Using the Principles of Bayesian Statistics to Improve the Performances of Medical Diagnostic Tests

摘要

The study analyzed diagnostic data from 432 patients evaluated in the emergency department for shortness of breath. Electronically recorded heart sounds and brain natriuretic peptide (BNP) were analyzed to determine whether left ventricular systolic dysfunction (LVSD) with heart failure caused each patient's symptoms. In each patient's computerized digital electrocardiogram, measurements of the duration of the QRS complex were made. As expected, the data show that the QRS complex duration itself does not detect LVSD with heart failure. However, for both the recorded heart sounds and the BNP data, the diagnostic sensitivities at 98% specificity for LVSD with heart failure are significantly greater in the subgroups with prolonged QRS complex duration than they are in the entire group of subjects. Evaluating QRS complex duration on the digital electrocardiogram can assess the prior probability of underlying heart disease and improves the performances of diagnostic tests for LVSD with heart failure.