A formula for estimating pretest probability: evaluation and clinical application.

Knowledge of the prevalence (or pretest probability) of a disease is necessary for the interpretation of the results of a diagnostic test in a specific population of patients. This paper evaluates a formula for estimating the prevalence of a disease in a population, based on the proportion of patients with abnormal test results in that population and the known sensitivity and specificity of the test. The authors tested the formula by using it to estimate the prevalence of myocardial infarction in 215 patients with chest pain admitted to a coronary care unit, based on results of initial total creatine kinase determinations. The estimated prevalence was 30%. The true prevalence of myocardial infarction, based on established diagnostic criteria, was 25% (95% confidence interval 19.2%-30.8%). To further evaluate the formula, a sensitivity analysis was performed. Errors in estimated prevalence were inversely related to test sensitivity and specificity, positively related to the magnitude of the differences between presumed and true test sensitivity and specificity, and complexly related to the true prevalence of disease. This formula permits the estimation of prevalence of a disease in a population without resorting to the use of a "gold standard" test, which is often invasive or impractical. Situations are presented where the formula could be used to evaluate and improve the utilization of laboratory tests.