|
Abstract:
|
When a new diagnostic device with continuous output is developed, we can evaluate its measurement accuracy by comparison with an established device by a so-called method comparison study. To evaluate method comparison studies, regression analysis is generally performed and regression slope ?1, intercept ?0, and their associated confidence intervals are reported. Sufficient sample size is needed such that 95% confidence intervals on ?1 and ?0 are narrow enough to conclude that slope ?1 is near unity and intercept ?0 is near zero to within a clinical insignificant deviation delta to support the equivalence of the two measurement procedures. To show the substantial equivalence of the new device to the predicate device, it is generally requested to provide pre-specified bias limit (their 95% confidence intervals) from regression analysis at medical decision points in addition to slope ?1 is near unity and intercept ?0 is near zero. CLSI EP-9A3 shows two methods that estimate bias. In this study, these two different methods will be compared with simulated data. Assumptions of each method will be reviewed and the results will be assessed when the assumption is violated.
|
