The most commonly referenced HIV screening tool is the HIV Dementia Scale (HDS) [18
]. This tool provides a rapid assessment of eye movements, motor skills, simple learning, and attention. The HDS performance characteristics are modest with acceptable sensitivity only for the most severe disease. For example, among hospitalized patients, 12% of whom met clinical criteria for HAD, the sensitivity at a cut-point of 10 (out of 16) was 0.92 and the specificity was 0.71, resulting in a positive predictive value (PPV) of 0.3 and a negative predictive value (NPV) of 0.98 [19
]. The performance characteristics are much worse for detecting milder impairment [20
]. A study of HIV-positive individuals interested in returning to work noted sensitivity and specificity for all impairment to be only 39% and 85%, respectively [22
Adjusting for age and education substantially improves performance but simultaneously complicates interpretation for primary care settings. Even with adjustment, the HDS demonstrates only modest ability to identify impairment (70% sensitivity) [23
]. A modified version (mHDS) excludes the antisaccade eye movement task, which may be the most sensitive aspect of the HDS. The mHDS was no better than a simple test of psychomotor speed and manual dexterity (the grooved pegboard test) in identifying dementia [24
]. Substantial educational attainment effects have also been noted [25
]. Taken together, these data are concerning that the HDS is suboptimal for most HAND patients given that HAD constitutes <5% of impaired cases among community-dwelling adults.
Reliance on patient self-report of symptoms is not ideal. Studies have revealed limited correlation between self-reported memory impairments and performance on objective neuropsychological tests, and tighter association with depressive symptoms. The Patient Assessment of Own Functioning (PAOF), for example, identified depressive symptoms and psychomotor inefficiency rather than neuropsychological testing performance, and when applied to HIV-negative substance abusers, there was no correlation to neuropsychological testing performance [26
]. Importantly, screening for symptoms would not identify ANI, the largest subset of impaired participants. It is possible that, combined with a clinical risk identification schema that includes risk factors such as CD4 nadir count, plasma viral load, and age, these metrics could be improved, given reports that such clinical variables may have some utility [28
Computer-delivered cognitive assessments have a theoretical benefit in that they alleviate the time burden on clinical staff. In the pre-cART era, the Sequential and Choice Reaction Time Program (CALCAP), a measure of reaction time, attention, psychomotor speed, and memory, identified advanced disease but not milder disease [29
]. More recently, the CogState was evaluated and similarly performed well in identifying advanced dementia [31
]. A pilot study revealed some promise in using the Computer-based Assessment of Mild Cognitive Impairment (CAMCI); however, the study design and size limited interpretation for the clinical setting [32
]. In general, computer approaches have several important limitations including an inability to test verbal learning efficiency, complicated outputs requiring interpretation, cost burdens, limitations in the face of literacy and non-English speakers, and a need to train staff.
Understanding which subtests within comprehensive batteries best correlate to cognitive impairment could provide pivotal information for screening tool development. A domestic study noted that a combination of tests tapping verbal memory (Hopkins Verbal Learning Test—revised total recall) and psychomotor speed (Grooved Pegboard or digit symbol modalities) emerged as best predictors of rater-determined impairment [33
]. The combined tests outperformed the HDS. A second international trial noted that verbal learning efficiency (WHO-UCLA auditory verbal learning sum of trials 1–5), psychomotor speed (digit symbol modalities test) and motor speed (timed gait) together best distinguished HAD from non-HAD cases among cART-naive advanced HIV-positive participants [34
]. However, the time needed to perform the list-learning task (verbal learning efficiency) in each study effectively precludes its use in simple screening instruments. A shorter battery that was limited to tests of psychomotor speed (digit symbol, Trails A) and cognitive flexibility (Trails B) generally performed worse in accurately categorizing impairment (60% of cases when less stringent cut-points were used) [35
There are considerable differences between screening tools for HIV and those with utility for other disorders encountered in older age. The most widely used screening tool for AD, the Mini Mental State Exam (MMSE), does not test executive function or motor skill, rendering it less sensitive to subcortical neuropathology, and limited investigations confirm poor sensitivity to HAND [36
]. Among the first 75 cases enrolled into the UCSF Memory and Aging HIV Cohort (all with age >60 years), the MMSE did not appear to provide sufficient variability by cognitive group to help diagnostically [mean score out of 30 possible points (sample size, standard deviation) were as follows: normal cognition: 29.3 (37, 0.75); ANI: 28.7 (13, 1.49); MND: 28.3 (20, 1.26) and HAD: 23 (4, 11.37) (unpublished data). Notably, the HAD group included 1 patient thought to have comorbid advanced AD and an MMSE score of 6]. Overall, the narrow distribution of these scores demonstrates poor likelihood of utility even in aged HIV-positive participants.
In summary, studies related to screening have broad limitations, are often underpowered, and attempt to identify only the most severe form of impairment (). Nearly all have been completed in younger populations; it is anticipated that the performance characteristics would be even less favorable in older participants. There are insufficient data to make firm recommendation on optimal screening tools, but it appears clear that the MMSE is not a good choice.
Screening Tools for Human Immunodeficiency Virus–Associated Neurocognitive Disorders