DXA longitudinal quality control: a comparison of inbuilt quality assurance, visual inspection, multi-rule Shewhart charts and Cusum analysis.

The performance of dual-energy X-ray absorptiometry (DXA) instruments can be monitored using various quality control (QC) procedures. It has not been established which of these is most appropriate. The aim of this study was to determine which of four QC procedures is the best to use for longitudinal monitoring. Eighteen centres with DXA instruments scanned an aluminium spine phantom weekly for up to 16 months, and the bone mineral density data were used for the QC procedures. The methods investigated were the instrument's inbuilt quality assurance (QA) procedure, visual inspection, multi-rule Shewhart control charts, and Cusum analysis using a truncated-V mask. True and false positive fractions (TPF and FPF) of each method were calculated, including those for a range of action levels for the Shewhart charts and dimensions of the Cusum mask. For Shewhart, the action levels giving the most desirable TPF and FPF were whole multiples of the standard deviation (SD). For Cusum, the most desirable mask dimensions were 3.6 SD for the total height of the vertical section and 0.9 SD per data point for the gradient of the wings. Predictive power of each method as a means of fault detection was decided by the number of faults detected out of a total of 8 non-mechanical faults subsequently diagnosed. The inbuilt QA detected 2, visual inspection 7, Shewhart chart 7 and Cusum analysis 7. The FPFs were: visual inspection 0.09, Shewhart 0.04, Cusum 0.08. At these levels of FPF, the average time in days (range) from onset of a fault to detection was 39 (6-82) for visual inspection, 39 (4-116) for Shewhart and 21 (1-49) for Cusum. All three "phantom" methods are excellent for DXA QC, with modified Cusum analysis being the most effective. The inbuilt QA appears of little use on its own for longitudinal QC.