Diagnosing dementia: interrater reliability assessment and accuracy of the NINCDS/ADRDA criteria versus CERAD histopathological criteria for Alzheimer's disease

We investigated the interrater reliability and accuracy of two independent medical doctors in using NINCDS/ADRDA criteria to classify 82 elderly subjects enrolled in OPTIMA, a longitudinal study investigating dementia. Kappa statistics revealed moderate agreement (0.5) in overall classification of dementia type, and almost perfect agreement (0.9) on the absence or presence of dementia. Combining NINCDS/ADRDA 'possible' and 'probable' Alzheimer's disease (AD) categories produced substantial agreement (0.7). Comparison with CERAD histopathological criteria for AD showed that combining 'possible' and 'probable' AD resulted in a high sensitivity and accuracy, but a low specificity. To increase specificity, the NINCDS/ADRDA 'probable AD' category should be used alone. An important finding was that the accuracy of diagnoses of AD made from the case notes alone was not different from the diagnoses obtained following active involvement with participants.