Dual X-ray absorptiometry quality control: comparison of visual examination and process-control charts.

Dual X-ray absorptiometry (DXA) is widely used to monitor treatment efficacy in reducing the rate of bone mineral loss. In order to assure the validity of these measurements, instrument quality control of the DXA scanners becomes very important. This paper compares five quality control procedures (visual inspection, Shewhart chart with sensitizing rules, Shewhart chart with sensitizing rules and a filter for clinically insignificant mean changes, moving average chart and standard deviation, and cumulative sum chart [CUSUM]) in their ability to identify scanner malfunction by means of (1) an analysis of five longitudinal phantom data sets that had been collected during a clinical trial and (2) an analysis of simulated data sets. The visual inspection method is relatively subjective and depends on the operator's experience and attention. The regular Shewhart chart with sensitizing rules has a high false alarm rate. The Shewhart chart with sensitizing rules and an additional filter for clinically insignificant mean changes has the lowest false alarm rate but a relatively low sensitivity. The CUSUM method has good sensitivity and a low false alarm rate. In addition, this method provides an estimate of the date a change in the DXA scanner performance might have occurred. The method combining a moving average chart and a moving standard deviation chart came closest to the performance of the CUSUM method. Comparing the advantages and disadvantages of all methods, we propose the use of the CUSUM method as a quality control procedure for monitoring DXA scanner performance. For clinical trials use of the more intuitive Shewhart charts may be acceptable at the individual sites provided their scanner performance is followed up by CUSUM analysis at a central quality assurance center.

Estimating the precision of estimates of genetic parameters realized from multiple-trait selection experiments.

The genetic change from multiple-trait selection experiments can be equated to the regression of genotype on phenotype. This gives rise to a method of obtaining estimates of additive genetic variances and covariances. The method requires the use of selection weights, derived by means of the index-in-retrospect, to provide invariant solutions. Solution variance estimates obtained from Monte Carlo simulation do not agree with variance estimates from ordinary least squares methods. This indicates that the errors are distributed with some structure V. A form of V is proposed which utilizes knowledge of the errors. Monte Carlo variance estimates from generalized least squares (GLS) methods agree closely with the average variance estimates from GLS when the proposed V is used. Use of an estimated V, derived after the initial estimation procedure, is shown to provide adequate information on the variance of the estimates.