Evaluation of competing diagnostic tests: sequences for the diagnosis of pulmonary embolism, Part II

Tests for pulmonary embolism (PE) and its most frequent source, deep vein thrombosis, include angiography, ventilation-perfusion nuclear medicine scans, and contrast and radionuclide venography. Although selective angiography is the definitive procedure for diagnosis of PE, its risk of death, although small, as well as the morbidity associated with contrast injections, is high enough that alternative, less accurate but safer, diagnostic procedures are preferred. Part I of this report described the methodology for selection of the single test having the lowest effective cost based on the test cost, its risks, and diagnostic performance. Other than the pulmonary angiogram, however, no single test has a sufficiently high diagnostic performance to be clinically realistic: incorrect diagnoses can result in sudden death. Sequences of tests or algorithms can be used to lower the overall risk of the tests to patients while increasing correct decision making to reach a clinically acceptable level. These points are illustrated by comparing three commonly recommended algorithms for the diagnosis of PE. Additionally, the prevalence of PE affects the choice of the algorithm, and we found that no single algorithm is best for all values of prevalence. In summary, we found that the most cost-effective strategy was to select the particular algorithm having the lowest effective cost for the relevant prevalence value. Use of algorithms also decreased the overall risk of diagnostic test complications and the number of incorrect diagnoses.