Ventilation-perfusion scanning for pulmonary embolism: refinement of predictive value through Bayesian analysis

Abstract

The diagnosis of pulmonary embolism (PE) by pulmonary perfusion and ventilation scintigraphy presents problems common to all imperfect diagnostic tests. Bayesian analysis indicates that the posttest probability of PE is a function of the prevalence (or pretest probability) of PE as well as the scintigraphic findings. The authors propose that Bayesian analysis allows an explicit refinement in communicating the implications of scintigraphic findings to referring clinicians. Recent data reported from a prospective study of ventilation-perfusion scanning compared with pulmonary angiography in patients suspected of PE were reviewed. Using the reported scintigraphic and chest radiographic findings, the sensitivity and specificity of each of the various test result combinations for angiographically proven PE were derived. The overall prevalence of PE was estimated to be 20% in patients suspected of PE who were referred for nuclear imaging. A Bayesian analysis was then performed for each category of test result to estimate the posttest probability of PE for different prevalence estimates. If a perfusion study alone is done which shows segmental or larger defects without corresponding chest radiographic changes, the sensitivity for PE is 80%, and the specificity is 86%. With a 20% prevalence of PE, the posttest probability of PE is 58%. The use of ventilation imaging improves the predictive power of the test by its effect on specificity. When a ventilation image shows preserved (mismatched) ventilation in concert with the above findings, the sensitivity for PE is 75%, and the specificity increases to 95%. For these findings the posttest probability increases to 79%, and for a prevalence of 50%, the posttest probability of PE is 94%.