Secondary Logo

Journal Logo

Epidemiology and Social

Neuropsychological phenotypes among men with and without HIV disease in the multicenter AIDS cohort study

Molsberry, Samantha A.a; Cheng, Yub,c; Kingsley, Lawrenced,e; Jacobson, Lisaf; Levine, Andrew J.g; Martin, Eileenh; Miller, Eric N.g; Munro, Cynthia A.i,j; Ragin, Annk; Sacktor, Nedj; Becker, James T.c,l,m for the Neuropsychology Working Group of the Multicenter AIDS Cohort Study

Author Information
doi: 10.1097/QAD.0000000000001865



Since early in the HIV epidemic, it has been clear that HIV infection affects the nervous system [1]. Immediately following infection, an aseptic meningitis can develop, which can be symptomatic, and which usually clears without treatment in 4–6 weeks [2,3]. With persistent infection, particularly in the absence of effective suppression of viral replication, neurological and cognitive symptomatology can manifest. The cluster of signs and symptoms fall under the rubric of HIV-associated neurocognitive disorder (HAND) [4,5], and in the absence of effective treatment, the condition may progress to an HIV-associated dementia, which can result in death within 6 months [6]. Although the use of combination antiretroviral therapies (cARTs) has reduced the prevalence of HAD, milder forms of HAND persist [6–9]. Indeed, some form of HAND is still present in approximately 40% of HIV-infected individuals [10]. Risk factors for HAND, both host and virus-based, have been identified and include age, low educational attainment, presence of depression, hepatitis C virus (HCV) coinfection, AIDS, IDU, alcohol abuse and the presence of other severe medical comorbidities, among others [1,4,11].

In the early days of the epidemic, the neurological and neuropsychological abnormalities resulted in a relatively coherent cluster of signs and symptoms such that the milder form of dysfunction was referred to as Minor Cognitive/Motor Disorder [12,13]. Patients became slow, weak and the cognitive deficits appeared similar to those found in ‘subcortical’ dementia syndromes [13,14]. More recently, however, even with access to large pools of data, it has been more difficult to identify an HIV-related phenotype of cognitive dysfunction [15–18]. This is important because patterns of impairment should, at least to some extent, reflect the central nervous system dysfunction that is the underlying cause of the cognitive impairment. Further, understanding the cognitive/behavioural phenotype and factors that influence risk is critical for developing rational pharmacological and nonpharmacological prevention and treatment/management strategies. This investigation analysed patterns of cognitive function among a large group of gay/bisexual men in order to determine how these men cluster in patterns of cognitive functions and the factors associated with cluster membership.

The Multicenter AIDS Cohort Study (MACS) is a four-centre study of the natural and treated history of HIV infection among gay/bisexual men. The MACS has tracked cognitive test performance for over 30 years in a subcohort, including participants infected with HIV and seronegative participants at risk of infection [19]. Participants in this Neuropsychological substudy are regularly assessed with a battery tests measuring performance in each of six cognitive domains.

We chose to use model-based clustering because it is an increasingly popular approach in diverse areas, including epidemiology [20] and clinical psychology [21]. Unlike more traditional clustering techniques, model-based clustering approaches the choice of the best clustering solution as a model choice problem that can be solved with goodness-of-fit techniques. Although previous analyses have used cluster analyses to identify patterns in neuropsychological data among HIV/AIDS populations [15–17], to the best of our knowledge, this is the first time it has been done with a model-based approach. The aim of this study is to identify neuropsychological phenotypes via model-based clustering of MACS participants based on cognitive domain scores at the first neuropsychological classification and to identify factors associated with identified phenotypes.

Materials and methods


The MACS enrolled a total of 6972 men, irrespective of serostatus, from sites in Baltimore, Washington, District of Columbia, Chicago, Los Angeles and Pittsburgh at three separate time points: 4954 men enrolled in 1984–1985, 668 men enrolled in 1987–1991 and 1350 men enrolled in 2001–2003 [22,23]. MACS participants return every 6 months for an interview, physical examination and collection of blood for laboratory testing. The interview covers physical health, medical treatments and sexual and substance use disorders. A more detailed description of MACS study enrollment has been previously reported [19,24]. This study uses data from the Neuropsychological substudy of the MACS that were collected on or before 30 September 2014.

