Interobserver disagreements on clinical dementia rating assessment: interpretation and implications for training

The Clinical Dementia Rating (CDR) is a widely used semiobjective instrument for staging dementia severity. A global CDR score is reported that is derived from individual scores in six domains. In this study, we examined both agreement and disagreement, among raters and with a gold standard, to identify domain-specific and global dementia severity level ratings that would most benefit from further training or greater emphasis in future training. We found that raters-in-training experienced the most difficulty with rating normal and questionable dementia. They also had the most trouble scoring the memory domain. When they disagreed with the gold standard, they nearly always gave higher ratings. A third, extremely experienced group of raters were uniform in their high levels of agreement on each domain and the global CDR and tended to give lower ratings if they disagreed with the gold standard. Analysis of the agreement and disagreement patterns suggested that greater emphasis on the memory, home and hobbies, and orientation domains during CDR training, and increasing the information provided for the judgment and problem solving domain on the standardized CDR worksheets, could improve the consistency of raters and increase the efficiency with which they are trained to use the CDR.