Diagnostiek van ernstige aandoeningen

Diagnosing pulmonary embolism is an important problem. Several risk factors have been combined in prediction models, such as the Wells en Geneva scores. These models were prospectively validated in a large Dutch study (Prometheus study). With respect to this study, two critical issues are noted. Firstly, the discriminative ability of the models was modest, with areas under the Receiver Operating Characteristic (ROC) curve around 0.7. Secondly, the cut-offs to define low risk groups for the prediction rules were rather liberal, defining 60-70% of the included patients as ‘low risk’, while around 15% of these had a pulmonary embolism. In combination with a D-dimer test only 22-23% were classified as negative. This test missed only 1 patient among 186 patients identified with a pulmonary embolism. Performing a D-dimer test is hence essential in all patients in the selection for a CT scan. Improvements in the prediction rules are necessary to make these meaningful in the diagnostic work-up for pulmonary embolism.

Conflict of interest: E.W. Steyerberg received royalties from Springer for his book Clinical Prediction Models. Financial support: none declared.