Neuropsychological evaluation

The MACS Neuropsychological evaluation includes tests from multiple cognitive domains related to the classification of HAND (Supplementary Table 1, [5,24–27]: Executive Functioning (Trails B, Stroop Interference), Speed of Information Processing (Symbol Digit Substitution Task, Stroop Color Naming), Attention and Working Memory (two indices of Choice Reaction Time), Learning (Rey Auditory Verbal Learning Test Sum of Trials 1-5, Rey Complex Figure Immediate Recall), Memory (Rey Auditory Verbal Learning Test Delayed Recall, Rey Complex Figure Delayed Recall) and Motor Speed/Coordination (Grooved Pegboard) [19]. Data from HIV seronegative participants were used to create statistical models to derive T-scores for each participant that are adjusted for age, years of education, ethnicity (white or nonwhite) and number of times the test has been administered. For each cognitive domain, a summary T-score was calculated by averaging all-available T-scores for that domain or, in the case of the Motor domain, using the lowest Grooved–Pegboard score [28]. Individuals who completed tests in at least four of the six domains were classified as cognitively normal, mildly impaired or severely impaired according to the Antinori criteria (Supplementary Table 2, [5]. Briefly, an individual was classified as Normal if one or fewer domains had T-scores 1 SD or more below the mean; Mildly Impaired if two or more domains had T-scores 1 SD or more below the mean and criteria for Severely Impaired were not fulfilled; and Severely Impaired if two or more domains had T-scores 2 SDs or more below the mean or one domain had a T-score of 2.5 SDs or more below the mean.

To be included in this study, participants needed to have a domain-specific T-score in each of the six domains at the time of their first cognitive classification. Further, men with extremely high or low scores (T-score less than 10 or greater than 90) on the motor and speed domains were truncated at 10 or 90. Out of the 6972 men in the MACS, 4318 have participated in the Neuropsychological study and completed enough testing to be cognitively classified. Of these, 2904 met the inclusion criteria. The differences between men who did and did not meet these criteria are summarized in Table 1. Briefly, the men included in these analyses are, on average, older, less likely to identify as white, have fewer years of education and lower estimated intelligence quotient (IQ), less likely to belong to the first enrollment cohort, more likely to use drugs and less likely to be classified as cognitively normal. Among those excluded from the analysis due to missing domain scores, the most commonly missing domain score was Working Memory followed by Executive Functioning, Memory, Learning, Motor and Speed, respectively. Men from the first enrollment cohort were more likely to be excluded in part because of the Reaction Time measures that contributed to the Working Memory domain and the Stroop Interference Test that contributed to the Executive Functioning and Speed of Information Processing domains that were not part of the original test battery.

Table 1:
Comparison of characteristics of excluded versus included participants.


Prior to implementing the latent profile analysis (LPA), we ran a variable selection procedure on the six domain scores in order to identify those variables that were most likely to make a significant contribution to the model. Including unnecessary variables can reduce model fit and/or make it difficult to identify a reasonable clustering solution. We utilized a ‘greedy search’ algorithm [29] that screens the variables to identify those that most improve the clustering solution as measured by the Bayesian Information Criterion (BIC) [20,30]. In a stepwise manner, it determines which of the clustering variables can be dropped from the model, and stops when there is no change in the clustering solution.

After the variable selection step, we implemented a LPA using the domain-specific T-scores from the time of participants’ first cognitive classification. The use of model-based clustering approaches is driven by the idea that the observed data from a population actually arose from two or more subpopulations. Finite mixture models estimate the overall population as a weighted combination of these subpopulations. In a Gaussian finite mixture model, we assume that each subpopulation is normally distributed and the overall population is a combination of these distributions. Bayes’ rule can be used to identify the subpopulations within the overall population by assigning each observation to the cluster for which it has a maximum posterior probability. This is done using the Expectation-Maximization algorithm to maximize the likelihood function. The BIC is used to select the appropriate number of clusters such that the model that corresponds to the largest BIC value has the best fit [20,30].

Following the LPA, chi-square tests for independence and analyses of variance (ANOVAs) were conducted to examine differences in demographic, HIV/AIDS, behavioural, educational and emotional factors of interest across cluster. For those factors that were found to be significantly associated with profile membership, posthoc analyses were conducted to determine which pairwise difference (s) were driving the effects.


The variable selection procedure [29] identified five (out of six) domain T scores that we used in subsequent analyses. Only the Working Memory domain score was excluded because it did not contribute significantly to the clustering solution.

