p(C) Analysis Facilitates Dementia Diagnosis

Abstract

A modified receiver operating characteristic (ROC) analysis technique was applied to a sample of 161 consecutive volunteers seen in a dementia clinic. Clinical, imaging, neuro- psychological, and laboratory evaluation guided experienced clinicians in clinical diagnosis, taken as the "gold standard." Two symptom inventories, the Hachinski Ischemic Score and the Dementia of the Alzheimer's Type Inventory, were obtained by clinicians who were blind to final clinical diagnosis; scores on these inventories correlate with the likelihoods of multi- infarct dementia and Alzheimer's disease, respectively. A disjunctive sequential testing strat egy was analyzed such that subthreshold scores on the first test identified patients for whom the second test was considered. Both tests were analyzed at all possible cutoff-point com binations and in both possible testing sequences. Diagnoses based on these tests were compared with the clinical "gold standard" diagnoses to determine the accuracy of the testing procedures. The best strategy correctly classified 154/161 (95.6%) of the dementia patients and required cutoff points (5 for the HIS and 10 for the Dementia of the Alzheimer's Type Inventory) that were lower than those usually recommended for either test used alone (i.e., 7 and 14, respectively). The Hachinski Ischemic Score-then Dementia of the Alzheimer's Type Inventory testing sequence was superior to the reverse strategy. A sensitivity analysis (varying prevalences of Alzheimer's disease, multi-infarct dementia, and other dementias) revealed similar test performances across a wide range of prevalences. These data suggest that simple clinical tests that take approximately 30 minutes to administer can produce diagnostic classifications of dementia that are similar to those of clinicians experienced in dementia diagnosis. Key words: dementia; multi-infarct; Alzheimer's disease; diagnostic tests; ROC analysis. (Med Decis Making 1994;14:393-402)