Abstract
Objective
The longitudinal trajectories that individuals may take from a state of normal cognition to HIV-associated dementia are unknown. We applied a novel statistical methodology to identify trajectories to cognitive impairment, and factors that affected the “closeness” of an individual to one of the canonical trajectories.
Design
The Multicenter AIDS Cohort Study (MACS) is a four-site longitudinal study of the natural and treated history of HIV Disease among gay and bisexual men.
Methods
Using data from 3,892 men (both HIV-infected and uninfected) enrolled in the neuropsychology substudy of the MACS, a Mixed Membership Trajectory Model (MMTM) was applied to capture the pathways from normal cognitive function to mild impairment to severe impairment. MMTMs allow the data to identify canonical pathways and to model the effects of risk factors on an individual’s “closeness” to these trajectories.
Results
We identified three distinct trajectories to cognitive impairment – one “normal aging” (low probability of mild impairment until age 60), one “premature aging” (mild impairment starting at age 45–50), and one “unhealthy” (mild impairment in 20s and 30s) profile. Second, clinically defined AIDS and not simply HIV Disease, was associated with closeness to the premature aging trajectory. And, third, Hepatitis-C infection, Depression, Race, Recruitment Cohort and Confounding Conditions all affected individual’s closeness to these trajectories.
Conclusions
These results provide new insight into the natural history of cognitive dysfunction in HIV disease and provide evidence for a potential difference in the pathophysiology of the development of cognitive impairment based on trajectories to impairment.
Keywords: HIV, Dementia, HIV Associated Neurocognitive Disorder, Epidemiology
Introduction
HIV infection affects the nervous system [1], frequently resulting in neurobehavioral disturbances and HIV-associated neurocognitive disorder (HAND)[2, 3]. The advent of combination antiretroviral therapy (cART) has reduced the prevalence of HIV-Associated Dementia, but not of the milder syndromes[4–7]. HAND is still present in approximately 40% of HIV-infected individuals [8], and a number of risk factors have been identified. These can be host or HIV-disease based, and include older age, less education, the presence of depression, hepatitis C virus (HCV) infection, AIDS, and the presence of other severe medical comorbidities[1, 2, 9].
The Multicenter AIDS Cohort Study (MACS) is a four-center study of the natural and treated history of HIV infection among gay/bisexual men. The MACS has tracked cognitive test performance for 25 years in a subcohort of men who formed the neuropsychological substudy[10]. Logistic regression, survival analysis, and similar methods have been used to determine how various risk factors influence the development of cognitive impairment in HIV disease[11–13], but they do not consider the possibility of multiple temporal patterns or trajectories to cognitive dysfunction.
In order to investigate the possibility of the existence of multiple patterns, or “trajectories” to cognitive impairment, we adapted the novel, data-driven Mixed Membership Trajectory Model (MMTM) technique [14] to fit these data. This technique has been used to model trajectories toward physical disability [15] as well as pathways to dementia in the elderly [16, 17]. MMTMs combine features of longitudinal Multivariate Latent Trajectory Models [18] to identify distinct, canonical profiles, with features of cross-sectional Grade of Membership Models[19] to allow individuals to have weighted memberships in each profiles.
An advantage of the MMTMs relative to other trajectory modeling techniques is that the MMTM also expresses each individual participant’s pathway as a weighted combination of the canonical trajectories. In addition to expressing an individual’s closeness to the canonical trajectories (or profiles), the membership weights can also be interpreted as reflecting each individual’s health propensities (i.e., disease and death).
The primary purpose of this study was to identify trajectories to impairment in a large cohort of gay/bisexual men with as much as 25 years of follow-up, and to determine the effects of HIV infection, AIDS, HCV infection, race, education and medical comorbidities on the closeness of each individual to the canonical trajectories.
Methods
This research was reviewed and approved by the Institutional Review Boards at each MACS site. Each participant signed a written statement of informed consent prior to starting any research-related activities.
Subjects
The MACS enrolled a total of 6972 men from sites in Baltimore, Washington, 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 (see [20, 21]). The men enrolled in 1984–1985 and 1987–1991 are considered cohort 1 (C1), and the men enrolled 2001–2003 are considered cohort 2 (C2). Details of the study enrollment have been reported previously [10] [22]. This study uses data from the NP study collected on or before March 31, 2012 [10].
Cognitive Classification
The NP evaluation includes measures from multiple cognitive domains related to classification of HAND[3, 22–25] including: executive functioning, speed of information processing, attention and working memory, learning, memory, and motor [10]. In order to classify a participant as normal, mildly or severely impaired, they needed to have completed at least one test in four of the six domains. If that criterion was met, a summary T-score was calculated for each domain; if only one test in a domain was completed, the T-score for that test was used as the Domain Score. For domains with two test scores, we averaged the T-scores to create the Domain Score. However, in the case of the Motor domain, we used the lowest T-score on either the dominant or nondominant hand of the Grooved Pegboard. If the T-score for one of the Grooved Pegboard measures was ≥ 40 and the other one was <40 then the Motor domain was assigned the lowest obtained T-score plus five.
Using the Antinori criteria [3], cognitive classifications were: 1) Normal, if one or fewer domains had T-scores 1 SD or more below the mean (i.e., T ≤ 40); 2) Mild Impairment, if two or more domains had T-scores 1 SD or more below the mean, and the individual did not meet criteria for the more severe category that follows; and 3) Severe Cognitive Impairment, if two or more domains had T-scores 2 SDs or more below the mean (i.e., T ≤ 30), or one domain had a T-score 2.5 SDs or more below the mean (i.e., T ≤ 25).
To be included in the MMTM, participants needed to have at least one cognitive classification. Although the MMTM technique is able to process time varying covariates, it is difficult to produce visual representations of trajectories that include multiple time-varying factors. Therefore, in this initial implementation of MMTMs in HIV disease, we limited our sample to men who did not become infected during the follow-up period, and age was the only time-varying factor. Of the 6,972 men in the MACS, 3,892 met these criteria. The differences between the men who did and did not meet these criteria are summarized in Supplemental Table E-1. Fifty-four percent (2099/3892) of the men included in this study were HIV infected at their first cognitive classification. Table 1 presents the characteristics of the participants by serostatus. Cognitive classifications were made at multiple time points (ages) for each individual. There were a total of 25,471 observations for the 3,892 participants, an average of 6.54 observations per individual.
Table 1.
Characteristics of Study Participants at Time of First Cognitive Classification
| Viral Status | HIV− | HIV+ | Effect Size |
|---|---|---|---|
| N= | 1793 | 2099 | |
| Cohort (C1/C2)1 | 71/29 (1275/518) | 69/31 (1451/648) | 0.022 |
| Years between Baseline Visit & First Cognitive Classification1 | 5.56 (7.1) | 3.59 (4.3) | −0.340* |
| Age1 | 40.96 (10) | 38.20 (8.0) | −0.299* |
| Education Years1 | 15.64 (2.7) | 14.94 (2.7) | −0.258* |
| Race (White)1 | 77.47 (1389) | 71.32 (1497) | −0.006 |
| Marijuana2 | 36.90 (655) | 45.44 (942) | 0.086* |
| Cocaine2 | 14.42 (256) | 21.08 (437) | 0.086* |
| IVD**2 | 2.95 (52) | 4.82 (98) | 0.048* |
| Opiates2 | 2.54 (45) | 2.17 (45) | 0.103* |
| CD4+ Count1 | n/a | 483.76 (280) | |
| Standardized Log10 Viral Load1 | n/a | 3.49 (1.4) | |
| AIDS2 | n/a | 6.10 (128) | |
| Cognitive Classification (none/mild/severe)2 | 78/18/4 (1392/327/74) | 67/18/5 (1629/368/102) | 0.019 |
| Combination ART2*** | n/a | 20.96 (440) | N/A |
| Speed of Processing3 | 50.5 (9.0) | 49.8 (9.1) | .06 |
| Executive Functions3 | 49.9 (9.4) | 49.9 (9.5) | .01 |
| Working Memory3 | 49.4 (9.3) | 49.2 (9.0) | .02 |
| Memory3 | 49.6 (9.4) | 49.9 (9.6) | −.03 |
| Learning3 | 49.5 (9.1) | 49.5 (9.3) | −.004 |
| Motor Function3 | 46.9 (10.4) | 46.5 (10.5) | .04 |
These characteristics are presented as Mean (SD), and their effect size is reported as Cohen’s d statistic.
These characteristics are presented as Percent (N) ‘Yes’, and their effect size is reported with the phi statistic.
Age and education adjusted T-scores (mean ± standard deviation); their effect size is reported as Cohen’s d statistic.
p<0.05
IVD = Intravenous drug use.
Antiretroviral Therapy
Trajectory Modeling
We assumed the existence of a small number of “canonical profiles” to describe distinct patterns in the development of cognitive outcomes over time (see Discussion for information regarding profile selection). The purpose of these analyses is to identify from the data, trajectories or pathways that individuals may take to one of the three cognitive outcomes (i.e., normal, mildly impaired, severely impaired) over time (and as they age). The MMTMs differ from other trajectory models in that each individual is not forced to follow a single trajectory exactly, but rather shares some properties of each trajectory to varying degrees. Thus, at the end of the analysis, we have membership weights for each individual for each trajectory.
At each study visit there is a cognitive classification that is coded for each individual: Normal, Mildly Impaired, Severely Impaired. For each of the three of the profiles, the probability of each diagnosis as a function of age is modeled as a multinomial ordered logistic regression [26] (See Eq. 1, Supplemental Materials), allowing the computation of the probability of each cognitive classification given age.
MMTMs are multilevel models which means that they offer insight both at a general level, and at the level of the individual pathway characterizing an individual participant. The trajectory of each individual can be modeled by the MMTM as a mixture of the canonical trajectories (See Eq. 3, Supplemental Materials). Because the membership weights capture general health status of each individual, it is reasonable to assume that particular outcomes, such as mental state or death [27, 28], are conditionally independent of one another, given these health status weights. This is analogous to the conditional independence of multiple indicators (observed scores, data), given factor scores, in typical factor analytic methodologies.
We tested the effects of seven time invariant, binary predictor variable: Cohort (C1), Race (White), Confounding Conditions (Ever) [29], Hepatitis C Infection (Ever), Depression (Ever), HIV (Present), and AIDS (Present Ever). AIDS was defined as occurring when HIV seropositive men developed an AIDS-defined illness (i.e., ignoring CD4+ cell count). We selected these seven variables in this first implementation of an MMTM either because we knew they were significant predictors within the MACS cohort (e.g., recruitment cohort, medical comorbidities), because they had been identified as risk factors by others (e.g., HCV, depression), or because they were related to HIV disease (i.e., infection status and AIDS).
In the MMTM, each of these predictor variables has a coefficient to describe its impact on the membership weight of each canonical profile. For example, Cohort (X1) will have a coefficient, α11, to describe the impact of Cohort on membership for canonical profile 1, a coefficient, α12, to describe the impact of cohort on membership for canonical profile 2, and coefficient, α13, to describe the impact of cohort on membership for canonical profile 3. The relative effect of a given predictor on the odds of an individual identifying more closely with one profile than the other, is described by the difference between these coefficients.
In addition to modeling the probabilities of the cognitive outcomes in each of the men while they are in the study, we also modeled the probability of death. This was important because we needed to account for censoring by death in order to obtain accurate estimates of the probabilities of mild and severe cognitive impairment at later ages, when these cognitive outcomes might be masked by (premature) death. For a more detailed description, see the Supplemental materials.
Results
There are three main findings from this study. First, we identified three distinct trajectories to cognitive impairment – one “normal aging”, one “premature aging”, and one “unhealthy” profile. Second, clinically defined AIDS and not simply HIV Disease, was associated with closeness to the premature aging trajectory. Third, HCV, Depression, Race, Cohort and Confounding Conditions all affected individual’s closeness to these trajectories.
Figure 1 shows the trajectories of the three canonical profiles: the probability of severe impairment as a function of age is shown by the red curves, the probability of mild impairment is shown by the black curves, and the probability of normal cognition is shown by the green curves. It is important to note that these curves should not be viewed or interpreted as if they were survival functions. These trajectories are the cross-sectional probabilities of the three states (normal, mildly and severely impaired) at each age; thus, at any given age, the sum of the three probabilities is equal to 1.00.
Figure 1.
The three canonical profiles with pointwise posterior 95% credible bands for each cognitive classification. The X-axis represents assembled cross-sectional probabilities of the three states (normal, mildly and severely impaired) across time/age. At any given age, the sum of the three probabilities is equal to 1.00. The Y-axis represents the age of the men in the cohort at the time of the examination.