The Mclust package [31,32] in R[33] was used to conduct the LPA. The five domain-specific T-scores from the first Neuropsychological cognitive classification visit were entered. The best fitting model was a three-cluster solution with variable distribution, variable volume, equal shape and orientation, a variable distribution, variable volume, and equal shape and orientation (VVE) solution and the solution had a BIC value of -100 650.2. The next best-solution was a four-cluster VVE solution with a BIC value of -100 700.1. The difference between these two models in BIC was 49.93, which is positive support for the better model [21]. All other solutions corresponded to smaller BIC solutions than the four-cluster VVE solution.

The best fitting model produced clusters with 578 (19.9%), 1012 (34.8%) and 1314 (45.2%) participants, respectively. The distribution of domain-specific T-scores is summarized in Table 2 and Fig. 1.

Table 2:
Characteristics of individual profiles.
Fig. 1:
The pattern of domain score distributions for each of the three latent profiles.The horizontal line within each box indicates the median score for the domain, while the hinges of each box indicate the scores corresponding to the 25th and 75th percentiles, respectively. The range of values within the whiskers represent scores that fall within 1.5*IQR of the nearest hinge.

In a normative sample, the cognitive domain T-score distribution has a mean of 50 with a standard deviation of 10. Scores of 45–55 are considered normal; 40–44 are borderline; 35–39 are Mildly Impaired; 30–34 are Mild/Moderately Impaired; 25–29 are Moderately Impaired; 20–24 are Moderately/Severely Impaired; a score less than 20 is Severely Impaired.

Participants in Profile 1 performed below average on all domains except for Motor skills where the average score was in the impaired range. For Profile 2, performance was within the normal range (at or slightly below a T score of 50) for all of the individual domain scores except for Memory and Learning, which were below the normal range with average scores of 43.6 and 43.8, respectively. By contrast, Profile 3 showed at or above average performance in all domains.

The participants classified in Profile 1 performed more poorly on measures of Motor speed than their other four domain scores, and had significantly lower motor speed than participants classified in Profiles 2 or 3. This group was the oldest, reported more symptoms of depression, had the lowest estimated IQ (based on the Shipley Institute of Living Scale Vocabulary Subtest) and had completed the fewest number of years of education (note, however, that the average was more than 2 years of college). Fifty-one percent of the participants classified in Profile 1 were drawn from the 2001–2003 enrollment cohort, which is the highest of the three profiles. In addition, only 56% of the individuals in this profile self-identified as white, which is significantly lower than the proportions in the other two profiles. Approximately half of the participants in Profile 1 were HIV-infected. Only 37% of the individuals are classified as cognitively normal, and 20% were classified as severely impaired [both proportions differ significantly from those seen in the other two profiles (for details see Table 2)].

For the individuals classified in Profile 2, their performance on measures of Learning and of Memory were lower than those for the other three domains and were significantly lower than the Learning and Memory domain scores seen in Profile 3 (but not Profile 1). On average, their age was midway between that of Profiles 1 and 3, as well as the number of items endorsed on the Center for Epidemiological Studies - Depression (CES-D). Their estimated IQ and years of education were also midway between the other two profiles. Seventy-four percent of the individuals in this group were classified as cognitively normal, and only 2% were classified as severely impaired. Like Profile 1, approximately 50% of individuals were HIV-infected. Only 36% of these participants were drawn from the 2001–2003 cohort, and 72% of the participants classified themselves as white.

Finally, Profile 3 represents what might be the ‘healthiest’ of the three profiles. Mean performance on all five domains was at or above normal, with Memory and Learning being one standard deviation unit higher than that seen in Profiles 1 and 2. This group of individuals was younger, reported fewer symptoms of depression, had the highest estimated IQ and the greatest number of years of education. Ninety-five percent of the individuals in this profile were classified as cognitively normal, and none were classified as severely impaired. Fifty-three percent of the sample were HIV-infected such as Profiles 1 and 2, 36% of the sample were drawn from Cohort Three and 73% self-identified as white.

Among the HIV-infected individuals only, there was a significant difference in nadir CD4+ cell counts, such that the individuals in Profile 3, the highest performing profile, had the lowest levels. In multinomial logistic regression models of profile membership, however, nadir CD4+ cell count no longer significantly contributed to the model after adjustment for recruitment cohort in the main analysis (Likelihood Ratio Test: Chi-square = 1.824, P = 0.402) or analyses restricted to HIV-infected individuals (Likelihood Ratio Test: Chi-square = 4.891, P = 0.087). The mean log10 viral load ranged from 3.05 to 3.19, with no significant difference among the three profiles. At first cognitive classification, 61% of the individuals in Profile 1 had used cART, whereas 70% of the individuals in Profiles 2 and 3 had used these medication regimens. The proportion of individuals who met criteria for AIDS ranged from 4 to 7%, and was not different across profiles.

