Assessing reproducibility by the within-subject coefficient of variation with random effects models.

In this paper we consider the use of within-subject coefficient of variation (WCV) for assessing the reproducibility or reliability of a measurement. Application to assessing reproducibility of biochemical markers for measuring bone turnover is described and the comparison with intraclass correlation is discussed. Both maximum likelihood and moment confidence intervals of WCV are obtained through their corresponding asymptotic distributions. Normal and log-normal cases are considered. In general, WCV is preferred when the measurement scale bears intrinsic meaning and is not subject to arbitrary shifting. The intraclass correlation may be preferred when a fixed population of subjects can be well identified.