Suspected pulmonary embolism and lung scan interpretation: trial of a Bayesian reporting method.

Objective: To determine whether a Bayesian method of lung scan (LS) reporting could influence the management of patients with suspected pulmonary embolism (PE).

Design: 1) A descriptive study of the diagnostic process for suspected PE using the new reporting method; 2) a non-experimental evaluation of the reporting method comparing prospective patients and historical controls; and 3) a survey of physicians' reactions to the reporting innovation.

Setting: University of Virginia Hospital.

Patients: Of 148 consecutive patients enrolled at the time of LS, 129 were completely evaluated; 75 patients scanned the previous year served as controls.

Intervention: The LS results of patients with suspected PE were reported as posttest probabilities of PE calculated from physician-provided pretest probabilities and the likelihood ratios for PE of LS interpretations.

Results: Despite the Bayesian intervention, the confirmation or exclusion of PE was often based on inconclusive evidence. PE was considered by the clinician to be ruled out in 98% of patients with posttest probabilities less than 25% and ruled in for 95% of patients with posttest probabilities greater than 75%. Prospective patients and historical controls were similar in terms of tests ordered after the LS (e.g., pulmonary angiography). Patients with intermediate or indeterminate lung scan results had the highest proportion of subsequent testing. Most physicians (80%) found the reporting innovation to be helpful, either because it confirmed clinical judgement (94 cases) or because it led to additional testing (7 cases).

Conclusions: Despite the probabilistic guidance provided by the study, the diagnosis of PE was often neither clearly established nor excluded. While physicians appreciated the innovation and were not confused by the terminology, their clinical decision making was not clearly enhanced.