Posthoc pairwise comparisons adjusted for multiple comparisons between the three profiles reveal that the significance of the associations between covariates and profile membership described above are generally driven by the differences between Profile 1 and the other two profiles (Table 3). Profile 1 was significantly different from Profiles 2 and 3 with respect to age, CES-D, IQ, HAART use, enrollment cohort membership, race and cognitive classification. Profile 1 was different from Profile 3 but not Profile 2 in terms of years of education, pack-years of smoking and alcohol use. Profile 2 was significantly different from Profile 3 with respect to age, IQ, smoking and cognitive classification. None of the pairwise comparisons between profiles were significant with respect to HIV status, CD4+ cell count or viral load.

Table 3:
Posthoc pairwise analyses.


There are three main findings from these analyses. First, there are three major groups of participants that cluster together based on performance on standard neuropsychological measures of Executive Functioning, Speed of Information Processing, Learning, Memory and Motor speed/coordination. Second, a variety of factors differed among the three profiles, including age, education, race, enrollment cohort and mood-related symptoms. Third, there were no differences among the clusters in terms of rate of HIV-infection, rate of AIDS diagnosis or markers of immunocompetence or viral replication.

The analyses described here support three distinct latent profiles of neuropsychological test performance, which were determined on the basis of cognitive domain scores in Learning, Memory, Speed, Motor Skills and Executive Functioning at the time of first cognitive classification. The results indicate that approximately 20% of participants identify most closely with Profile 1, 35% identify most closely with Profile 2 and 45% identify most closely with Profile 3. Profile 1 is characterized by below average or impaired functioning across domains, Profile 2 is characterized by average to slightly below average functioning across domains and Profile 3 is characterized by average or above average functioning across domains.

One of the key strengths of this study is that it was conducted using model-based clustering rather than more traditional hierarchical or k-means based clustering. Model-based clustering is advantageous in that it allows the data to determine the optimal number of clusters, and therefore is less subject to a priori assumptions about the latent structure. In addition, model-based clustering assigns membership weights for each profile rather than strictly assigning each subject to a specific profile, which is more flexible and allows participants to be ideal members of a single profile or mixed members of multiple profiles.

Another strength of this study is the utilization of data from the MACS. The MACS is a large, longitudinal cohort study with rich data that has enrolled multiple subcohorts over time. Because of these characteristics, we are able to differentiate effects of HIV-related factors from effects of demographic and other factors that have shifted among the HIV-infected population over the course of the HIV epidemic.

A key limitation of this study is that those who met inclusion criteria were significantly different than those who did not with respect to many factors, including age, race, years of education, illicit drug use, enrollment cohort and neurocognitive test performance. The differences between the included versus excluded individuals are likely a reflection of the differences between the enrollment cohorts. Due to strong competing risks during the initial epidemic and the timing of the initiation of the Neuropsychological substudy, fewer participants in the first enrollment cohort were able to participate in the Neuropsychological substudy. As previously described [19], the first enrollment cohort differs greatly from the second enrollment cohort with respect to many of the same factors that differ between individuals included versus excluded from this study.

Another potential limitation of this study is that the participants are exclusively MSM and, as such, their primary risk factor for becoming HIV-infected is through unprotected sexual intercourse with other men. If either sex or the route of infection modifies the effect of HIV on neurocognitive performance, this would limit the generalizability of these results. As our cohorts consist solely of MSM, we are unable to explore the extent to which these results would differ in other at-risk populations.

In spite of attempts, we were unable to develop a model that accurately predicted profile membership. We believe this was largely due to a limitation of the data, as different variables began to be collected at different time points over the course of the MACS such that many variables are missing at the time of first neurocognitive classification for participants enrolled in the earlier subcohorts. For instance, although variables known to be related to cognitive performance, such as markers of cardiometabolic health, may have improved the predictive performance of our model, we had insufficient data at the time of the relevant neurocognitive classification to consider these factors. This limitation reinforces the need to consistently collect information beyond the variables traditionally considered in NeuroHIV work, as these alone appear to be inadequate for predicting cognitive performance.

