Accurate confidence intervals for measures of test performance.

ABSTRACT

The use of confidence intervals to estimate population parameters is briefly reviewed. Exact binomial confidence intervals can be calculated through the use of tables or statistical software packages. As an alternative, a microcomputer program to calculate sensitivity and specificity, point estimates and binomial confidence intervals for false-negative and -positive rate, positive and negative predictive power, prevalence of cases and non-cases, correct classification rate, and misclassification rate has also been developed. Characteristics of the computer program, 'AccuCon', which is available from the authors, are described.