Which to use to evaluate change in BMD at follow-up: RMS-SD or RMS-%CV?

There is bias toward the use of root-mean-square standard deviation (RMS-SD) over root-mean-square coefficient of variation (RMS-%CV) to calculate a least significant change (LSC) value that is used to determine if change in bone mineral density (BMD) at follow-up is statistically significant. In part, this bias is based on the assumption that SD is relatively constant over a wide range of BMD as opposed to %CV, which is assumed to increase as BMD decreases. However, evidence to support these assumptions is scant. The purpose of this project was to determine the frequency of discrepancies in interpretation of follow-up data using an LSC based on RMS-SD (LSC-SD) vs RMS-%CV (LSC-%CV). A convenience sample of 100 patients returning for follow-up dual-energy X-ray absorptiometry scans was used to compare the frequency of change in BMD exceeding LSC-SD vs LSC-%CV. Results showed that agreement in exceeding (or not) both LSC-SD and LSC-%CV was 94% for L1-L4 (lumbar spine), femoral neck, and total hip. Disagreements were characterized by BMD changes that usually marginally exceeded either LSC-SD or LSC-%CV but not the other, and they occurred at both high and low absolute BMD. We conclude that significant change in BMD typically exceeds both LSC-SD and LSC-%CV. Thus, there seems to be no basis for the bias in favor of using LSC-SD as long as RMS-SD and RMS-%CV are calculated in the same manner. Furthermore, a criterion that both LSC-SD and LSC-%CV must be exceeded may help avoid overcalling marginal change as being significant, especially at high and low BMD.