Comparison of four methods for cross‐calibrating dual‐energy x‐ray absorptiometers to eliminate systematic errors when upgrading equipment

Abstract Dual‐energy x‐ray absorptiometry (DXA) is a widely employed technique for making noninvasive measurements of bone mineral density (BMD). Advances in DXA technology have resulted in the development of new densitometers that offer increased scan speed, improved spatial resolution, and the ability to make measurements at additional skeletal sites. However, changing from a first to a second‐generation DXA system generates two additional potential sources of error. First, if the densitometers produce results that are substantially different, diagnostic errors occur if the results from both instruments are compared to the same normative database. Second, even if the densitometers produce results that are nearly identical, small systematic errors may influence interpretation of serial bone density measurements in individual patients. To assess the impact of changing from a first‐ to a second‐generation DXA scanner, we made measurements using the standard “pencil beam” mode on 133 consecutive patients using both a Hologic QDR‐1000 and a QDR‐2000 densitometer when the latter instrument was calibrated according to the manufacturer's routine procedure using a single anthropomorphic spine phantom. We then recalculated the results for the QDR‐2000 densitometer using cross‐calibration factors based on (1) a regression line generated by scanning three anthropomorphic spine phantoms whose BMD ranged from osteoporotic to high normal on each instrument, (2) an adult human lumbar spine embedded in tissue‐equivalent plastic, or (3) a regression line derived from scans of the first 83 patients that was then applied to the last 50 patients. Next, we made measurements in 200 consecutive patients with the QDR‐2000 densitometer using both the pencil beam mode and the “fan beam” mode when the instrument was calibrated according to the manufacturer's routine procedure. The results for the “fan beam” measurements of the last 100 patients were then recalculated using cross‐calibration factors based on a regression line derived from scans of the first 100 patients. Measurements of bone area, bone mineral content (BMC), and BMD on the QDR‐1000 and QDR‐2000 densitometers were highly correlated ( r ≥ 0.997). Still, there were small, significant systematic differences in bone area and BMD, so that BMD values on the QDR‐1000 were 0.007 g/cm higher ( p < 0.005). Recalculation of the QDR‐2000 pencil beam values using cross‐calibration factors generated by scanning the three anthropomorphic spine phantoms decreased, but did not eliminate, these systematic differences. Recalculation of the QDR‐2000 pencil beam values using calibration factors based on the adult human lumbar spine phantom failed to eliminate the differences in bone area and BMD and introduced a new systematic difference in BMC. In contrast, recalculation of the QDR‐2000 pencil beam values using cross‐calibration factors generated by scanning a series of patients eliminated these systematic differences. Similar results were obtained when comparing the results of scans performed with the QDR‐2000 densitometer in the pencil and fan beam modes. The residual variability around the regression line relating the QDR‐2000 pencil and fan beam measurements was twice as great as that relating the QDR‐1000 and QDR‐2000 pencil beam measurements ( p < 0.001). These results demonstrate that routine calibration methods based on phantoms are adequate for diagnosing osteopenia when changing from a first‐ to a second‐generation DXA system. However, such calibration methods may introduce systematic errors that have important effects on the interpretation of serial BMD measurements. These errors can be eliminated by calibrating the densitometers based on measurements of a large number of heterogeneous patients from a routine referral population.