The analyses described here are not dissimilar to the results of the mixed membership trajectory model (MMTM) previously described by our group [34], which identified three canonical trajectories describing the probabilities of normal, mildly impaired and severely impaired cognition over time in the MACS. Figure 2 displays the proportion of participants identifying with each MMTM trajectory in each LPA profile among the nearly 2700 participants included in both analyses. Approximately half of participants who most closely identify with the ‘unhealthy’ MMTM trajectory also most closely identify with LPA Profile 1, the profile with the lowest average cognitive domain scores. Similarly, participants who most closely identify with the ‘healthy’ or ‘premature aging’ MMTM trajectories are most likely to identify with LPA Profile 3 followed by LPA Profile 2, the highest and second highest performing LPA profiles, respectively. The similarities between the two different methodologies suggest some degree of convergent validity.

Fig. 2:
The proportion of individuals identifying most closely with each MMTM canonical trajectory among those who identify most closely with a given latent profile.The bars for each respective MMTM trajectory sum to 1.

Several cluster analyses using neuropsychological test performance have been conducted among other cohorts that included HIV-infected individuals [15–17]. These analyses all use different neuropsychological batteries and different cluster analysis techniques. As such, the different studies are not directly comparable, but overall, this body of literature generally supports the idea of a small number of latent neuropsychological performance profiles among people living with HIV. The analyses reported here are the first to indicate that such profile membership may be driven by demographic, mood and lifestyle characteristics rather than factors related to HIV infection.

These data reinforce the importance for clinicians to consider all possible causes of cognitive impairment in their HIV-infected patients. Although these results do not show that HIV has no impact on cognitive function, they add to the growing body of literature that shows that other factors, some related to HIV disease (or treatment) and others not, may be more important than the infection in terms of driving cognitive impairment. Self-reported mood state, use of cART and predispositional risk factors should be considered in designing treatment plans.


Data in this manuscript were collected by the MACS with centres at Baltimore (U01-AI35042): The Johns Hopkins University Bloomberg School of Public Health: Joseph B. Margolick (PI), Jay Bream, Todd Brown, Adrian Dobs, Michelle Estrella, W. David Hardy, Lisette Johnson-Hill, Sean Leng, Anne Monroe, Cynthia Munro, Michael W. Plankey, Wendy Post, Ned Sacktor, Jennifer Schrack, Chloe Thio; Chicago (U01-AI35039): Feinberg School of Medicine, Northwestern University, and Cook County Bureau of Health Services: Steven M. Wolinsky (PI), Sheila Badri, Dana Gabuzda, Frank J. Palella, Jr., Sudhir Penugonda, John P. Phair, Susheel Reddy, Matthew Stephens, Linda Teplin; Los Angeles (U01-AI35040): University of California, UCLA Schools of Public Health and Medicine: Roger Detels (PI), Otoniel Martínez-Maza (PI), Peter Anton, Robert Bolan, Elizabeth Breen, Anthony Butch, Shehnaz Hussain, Beth Jamieson, John Oishi, Harry Vinters, Dorothy Wiley, Mallory Witt, Otto Yang, Stephen Young, Zuo Feng Zhang; Pittsburgh (U01-AI35041): University of Pittsburgh, Graduate School of Public Health: Charles R. Rinaldo (PI), James T. Becker, Phalguni Gupta, Kenneth Ho, Lawrence A. Kingsley, Susan Koletar, Jeremy J. Martinson, John W. Mellors, Anthony J. Silvestre, Ronald D. Stall; Data Coordinating Center (UM1-AI35043): The Johns Hopkins University Bloomberg School of Public Health: Lisa P. Jacobson (PI), Gypsyamber D'Souza (PI), Alison Abraham, Keri Althoff, Michael Collaco, Priya Duggal, Sabina Haberlen, Eithne Keelaghan, Heather McKay, Alvaro Muñoz, Derek Ng, Anne Rostich, Eric C. Seaberg, Sol Su, Pamela Surkan, Nicholas Wada. Institute of Allergy and Infectious Diseases: Robin E. Huebner; National Cancer Institute: Geraldina Dominguez. The MACS is funded primarily by the National Institute of Allergy and Infectious Diseases (NIAID), with additional cofunding from the National Cancer Institute (NCI), the National Institute on Drug Abuse (NIDA) and the National Institute of Mental Health (NIMH). Targeted supplemental funding for specific projects was also provided by the National Heart, Lung, and Blood Institute (NHLBI), and the National Institute on Deafness and Communication Disorders (NIDCD). MACS data collection is also supported by UL1-TR001079 (JHU ICTR) from the National Center for Advancing Translational Sciences (NCATS) a component of the National Institutes of Health (NIH), and NIH Roadmap for Medical Research. The contents of this publication are solely the responsibility of the authors and do not represent the official views of the NIH, Johns Hopkins ICTR or NCATS. The MACS website is located at

