Abstract
Test results from "normal" and "non-normal" individuals frequently overlap. Individuals with test results near the region of overlap have a high risk of being misdiagnosed. We present a statistical method for quantifying the certainty of diagnoses and defining a normal range, and illustrate its application with a specific example from a Tay-Sachs disease carrier screening program. This method can be applied to any test result based on a continuous variable and is particularly well suited to screening programs where the risk or incidence of a disease is known. We use an inconclusive range to reduce the likelihood of incorrect diagnoses resulting from measurement error and borderline results. The limits of the normal, inconclusive, and nonnormal ranges are based on three considerations: (1) the probability of misdiagnoses, (2) the expected frequency of inconclusive diagnoses, and (3) the reproducibility of the test results.
