Clinical prediction rules for pulmonary embolism: a systematic review and meta-analysis

SUMMARY BACKGROUND: Pretest probability assessment is necessary to identify patients in whom pulmonary embolism (PE) can be safely ruled out by a negative D-dimer without further investigations. OBJECTIVE: Review and compare the performance of available clinical prediction rules (CPRs) for PE probability assessment. PATIENTS/METHODS: We identified studies that evaluated a CPR in patients with suspected PE from Embase, Medline and the Cochrane database. We determined the 95% confidence intervals (CIs) of prevalence of PE in the various clinical probability categories of each CPR. Statistical heterogeneity was tested. RESULTS: We identified 9 CPR and included 29 studies representing 31215 patients. Pooled prevalence of PE for three-level scores (low, intermediate or high clinical probability) was: low, 6% (95% CI, 4-8), intermediate, 23% (95% CI, 18-28) and high, 49% (95% CI, 43-56) for the Wells score; low, 13% (95% CI, 8-19), intermediate, 35% (95% CI, 31-38) and high, 71% (95% CI, 50-89) for the Geneva score; low, 9% (95% CI, 8-11), intermediate, 26% (95% CI, 24-28) and high, 76% (95% CI, 69-82) for the revised Geneva score. Pooled prevalence for two-level scores (PE likely or PE unlikely) was 8% (95% CI,6-11) and 34% (95% CI,29-40) for the Wells score, and 6% (95% CI, 3-9) and 23% (95% CI, 11-36) for the Charlotte rule. CONCLUSION: Available CPR for assessing clinical probability of PE show similar accuracy. Existing scores are, however, not equivalent and the choice among various prediction rules and classification schemes (three- versus two-level) must be guided by local prevalence of PE, type of patients considered (outpatients or inpatients) and type of D-dimer assay applied