P2‐128: Cardiovascular factors and white matter lesions affect on extrapyramidal signs in Alzheimer's disease

(CA) was applied to the HIS data on all 2,968 subjects to generate optimally scaled item responses and patient scores. CA was performed separately on different configurations of the HIS_1) the original 13 items, subsets of items, and subsets using composite items. We used the optimal or raw HIS item scores plus covariates of age, sex, education, history of hypertension and diabetes to predict VD vs. non-VD diagnosis. Using bootstrap sampling, we generated 100 data sets of an equal number of randomly sampled VD and non-VD subjects, and then performed logistic regression on each data set to obtain stable parameter estimates. Performance statistics were determined using non-parametric receiver operating characteristic curve methods. Results: Two dimensions of the optimally scaled HIS item responses, and none of the covariates, significantly contributed to VD/non-VD prediction. Compared to the original, 13-item HIS instrument scoring, an optimally scored, composite item model consisting of 5 questions was more accurate (96.5 vs. 97.3%), more sensitive (91.7% vs. 92.2%) and more specific (89.6% vs. 92.1%). The best, optimally scored subset of 6 HIS items performed comparable to the original HIS instrument. Conclusions: 1) The HIS instrument can be substantially shortened and improved by optimal scaling and by creating composite items. 2) The HIS instrument measures two dimensions of information. 3) The composite HIS item model does not require a neurological exam to accurately classify VD and non-VD.