The ability to predict the RR-QT relation over a range of heart rates was evaluated in 10 patients with atrial fibrillation (AF) and in 10 control subjects in sinus rhythm. The data from each subject were fitted by regression into 3 QT prediction formulas (the square root formula of Bazett, the cube root formula of Fridericia and the exponential formula of Sarma) applied in standard form and modified with a weighted average of the preceding 5 RR intervals. The goodness-of-fit of each formula was evaluated using mean square residual and Akaike information criterion. For AF, the mean square residuals did not differ among the 3 standard QT prediction formulas (Bazett 624 +/- 274, Fridericia 625 +/- 274 and Sarma 611 +/- 267) and among the 3 modified QT prediction formulas (Bazett 507 +/- 325, Fridericia 496 +/- 255 and Sarma 495 +/- 328). The weighted average modification produced a significant decrease in mean square residuals for all 3 equations (p less than 0.05) in all patients. These findings were confirmed by Akaike information criterion. Goodness-of-fit in sinus rhythm was similar to previously published reports, and significantly better than the fit for AF (p less than 0.0001). For 9 of the 10 patients with AF, sinus rhythm electrocardiograms were obtained and the above regression equations were used to predict QT intervals.(ABSTRACT TRUNCATED AT 250 WORDS)