return to HNRP

Publication Abstract Display
Type: Poster
Title: Norms and alternative cutpoints improve sensitivity/specificity tradeoff for the HIV Dementia Scale: The CHARTER study.
Authors: Sakamoto M, Marcotte TD, Umlauf A, Franklin D, Heaton RK, Ellis RJ, Alexander T, Morgan EE, Woods S, Collier A, Clifford D, Gelman B, McCutchan A, McArthur J, Morgello S, Grant I, and the CHARTER Group
Date: 02-15-2012
Abstract:Objective: The HIV Dementia Scale (HDS) is a brief screen designed to detect HIV-­related neuropsychological (NP) impairment. Utilizing a large, diverse cohort, the aims of this study were to 1) examine the classification accuracy of the HDS for NP impairment using raw and norm-­‐based cutpoints and 2) evaluate the contribution of the HDS subtests to predicting NP impairment. Participants & Methods: 1,580 HIV-­‐infected participants from 6 U.S. clinics completed the HDS and a comprehensive NP battery. Based on a gold standard of clinical ratings, 51% (n=810) of the sample was NP impaired. Results: Sensitivity and specificity of the standard raw cutpoint (≤ 10) was 27% and 92%. Using published age-­‐ and education-­‐adjusted normative standards (T<40), sensitivity significantly improved to 69% (p < .001), but specificity declined to 57%. Within a training subgroup, a cutpoint of 14 yielded optimal sensitivity (66%) and specificity (61%), which was confirmed in the validation subgroup. Individuals incorrectly classified as impaired by the raw score tended to be older, less educated, and African American. Among the subtests, attention, 4-­‐word recall, and speed significantly (p < .001) contributed to detection of NP impairment, while construction (p= .55) did not. Conclusions: HDS sensitivity to NP impairment is significantly improved by using norms. Raising the raw cutpoint for impairment to 14 also improves sensitivity, which may be more appropriate for the milder impairments in the era of effective treatments, but the use of norms, when available, is recommended to avoid demographic biases in impairment classification.

return to publications listing