USING LINEAR-REGRESSION FOR METHODS COMPARISON

Laboratories use linear. regression as a tool to compare methods, either to decide whether to replace one method with another, to compare the analytical performance of two methods, or to make the results of the two methods equivalent. Linear regression is a requirement in the National Committee on Clinical Laboratory Standards (NCCLS) document EP9-T, Method Comparison and Bias. Computer software programs can simplify the process. Technologists should be aware of four patterns of bias-constant error, proportional bias, random error, and outliers. When determiningg whether the methods are adequate the technologist evaluates against a predetermined medical decision point, and has predetermined an acceptable range of bias. With these pieces of information, he or she can calculate the bias and its 95% confidence interval using equations in the NCCLS document. This is the second article in a continuing education series on statistics. Other articles in this series focus on the control chart, assessment of linearity and the creation of reference ranges.