Factors Associated With Incident Human Immunodeficiency Virus–Dementia FREE

FEW STUDIES have examined the risk factors for dementia in patients with human immunodeficiency virus (HIV) infection and these have focused primarily on demographic and medical factors. Previous studies have suggested that older age, history of HIV-related medical symptoms, lower hemoglobin levels, higher plasma viral load,¹ lower CD4⁺ cell counts,² and intravenous drug use³ are associated with a greater risk of developing dementia. There has been some suggestion that zidovudine treatment is protective.⁴ While many studies have demonstrated subtle cognitive changes in patients with HIV, and psychomotor slowing has been associated with an increased risk of dementia,⁵ it is unclear whether these are unrelated to later dementia or represent the early stages of a dementing process. Similarly, diagnostic criteria have been proposed for HIV-associated minor cognitive/motor disorder (MCMD), which applies to HIV-positive patients with subtle cognitive, neurologic, and psychiatric symptoms but minimal functional complaints.⁶ To our knowledge, the relationship between this diagnosis and a later diagnosis of dementia has been unexplored. This study prospectively followed up HIV-positive patients to evaluate which clinical features noted at the initial visit were associated with an increased risk of developing dementia. Potential correlates of incident dementia included demographic, medical, neurologic, functional, and psychiatric features as well as laboratory study results. In addition, all participants completed a battery of neuropsychological tests, allowing us to assess the predictive use of poor test performance and the presence of MCMD.

The Dana Consortium on the Therapy of HIV–Dementia and Related Cognitive Disorders was formed in 1994 to recruit a cohort of HIV-infected individuals and examine risk factors for human immunodeficiency virus–dementia (HIV-D). Subjects were recruited from infectious disease clinics or through targeted advertising at 3 sites: Columbia University, New York, NY; Johns Hopkins University, Baltimore, Md; and the University of Rochester, Rochester, NY. Subjects were eligible for inclusion in the cohort if they were positive for HIV, had subjective reports of memory or concentration problems, and had CD4⁺ cell counts either below 200/µL or below 300/µL, and cognitive impairment on neuropsychological testing (see "Neuropsychological Testing" subsection). They were excluded from the cohort if they were not ambulatory or if they had other neuropsychiatric conditions (eg, current or past central nervous system infection, head injury, or schizophrenia) that might cause cognitive impairment.

Assessments were performed semiannually for up to 30 months. At each visit, subjects underwent a neurologic examination; completed a battery of neuropsychological, functional, and psychiatric assessments; and had blood drawn for laboratory studies. Data collected from all of these evaluations were used to examine potential risk factors for incident HIV-D.

Age at initial visit, number of years of education, sex, race (white vs other), use of antiretroviral therapy at the initial visit, consumption of alcohol (at least weekly, less than weekly), and intravenous drug use during the year prior to enrollment (all intravenous drug users in our study had used drugs during that year) were used as independent variables.

A standardized history of HIV-related diagnoses was collected. The presence of any one diagnosis (yes, no) was used as an independent variable in our analyses. The neurologic examination was designed to capture signs associated with HIV-D. The macroneurologic examination created for the AIDS (Acquired Immunodeficiency Syndrome) Clinical Trials Group was used. The motor subscale (Part III) of the Unified Parkinson Disease Rating Scale⁷ was administered to rate the presence or absence of extrapyramidal signs. Total scores for the macroneurologic examination and the Unified Parkinson Disease Rating Scale were dichotomized at or near their median values for analysis.

The neuropsychological battery was designed to delineate HIV-D and MCMD. Where possible, tests recommended by the National Institute of Mental Health⁸ and the AIDS Clinical Trials Group were included.

The 8 tests included in the core neuropsychological battery covered 6 domains. Verbal memory was assessed with the Rey Auditory Verbal Learning Test.⁹ Visual memory was assessed with the Rey-Osterrieth Complex Figure 1 Delayed Recall Test.⁹ Constructional skills were assessed with the Rey-Osterrieth Complex Figure 1 Immediate Recall Test. Psychomotor skills were measured with the Digit Symbol Test.¹⁰ Motor skills were assessed with the Grooved Pegboard (dominant and nondominant hand)¹¹ and Timed Gait tests. Frontal systems were assessed with Verbal Fluency¹² and the Odd-Man-Out tests.^13- 14

Performance for each test was referenced to age- and education-appropriate norms. Norms for subjects with an educational level above high school were those reported for the Multicenter AIDS Cohort Study, a cohort of homosexual men¹⁵; norms for subjects with an education at the high school level or below were those reported for the ALIVE (AIDS Link to Intravenous Experience) cohort, a cohort of intravenous drug users.¹⁶ For the Odd-Man-Out Test, unpublished norms from the homosexual men and intravenous drug user cohorts followed up at Columbia University^17- 18 were used.

