Background:

The 200S CHRISTOPHER trial validated a diagnostic algorithm for pulmonary embolism (PE) using the Wells score, D‐dimer, and CT angiography (CTA). Thirty‐two percent of patients had a combination of low Wells score and negative D‐dimer ruling out PE with a negative predictive value of 99.4%. Two recent studies concluded that clinicians are not using the CHRfSTOPHER algorithm correctly with resultant overuse of CTA. These studies did not include patients diagnosed using ventilation/perfusion (VQ) scan or D‐dimer alone. In addition, assumptions used to impute missing Wells scores introduced unacceptable bias. Our study included all patients in whom the diagnosis of PE was considered and avoided imputation of the Wells score by comparing the output predicted by a mathematical model with observed diagnostic outcomes.

Methods:

Administrative data were used to identify patients who had CTA or VQ or D‐dimer during a 6‐month study period Radiology reports were reviewed by 2 hospitalists and adjudicated by a third in cases of disagreement. When D‐dimer was the only diagnostic test performed, clinical notes were reviewed to ensure it was used for diagnosis of PE. A decision analysis model was constructed (Fig. 1), and the probabilities for nodes 2‐4 were assigned based on data from CHRISTOPHER. A 1‐way sensitivity analysis was performed by varying the proportion of high Wells score patients at node 1 from 5% to 95%. The output was used to calculate expected values for (1) the proportion of patients with a positive radiographic study, (2) the proportion of patients ruled out with a low Wells score and negative D‐dimer, and (3) the incidence of PE.

Results:

Fifteen hundred and forty patients were considered for the diagnosis of PE. One thousand and thirty had CTA: 11% positive, 84.5% negative, 4.5% indeterminate, Two hundred and sixty‐nine had VQ; 3% high probability; 89% normal, very low, or low; and 8% intermediate. Two hundred and forty‐one had only a D‐dimer, and pretest clinical probability was documented in 42, all of whom were low risk. The model predicted a proportion of positive study ranging from 24% to 37%. We observed a rate of 9%. The proportion of patients ruled out with a low Wells score and negative D‐dimer ranged from 46% to 2%, with an observed rate of 16%. Incidence of PE ranged from 13% to 36%; the observed incidence was 7%.

Conclusions:

Thirty‐three percent of the patients in our study would had been excluded in previous studies. The observed proportion of positive radiographic studies was far below the range predicted by our model. The proportion of patients ruled out with a low Wells score and negative D‐dimer is within the range predicted by our model, but not al the incidence of PE that was actually observed. The diagnostic decision makers in our study could not have been using the CHRISTOPHER algorithm appropriately.

Author Disclosure:

R. Patrick, none; P. Patel, none; I. Masood. none; M. Auron, none; C. Whinney, Sanofi Aventis, speakers bureau; H. Gornik, none; A. Fu, none.