Statistical concepts in the interpretation of serial bone densitometry.

The precision of a measurement can be expressed as the variance of multiple measurements. The coefficient of variation is a dimensionless expression of precision that is prevalent in the radiology literature. However, the coefficient of variation has important limitations. The question whether measured change is significant arises whenever any quantitative clinical parameter is followed over time. When serial bone mineral density measurements are made, change is commonly estimated as the slope of a line fitted to the serial data by linear regression. Confidence intervals for change based on this method usually assume that the precision error in measurement remains constant over time and that change is truly linear. Estimates of long-term precision may be elusive, and if constant, may vary for different individuals. If separate measurement precisions are known or are indeed constant, one can assess the level of statistical agreement of longitudinal data to linear, or other theoretical, models.