At the initial visit, the presence of cognitive impairment sufficient to meet study enrollment criteria was defined as performance 2 SDs below the appropriate mean on 1 test, or 1 SD below the mean on 2 tests. If Timed Gait performance was the only measure in which subjects scored 2 SDs below the mean, subjects were not considered to have met criteria for cognitive impairment.

For statistical analysis, neuropsychological test scores were dichotomized at 1 SD below the mean. For the Grooved Pegboard Test, a score greater than 1 SD below the appropriate norm for either the dominant or nondominant hand was considered to be a deficit. For Timed Gait performance, 1.5 SDs below the mean was used as the cutoff because of the nature of the available normative data.

Composite scores were created from the neuropsychological test battery with the aid of a principal component factor analysis using Varimax rotation. Three factors were identified that accounted for 68.4% of the total variance: attention-memory (Rey Auditory Verbal Learning Test—total score, trial 5 score, recall after interference, delayed recall, and correct recognition), psychomotor speed (Grooved Pegboard, dominant and nondominant hands and Digit Symbol Test), and executive function (Rey-Osterreith Complex Figure 1 Immediate Recall Test, Odd-Man-Out Test, and Verbal Fluency Test). Composite scores for each of these factors were formed by first creating a z score for each neuropsychological test variable (using the Multicenter AIDS Cohort Study and ALIVE age- and education-based norms) and then averaging the z scores across the appropriate variables. These composite scores (average z scores) were dichotomized at 0 for statistical analysis.

We also used the presence of HIV-associated MCMD at the initial study visit as an independent variable. Subjects received this diagnosis if they had at least 2 deficits in cognitive tests (Digit Symbol Test, Rey Auditory Verbal Learning Test) or neurologic examination (finger agility, alternating movement, gait and coordination, limb coordination, emotional lability) and a deficit in at least 1 role function. These are the same criteria used in the diagnostic algorithm previously published based on American Academy of Neurology guidelines,¹⁹ except in this study we did not require the role function deficit to be attributed to a cognitive source.

Functional measures were chosen that would reflect the degree to which cognitive deficits compromised everyday function. Measures were derived from the Instrumental Activities of Daily Living (IADL) scales of Lawton and Brody,²⁰ the Katz Instrumental Activities of Daily Living/Lawton Personal Self-Maintenance Scale,²¹ and the role functioning items of the Medical Outcomes Study.²² Two functional outcomes that reflect stress and stamina were also included: the Karnofsky performance scale²³ and the Medical Outcomes Study physical function subscale.²² All scores were dichotomized at or near their median values for analysis except for the Karnofsky performance scale, which was categorized as less than 80, 80 to 89, and 90 to 100. A score of less than 80 on this scale would suggest the need for assistance with some Instrumental Activities of Daily Living and an inability to work outside the home.

The 20-item Center for Epidemiologic Studies–Depression Scale²⁴ was used to assess mood. The Center for Epidemiologic Studies–Depression Scale score was considered a continuous variable in the statistical analyses.

The CD4⁺ cell counts, hemoglobin and β₂-microglobulin levels, and hematocrit were obtained. For analysis, these were treated as continuous variables. The CD4⁺ cell count and β₂-microglobulin level were transformed using the natural logarithm.

The primary end point for all analyses was the diagnosis of HIV-D. A standardized algorithm was used to diagnose HIV-D, based on data from the neurologic, neuropsychological, psychiatric, and functional assessments.¹⁹ To receive the HIV-D diagnosis, subjects had to fulfill cognitive criteria based on a set of test cutoff scores, as well to report a deficit in at least 1 of 8 Instrumental Activities of Daily Living items. In addition subjects were required to meet fixed criteria for either neurologic or neuropsychiatric problems characteristic of HIV-D.