The members of the Neuropsychology Working Group include J.T.B., Pim Brouwers, Christopher Cox, Jenna Fahey, Rebecca Godfrey, K.G., Robin Huebner, A.J.L., E.M., Donna M. Martineck, Eric N. Miller, Ann Ragin, S.R., JoanaDarc Roe, Ned Sacktor, Janet Schollenberger, Eric Seaberg and Matthew Wright.

S.A.M. did the conceptualization or design of the study, analysis and interpretation of data, drafting the manuscript; Y.C. did the conceptualization or design of the study, analysis and interpretation of data, revising the manuscript; L.K. did the acquisition of study data, analysis and interpretation of data, revising the manuscript; L.J. revised the manuscript; A.J.L., E.M., E.N.M., C.M., A.R., N.S. did the acquisition of study data, revising the manuscript; J.T.B. did the acquisition of study data, conceptualization or design of the study, analysis and interpretation of data, revising the manuscript.

Additional support for the analysis of these data and preparation of the manuscript was provided by funds from the NIH to J.T.B. (AG034852 and MH098745).

E.N.M. is the author of the reaction time software used in this study (CalCAP) and has a financial interest in the software.

Conflicts of interest

There are no conflicts of interest.


1. Clifford DB, Ances BM. HIV-associated neurocognitive disorder. Lancet Infect Dis 2013; 13:976–986.
2. Michaels J, Sharer LR, Epstein LG. Human immunodeficiency virus type 1 (HIV-1) infection of the nervous system: a review. Immunodefic Rev 1988; 1:71–104.
3. Schacker T, Collier AC, Hughes J, Shea T, Corey L. Clinical and epidemiologic features of primary HIV infection. Ann Intern Med 1996; 125:257–264.
4. Letendre S. Central nervous system complications in HIV disease: HIV-associated neurocognitive disorder. Top Antivir Med 2011; 19:137–142.
5. Antinori A, Arendt G, Becker JT, Brew BJ, Byrd DA, Cherner M, et al. Updated research nosology for HIV-associated neurocognitive disorders. Neurology 2007; 69:1789–1799.
6. Liner KJ 2nd, Ro MJ, Robertson KR. HIV, antiretroviral therapies, and the brain. Curr HIV/AIDS Rep 2010; 7:85–91.
7. Al-Khindi T, Zakzanis KK, van Gorp WG. Does antiretroviral therapy improve HIV-associated cognitive impairment? A quantitative review of the literature. J Int Neuropsychol Soc 2011; 17:956–969.
8. Joska JA, Gouse H, Paul RH, Stein DJ, Flisher AJ. Does highly active antiretroviral therapy improve neurocognitive function? A systematic review. J Neurovirol 2010; 16:101–114.
9. Spudich S. HIV and neurocognitive dysfunction. Curr HIV/AIDS Rep 2013; 10:235–243.
10. Heaton RK, Franklin DR, Ellis RJ, McCutchan JA, Letendre SL, Leblanc S, et al. HIV-associated neurocognitive disorders before and during the era of combination antiretroviral therapy: differences in rates, nature, and predictors. J Neurovirol 2011; 17:3–16.
11. De Ronchi D, Faranca I, Berardi D, Scudellari P, Borderi M, Manfredi R, et al. Risk factors for cognitive impairment in HIV-1-infected persons with different risk behaviors. Arch Neurol 2002; 59:812–818.
12. Woods SP, Moore DJ, Weber E, Grant I. Cognitive neuropsychology of HIV-associated neurocognitive disorders. Neuropsychol Rev 2009; 19:152–168.
13. McArthur JC, Brew BJ, Nath A. Neurological complications of HIV infection. Lancet Neurol 2005; 4:543–555.
14. Ances BM, Ellis RJ. Dementia and neurocognitive disorders due to HIV-1 infection. Semin Neurol 2007; 27:86–92.
15. Dawes S, Suarez P, Casey CY, Cherner M, Marcotte TD, Letendre S, et al. Variable patterns of neuropsychological performance in HIV-1 infection. J Clin Exp Neuropsychol 2008; 30:613–626.
16. Fazeli PL, Crowe M, Ross LA, Wadley V, Ball K, Vance DE. Cognitive functioning in adults aging with HIV: a cross-sectional analysis of cognitive subtypes and influential factors. J Clin Res HIV AIDS Prev 2014; 1:155–169.
17. Lojek E, Bornstein RA. The stability of neurocognitive patterns in HIV infected men: classification considerations. J Clin Exp Neuropsychol 2005; 27:665–682.
18. Saylor D, Dickens AM, Sacktor N, Haughey N, Slusher B, Pletnikov M, et al. HIV-associated neurocognitive disorder: pathogenesis and prospects for treatment. Nat Rev Neurol 2016; 12:234–248.
19. Becker JT, Kingsley LA, Molsberry S, Reynolds S, Aronow A, Levine AJ, et al. Cohort Profile: recruitment cohorts in the neuropsychological substudy of the Multicenter AIDS Cohort Study. Int J Epidemiol 2015; 44:1506–1516.
20. Flynt A, Daepp MI. Diet-related chronic disease in the northeastern United States: a model-based clustering approach. Int J Health Geogr 2015; 14:25.
21. Suveg C, Jacob ML, Whitehead M, Jones A, Kingery JN. A model-based cluster analysis of social experiences in clinically anxious youth: links to emotional functioning. Anxiety Stress Coping 2014; 27:494–508.
22. Kaslow RA, Ostrow DG, Detels R, Phair JP, Polk BF, Rinaldo CR Jr. The Multicenter AIDS Cohort Study: rationale, organization, and selected characteristics of the participants. Am J Epidemiol 1987; 126:310–318.
23. Kingsley LA, Detels R, Kaslow R, Polk BF, Rinaldo CR Jr, Chmiel J, et al. Risk factors for seroconversion to human immunodeficiency virus among male homosexuals. Results from the Multicenter AIDS Cohort Study. Lancet 1987; 1:345–349.
24. Miller EN, Selnes OA, McArthur JC, Satz P, Becker JT, Cohen BA, et al. Neuropsychological performance in HIV-1-infected homosexual men: the Multicenter AIDS Cohort Study (MACS). Neurology 1990; 40:197–203.
25. Miller EN, Satz P, Visscher B. Computerized and conventional neuropsychological assessment of HIV-1-infected homosexual men. Neurology 1991; 41:1608–1616.
26. Selnes OA, Miller EN. Grant I, Martin A. Development of a screening battery for HIV-related cognitive impairment. Neuropsychology of HIV infection. New York: Oxford University Press; 1994. 176–190.
27. Woods SP, Rippeth JD, Frol AB, Levy JK, Ryan E, Soukup VM, et al. Interrater reliability of clinical ratings and neurocognitive diagnoses in HIV. J Clin Exp Neuropsychol 2004; 26:759–778.
28. Becker JT, Martinson JJ, Penugonda S, Kingsley L, Molsberry S, Reynolds S, et al. No association between Apoepsilon4 alleles, HIV infection, age, neuropsychological outcome, or death. J Neurovirol 2015; 21:24–31.
29. Scrucca L, Raftery A. clustvarsel: a package implementing variable selection for model-based clustering in R. ArXiv e-prints. 2014.
30. Scrucca L, Fop M, Murphy TB, Raftery AE. mclust 5: clustering, classification and density estimation using Gaussian finite mixture models. R J 2016; 8:289–317.
31. Fraley C, Raftery A, Murphy T, Scrucca L. mclust Version 4 for R: normal mixture modeling for model-based clustering, classification, and density estimation technical report no. 597. Department of Statistics, University of Washington; 2012.
32. Fraley C, Raftery A. Model-based clustering, discriminant analysis and density estimation. J Am Stat Assoc 2002; 97:611–631.
33. Team RC. R: a language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2016.
34. Molsberry SA, Lecci F, Kingsley L, Junker B, Reynolds S, Goodkin K, et al. Mixed membership trajectory models of cognitive impairment in the multicenter AIDS cohort study. AIDS 2015; 29:713–721.

cognition; epidemiology; HIV; latent profile analysis; model-based clustering

Supplemental Digital Content

Copyright © 2018 Wolters Kluwer Health, Inc.