Objectives: The objectives of this study were to evaluate the magnitude and sources of site differences in a multisite study of rates of cognitive decline among patients with Alzheimer disease and to seek strategies to reduce the magnitude of site differences in this and future such studies.
Methods: A total of 3,280 participants from 15 different sites was analyzed. For each participant, the average rate of change in the Mini-Mental State Examination (MMSE) was calculated. Participants who declined at least three MMSE points per year were classified "rapid decliners." Site differences in sociodemographic distributions and the percentage of rapid decliners were examined, and a signal detection approach was used to identify the main correlates of rapid decline.
Results: The percentage of rapid decliners for the 15 sites initially varied from 8%-40%. Two of the correlates of rapid decline were largely the result of different sampling protocols, namely baseline MMSE and elapsed time between the first and last MMSE. By selecting only those participants at each site with a baseline MMSE between 15 and 23, and limiting the follow-up time to a period of 11-24 months, the authors created greater homogeneity in the protocols across sites and reduced site variability of rapid decliners from 27%-50%.
Conclusion: Results of single-site studies are often nonreproducible, and multisite studies that follow different protocols and do not take site differences into account may be misleading. This study indicates the importance of site differences and how relatively simple efforts to impose common sampling, measurement, and design criteria can reduce, if not totally remove, site differences.