Effect of regression to the mean in serial echocardiographic measurements of left ventricular mass. Quantification and clinical implications.

When a study sample is selected on the basis of an increased value of a given parameter, subsequent serial measurements are likely to show a decrease in this measured parameter, that is, the "regression to the mean." This statistical phenomenon undoubtedly affects the results of echocardiographic follow-up studies. Its magnitude is linked to that of the intraindividual variability of the measurements. Therefore, we undertook a prospective study aimed at assessing the different components of the variability of repeated measurements of left ventricular (LV) mass index. For this purpose, 25 consecutive patients underwent two echocardiograms 2 weeks apart. An analysis of variance for the nested design (random-effect model) was performed: inter- and intraindividual variabilities accounted for 70% and 30% of the total variability, respectively. Moreover, the contribution of the readings to the total variability did not exceed 2%, whereas the contribution of both recordings and time reached 28%. Further calculations allowed the expected magnitude of the regression to the mean to be quantified according to the baseline left ventricular mass index. For instance, when patients are selected on the basis of an increased LV mass index with a mean baseline value of 150 g/m2, a spontaneous decrease of 12% (related solely to the regression to the mean) has to be expected for the following measurement. Accordingly, the results of open uncontrolled echocardiographic follow-up studies should be reevaluated: an adjustment for the potential influence of the regression to the mean has to be done.