Bias Estimation in Method Comparison Studies

Publisher Website
Google Scholar
Abstract
A test and a reference analytical method are usually compared for agreement based on paired data obtained from several independent subjects. Bias between two methods can be classified as constant and proportional. In this article, we provide an approach for maximum likelihood estimation of total bias between two methods and partitioning it into constant and proportional bias for each subject. Normal, binomial, or Poisson distribution are the conditional distributions of the response variable that we have considered here, whereas subjects are considered to be random sample from a normally distributed population. Real data on blood cell counts and hemoglobin are used for demonstration. The estimate of biases can be used to test different statistical hypotheses and/or for graphical interpretation of the agreement. The partitioning of total biases in terms of constant and proportional gives an insight on the sources of disagreement between two methods and helps designers and manufacturer's define a remedial strategy.

This publication has 9 references indexed in Scilit: