The accuracy of radionuclide ventriculography in the analysis of left ventricular ejection fraction

To assess the reliability of using ejection fraction values derived from gated blood pools scans as a way to classify patients, we reviewed 76 radionuclide studies. Each study was processed in 10 different ways (varying or fixed left ventricular region of interest in association with 5 different backgrounds), and compared with contrast ventriculograms performed within one week of the radionuclide study. Correlation values with contrast results varied from 0.76 to 0.94 (mean 0.84 +/- 0.06, +/- 1 SD), and linear fit slopes varied from 0.39 to 0.90 (mean 0.75 +/- 0.15). Based on contrast results, patients were divided first into four groups (severely reduced, reduced, normal, and high ejection fractions). Predictive accuracy of the radionuclide estimates for these four groups was poor, ranging from 0.39 +/- 0.08 to 0.82 +/- 0.06%. Sensitivity was somewhat better (range 0.53 +/- 0.11 to 0.78 +/- 0.08%), and specificity better yet (0.78 +/- 0.08 to 0.95 +/- 0.02%). The results were better with techniques resulting in higher correlation values, and further improved when patients were regrouped into two populations (normal or reduced), though 90% of subjects were correctly identified only when correlation values exceeded 0.90. We conclude that care must be taken in the identification of patients using radionuclide angiography who have normal and abnormal left ventricular function; it can be done with confidence only when linear fits with contrast ventriculographic results have high correlation coefficients. Stratification attempts should be reduced to a minimum, as multiple population subgroups result in substantial patient mislabeling.