Purpose: Referral bias, in which the result of a diagnostic test affects the subsequent referral for a more definitive test, influences the accuracy of noninvasive tests for coronary artery disease. This study evaluates the effect of referral bias on the apparent accuracy of single photon emission computed tomography (SPECT).
Methods: Over a 10-year period, 14,273 patients without known coronary artery disease underwent stress SPECT. Coronary angiography was performed within 3 months after the stress test in 1853 patients (13%). The apparent sensitivity, specificity, and likelihood ratios of SPECT were determined in these patients, and then adjusted for referral bias using two different formulas.
Results: The overwhelming majority (95%) of patients who underwent angiography had abnormal SPECT images. Apparent values for test indices were a sensitivity of 98%, a specificity of 13%, a likelihood ratio for a positive test of 1.1, and a likelihood ratio for a negative test of 0.15. Test indices adjusted for referral bias (using the two methods) were a sensitivity of 65% or 67%, a specificity of 67% or 75%, a likelihood ratio for a positive test of 2.0 or 2.7, and a likelihood ratio for a negative test of 0.44 or 0.52.
Conclusion: Referral bias has a marked effect on the apparent accuracy of stress SPECT for the diagnosis of coronary disease. Adjustment for referral bias yields estimates for sensitivity and specificity and likelihood ratios that better reflect the accuracy of the technique.