Objectives: To determine whether the accuracy of the Mini-Mental State Examination (MMSE) in predicting future Alzheimer's disease (AD) could be improved by the addition of patient and informant ratings of cognitive difficulties.
Design: An inception cohort of nondemented patients followed longitudinally for 2 years.
Setting: Patients referred to a university teaching hospital research investigation by their family physicians because of concerns about memory impairment.
Participants: One hundred sixty-five community-residing patients were included who did not have dementia or any identifiable cause for memory impairment. After 2 years, 29 met criteria for AD, and 95 were not demented.
Measurements: Baseline assessments included MMSE, an Informant Rating Scale, and a Patient Rating Scale of cognitive difficulties. After 2 years, patients were diagnosed following the reference standard for probable AD. Diagnosticians were blind to baseline scores.
Results: Age and education were included in all analyses as covariates. The best logistic regression model included the Informant Rating Scale and the MMSE (sensitivity = 83%, specificity = 79%). An empirically reduced six-item model that included two items each from the MMSE, the Patient Rating Scale, and the Informant Rating Scale produced a significantly better model than the one with the full test scores (sensitivity = 90%, specificity = 94%).
Conclusion: Results indicate that inclusion of informant ratings with the MMSE significantly improved its accuracy in the prediction of probable AD. Replication in a new prospective cohort of nondemented patients is necessary to confirm these findings.