Concurrent classification accuracy of the HIV dementia scale for HIV-associated neurocognitive disorders in the CHARTER cohort

Background: The HIV Dementia Scale (HDS) was developed to screen for HIV-associated neurocognitive disorders (HAND), but concerns have persisted regarding its substandard sensitivity. This study aimed to examine the classification accuracy of the HDS using raw and norm-based cut points and to evaluate the contribution of the HDS subtests to predicting HAND. Methods: A total of 1580 HIV-infected participants from 6 US sites completed the HDS, and a gold standard neuropsychological battery, on which 51% of participants were impaired. Results: Sensitivity and specificity to HAND using the standard raw HDS cut point were 24% and 92%, respectively. The raw HDS subtests of attention, recall, and psychomotor speed significantly contributed to classification of HAND, whereas visuomotor construction contributed the least. A modified raw cut point of 14 yielded sensitivity of 66% and specificity of 61%, with cross-validation. Using norms also significantly improved sensitivity to 69% with a concomitant reduction of specificity to 56%, whereas the positive predictive value declined from 75% to 62% and negative predictive value improved from 54% to 64%. The HDS showed similarly modest rates of sensitivity and specificity among subpopulations of individuals with minimal comorbidity and successful viral suppression. Conclusions: Findings indicate that while the HDS is a statistically significant predictor of HAND, particularly when adjusted for demographic factors, its relatively low diagnostic classification accuracy continues to hinder its clinical utility. A raw cut point of 14 greatly improved the sensitivity of the previously established raw cut score, but may be subject to ceiling effects, particularly on repeat assessments.