Time related bias in longitudinal studies using dual photon absorptiometry

Abstract

Examining the bone mineral density (BMD’s) slope of patients regularly followed in our department, we observed recently that the group of patients who had their last BMD during the last 6 months of 1989, had a different slope than patients who had their last BMD during the following 6 months. In order to investigate if a small time-related bias of measurement, unsuspected by the former quality control investigations, could exist, we performed the following analyses. A regression equation between BMD and time was calculated and a slope was obtained for 95 women who had been followed for at least 3 yr and had had at least 3 BMD measurements during that time. The women were divided in 3 groups according to when the last BMD measurement had been performed (July–December 1989, January–June 1990 or July–December 1990). The slopes of the 3 groups of patients were compared. For each value of BMD of every patient, a predicted BMD (BMDp) was calculated using the regression equation and the relative difference (RD) between BMDp and BMD was calculated and analysed in relation to time. There was a significant difference (p<0.05) between the slopes of patients in relation to the time when the last BMD had been measured. Significant fluctuations (p<0.001) in RD were observed in relation to time. These RD variations suggested the existence of a time-related error. The presence of this error is also substantiated by the fact that a parallelism existed between the curve of the RD variations and the curve of the mean values of BMD of all patients referred to our department, calculated per period of 4 months. Although the fluctuation of the latter curve was not statistically significant. This study revealed that despite regular quality checkings, an unsuspected not random error may exist. The methodology described in this study can be used as a supplementary long-term quality assessment.