The primary outcome variable for this investigation was the time from enrollment to the development of HIV-D, as determined at a subsequent follow-up visit. The original cohort consisted of 272 subjects; however, 71 of these were diagnosed as having HIV-D at the baseline visit. An additional 55 subjects did not have follow-up (postbaseline) assessments for HIV-D. All analyses were performed on the resulting cohort of 146 subjects. For subjects who did not develop HIV-D, the follow-up time was censored at the last available visit at which an assessment for HIV-D was performed.

The analytic strategy focused on examination of the associations between independent variables measured at the initial study visit (baseline) and the time to development of HIV-D. The Cox proportional hazards regression model was used.²⁵ An initial set of analyses was performed to examine these associations separately for each of the independent variables described earlier. A best subsets model selection procedure based on the score statistic²⁶ was used in conjunction with clinical judgment to help identify reasonable Cox proportional hazards regression models for time to development of HIV-D. Two sets of multiple regression analyses were performed: one that included all of the individual neuropsychological test scores as independent variables, and another that instead used the composite neuropsychological test scores. All Cox proportional hazards regression analyses included study site as a stratification factor.

Baseline variables were compared among subjects who did and did not reach the end point of HIV-D using t tests or χ² tests, as appropriate. A Kaplan-Meier curve was constructed for the time to dementia outcome.²⁵ All analyses were performed at the 5% level of significance (2-tailed).

The follow-up experience of the 146 subjects was as follows: 146 at 6 months, 126 at 12 months, 99 at 18 months, 58 at 24 months, and 29 at 30 months. There were 33 subjects who prematurely dropped out of the study, 23 because of death.

During the course of follow-up, 45 of the 146 subjects met criteria for incident HIV-D. Baseline characteristics of the subjects who did and did not become demented are listed in Table 1 and Table 2. Compared with the subjects who did not become demented, those who became demented were more likely to be female, nonwhite, have higher macroneurologic and motor Unified Parkinson Disease Rating Scale scores, have lower hematocrit and hemoglobin levels, and have higher β₂-microglobulin levels. They were also more likely to have poorer role function, Karnofsky performance scale²³ and physical function scores, and higher Center for Epidemiologic Studies–Depression Scale scores. A Kaplan-Meier curve summarizing time to reach the dementia end point across the entire group of patients is shown in Figure 1. In our cohort of 146 subjects, the estimated probability of incident HIV-D was approximately 25% at 12 months and 40% at 24 months.

We first examined the associations between the presence of individual features noted on the initial evaluation and incident dementia; results are summarized in Table 3 and Table 4. For individual tests in the neuropsychological battery, defective scores on Timed Gait performance, Verbal Fluency, Grooved Pegboard, and Digit Symbol tests were associated with an increased risk of incident dementia (Table 3). Also, the composite z scores for attention-memory, psychomotor speed, and executive function, as well as the diagnosis of MCMD at baseline, were associated with an increased risk of incident dementia (Table 3). Survival curves for patients with and without a diagnosis of MCMD at baseline are shown in Figure 1.

For demographic and medical variables, female sex, race (nonwhite), increased abnormalities on the neurologic examination, the presence of extrapyramidal signs, and a history of HIV-related medical symptoms were all associated with an increased risk of incident HIV-D. Increased risk was also noted in subjects with higher reported role or physical function difficulties and with higher depression scores. Finally, lower hematocrits and hemoglobin levels and higher β₂-microglobulin levels were associated with an increased risk of HIV-D (Table 4).

Best-subsets model selection was used to generate a series of multivariate models for the HIV-D end point. A series of models containing 1 to 8 independent variables were generated. The first model considered demographic and medical variables and individual neuropsychological test results as potential covariates. The selected model included depression as assessed with the Center for Epidemiologic Studies–Depression Scale, sex, Digit Symbol Test, Verbal Fluency Test, history of HIV-related diagnoses, and Timed Gait performance (Table 5). A second, similar model was created using the cognitive domain scores. The selected model included depression, sex, executive function, history of HIV-related diagnoses, Timed Gait performance, and the presence of MCMD (Table 6).

