Scanner-induced variability and quality assurance in longitudinal dual-energy x-ray absorptiometry measurements

Characteristics of typical malfunctions and scanner-induced variability observed in dual-energy x-ray absorptiometry (DXA), and their potential effects on longitudinal reliability of DXA were evaluated. According to extensive, cumulative quality assurance (QA) data obtained from two successive x-ray sources during a 3-yr period, the scanner-induced variability may derive from long-term drift (approximately 0.5%/year), short-term drift (approximately 0.2%-2.2%/day), inhomogeneity of the x-ray beam intensity over the tabletop (approximately 1%), and changes in internal filtration (approximately 0.5%). The absolute magnitudes of these effects may be considerable with respect to expected small changes in bone characteristics observed in intervention studies. Furthermore, these effects may not be discriminated from each other. Therefore it may not be possible to correct their cumulative effect using long-term QA data only. The observed drifts are fortunately negligible with respect to the precision adequate for clinical decision making. In contrast, the evaluation of these multiform scanner-induced variability is warranted in longitudinal intervention studies using DXA. It is emphasized that this study was performed with a single DXA system, and the results should be considered accordingly; nevertheless, it is believed that the issues raised would apply to other systems too, at least in the sense of stringent QA. In this respect, single daily phantom measurement appeared to be occasionally ineffective, whereas a remeasurement of the phantom after subject measurements significantly improved the effectiveness of QA. Altogether, the QA procedures, which consider both the short-term and long-term variability as well as the spatial variability over the tabletop, may provide a more effective method to detect silently degrading scanner performance and evaluate its effects on the subject measurements. High-quality operator performance is, however, a prerequisite for proper QA in any setting.