A study to develop and validate a clinical prediction rule to exclude pulmonary embolism.

Abstract

Statement of the problem. Pulmonary Embolism (PE) is a common, lethal and treatable condition that is only present in a minority of those investigated with suspected PE. A clinical prediction rule that safely excludes PE could avert the need for many expensive diagnostic tests and improve patient management. Methods. Phase I - Using univariate techniques potential predictor variables were identified from a prospective cohort study of patients with suspected PE who subsequently had outcomes measured (Derivation set). Phase II - The inter-observer reliability of these individual predictor variables were determined in a second prospective cohort study of patients with suspected PE (Inter-observer reliability study). Phase III - Two multivariate techniques (recursive partitioning and logistic regression) were used to derive clinical prediction rules in the derivation set based on those variables that were both significant in univariate analysis (p < 0.20) (Phase I) and had good inter-observer reliability (kappa > 0.5) (Phase II). The clinical prediction rule with the highest specificity with close to 100% sensitivity was then identified using classification analysis in the derivation set. Phase IV - The rule was then retrospectively validated in a third dataset from a previously published prospective cohort study of patients with suspected PE (Validation set). (Abstract shortened by UMI.)