A structured clinical model for predicting the probability of pulmonary embolism

Purpose To develop a structured model to predict the clinical probability of pulmonary embolism. Methods We studied 1100 consecutive patients with suspected pulmonary embolism in whom a definite diagnosis had been established. We used logistic regression analysis to estimate the probability of pulmonary embolism based on patients’ clinical characteristics; the probability was categorized as low (≤10%), intermediate (>10%, ≤50%), moderately high (>50%, ≤90%), or high (>90%). Results The overall prevalence of pulmonary embolism was 40% (n = 440). Ten characteristics were associated with an increased risk of pulmonary embolism (male sex, older age, history of thrombophlebitis, sudden-onset dyspnea, chest pain, hemoptysis, electrocardiographic signs of acute right ventricular overload, radiographic signs of oligemia, amputation of the hilar artery, and pulmonary consolidation suggestive of infarction), and five were associated with a decreased risk (prior cardiovascular or pulmonary disease, high fever, pulmonary consolidation other than infarction, and pulmonary edema on the chest radiograph). With this model, 432 patients (39%) were rated a low probability, of whom 19 (4%) had pulmonary embolism; 283 (26%) were rated an intermediate probability, of whom 62 (22%) had pulmonary embolism; 72 (7%) were rated a moderately high probability, of whom 53 (74%) had pulmonary embolism; and 313 (28%) were rated a high probability, of whom 306 (98%) had pulmonary embolism. Conclusion This prediction model may be useful for estimating the probability of pulmonary embolism before obtaining definitive test results.