericia's > Bazett's (not recommended). Analysis of covariance based on pretest or control data is preferred for moderate to large studies where there are adequate data for estimation of the slope parameter beta, the investigator does not have sufficient control over HR, or treatment alters the association between HR and the QT interval. Conversely, for smaller studies a fixed rate adjustment formula from the literature (such as Van de Water's or Fridericia's equations) may be preferable since the bias from using a fixed formula is likely to be smaller than the variance resulting from estimating beta from a small sample.