This prospective study identified a set of variables that were associated with increased risk of dementia in a cohort of HIV-positive subjects identified as having cognitive deficits and/or severe immunosuppression. In the univariate analyses, variables were identified in all of the categories evaluated. For medical variables, similar to observations in a separate cohort studied by McArthur et al,¹ an increased risk of dementia was associated with history of HIV-related medical symptoms and lower hemoglobin levels. In addition, lower hematocrit and increased β₂-microglobulin levels were associated with greater risk, consistent with previous observations linking these variables with cognitive deficits and dementia in HIV-D. The observation that abnormalities on the neurologic examination and the presence of extrapyramidal signs were associated with an increased risk is also consistent with the observation that these features are typically noted as parts of the AIDS dementia complex. It is, therefore, logical to expect that in some cases their presence would precede the actual onset of dementia. Similarly, depression is one of the defining features of the dementia complex and, thus, logically might precede the onset of the entire syndrome. Lower role and physical function scores were associated with an increased risk of dementia; this may represent the earliest manifestation of the dementing process.

Poor performance on a series of neuropsychological tests, including Timed Gait, Verbal Fluency, Grooved Pegboard, and Digit Symbol, was also associated with an increased risk of incident dementia. Similarly, poorer performance in the following cognitive domains was associated with greater risk: attention-memory, psychomotor, and executive function. In an analysis of data from the Multicenter AIDS Cohort Study, Sacktor et al⁵ reported that sustained psychomotor slowing was associated with an increased hazard of dementia, AIDS, and death. Cognitive decline in other domains was not predictive, however. Their analysis differs from ours because we considered only performance at the initial visit, as opposed to change in performance over time. In addition, the current study specifically recruited individuals with cognitive deficits or advanced immunosuppression, perhaps increasing the probability that they would become demented.

The association between cognitive performance at baseline and later dementia suggests that poorer performance could represent the earliest sign of a dementing process. However, the significance of mild cognitive changes in individuals who are HIV positive has been unclear. To our knowledge, it has not been previously established whether the presence of subtle cognitive deficits has any implications for the later development of dementia. Further, it has not been clear whether the presence of MCMD has any relation to dementia. This study suggests that at least in some cases, cognitive deficits and MCMD are related to the later advent of dementia.

Female sex may also be a risk factor for dementia. This observation is consistent with previous reports that HIV-positive women have more rapid progression of neurologic signs and symptoms.²⁷

The multivariate models presented herein were selected from several candidate models and require validation in other samples. Still, they provide some insight into the independent contribution of the variables identified in the univariate analyses. It is notable that sex, depression, Timed Gait performance, and a history of HIV-related diagnoses are included in both models. In addition, both models implicate poorer executive function as an independent risk factor for dementia. Since executive dysfunction is often one of the earliest cognitive changes noted in HIV, this suggests that there may be a common underlying mechanism between dementia and early cognitive change. Finally, in the second model presented, MCMD remained a significant risk factor for dementia, independent from depression and executive dysfunction. Thus, the presence of this syndrome seems to have unique predictive value, over and above the presence of some features that contribute to its diagnosis.

Accepted for publication July 28, 2000.

This study was supported by the Charles A. Dana Foundation, New York, NY, and grants RR00645 and 5MO1 00044 from the National Institutes of Health, Bethesda, Md.

Columbia University, New York, NY: Karen S. Marder, MD, MPH; Yaakov Stern, PhD; Steven Albert, PhD; George Todak, MSW; Ronda Clouse, RN; Carmen Polanco, MS; Richard Mayeux, MD, MS. Johns Hopkins University, Baltimore, Md: Ned Sacktor, MD; Ola A. Selnes, PhD; Deborah Hasenaue, MA; Deneen Esposito, BA; Justin McArthur, MB,BS, MPH. University of Rochester, Rochester, NY: Giovanni Schifitto, MD; Michael P. McDermott, PhD; Donna Palumbo, PhD; Connie Orme, BA; Charlyne Hickey, RN; Cindy Casaceli, MBA; Lisa Rumfola, RN; Carol Zimmerman, RN; Karl Kieburtz, MD, MPH. Northwestern University Medical School, Chicago, Ill: Leon Epstein, MD.

Corresponding author and reprints: Yaakov Stern, PhD, Sergievsky Center, 630 W 168th St, New York, NY 10032 (e-mail: ys11@columbia.edu).