A) The trajectories of the “normal aging profile” for the probability of normal cognition (green), mildly impaired cognition (black), and severely impaired cognition (red). B) The trajectories of the “premature aging profile” for the probability of normal cognition (green), mildly impaired cognition (black), and severely impaired cognition (red). C) The trajectories of the “unhealthy profile” for the probability of normal cognition (green), mildly impaired cognition (black), and severely impaired cognition (red).
We arbitrarily refer to canonical profile 1, for which the probability of normal cognition is initially very high, as the “normal aging” profile; profile 2, for which the probability of mild impairment begins to climb at age 45–50 years, as the “premature aging” profile; and profile 3, for which the probability of normal cognition is near zero even at the youngest age, as the “unhealthy” profile.
For the normal aging profile, the probability of cognitive impairment is virtually zero until the age of 60. Thereafter, the likelihood of mild impairment increases, while that of normal cognition necessarily decreases. At age 80 the individuals close to the canonical profile have approximately equal likelihoods of having normal cognition or being mildly impaired. The likelihood of a severe impairment remains close to zero, reflecting, in part, the relatively small proportion of the sample in the 70+ year age range. Approximately 60% of the study sample is closest to this canonical profile.
The “premature aging” profile is similar to the “normal aging” profile, except that it is offset to the left by 25+ years. Individuals close to this profile (approximately 21% of the sample) have a approximately equal probability of normal cognition and mild impairment at age 50–55, and by age 80 the likelihood is greatest that these individuals will have a severe cognitive impairment.
Individuals closest to the “unhealthy” profile (approximately 19% of the sample) are either mildly or severely impaired across the entire age range. Even those individuals in their 20s have >80% likelihood of a mild impairment; by age 70 the probability of mild vs. severe impairment is approximately 50%.
The parameters of the MMTM, including the significant time invariant covariates, as well as the survival component are presented in Table 2. Cohort, AIDS, depression, HCV infection, confounding conditions, and the cohort by race interaction significantly affected closeness to the canonical profiles. Of particular importance were the findings related to HIV disease, AIDS, and HCV infection. The effect of AIDS on closeness to the canonical trajectories was the largest of all of the time invariant covariates that were tested. While infection with HIV moved individuals towards the premature aging profile, those individuals with AIDS moved away from the healthy profile towards the unhealthy, and also towards the premature aging profile. Infection with hepatitis C tended to move individuals away from the normal aging profile and towards the premature aging and unhealthy profiles. The effect of having been recruited into C2 was to move individuals away from the healthy profile.
Table 2.
Parameters for the MMTM with all of the tested covariates included.
| Canonical Trajectory Parameters | Estimate (SD) | |||
|---|---|---|---|---|
| Healthy | Premature Aging | Unhealthy | ||
| Intercept | B0* | 13.629 (5.816) | 5.500 (0.608) | 1.061 (0.044) |
| Effect of Age | B1* | −0.189 (0.027) | −0.155 (0.016) | −0.039 (0.004) |
| Mild/Severe Impairment Threshold | c* | 6.540 (3.882) | 3.985 (0.540) | 28.872 (18.931) |
| Covariate | Estimate [95% CI] for difference between Healthy and Unhealthy Profiles1 | Estimate [95% CI] for difference between Healthy and Premature Aging Profiles2 | Estimate [95% CI] for difference between Unhealthy and Premature Aging Profiles3 |
|---|---|---|---|
| Cohort | −0.805 (−1.131,−0.484)* | 0.738 (0.317,1.269)* | 1.588 (1.124, 2.127)* |
| Race (White) | −0.046 (−0.364, 0.277) | −0.318 (−0.688, 0.096) | −0.272 (−0.644, 0.171) |
| Confounding Conditions | −0.154 (−0.030, −0.014)* | 0.592 (0.390, 0.798)* | 0.746 (0.533, 0.959)* |
| Hepatitis C Virus Infection | −0.321 (−0.575, −0.058)* | −0.344 (−0.656,−0.020)* | −0.024 (−0.346, 0.287) |
| Depression | −0.433 (−0.606, −0.262)* | 0.196 (−0.032, 0.402) | 0.629 (0.365, 0.874)* |
| HIV | −1.267 (−1.569, −0.977)* | −3.795 (−4.176, −3.458)* | −2.529 (−2.963, −2.173)* |
| AIDS | −0.194 (−0.351, −0.034)* | −2.57 (−2.863,−2.253)* | −2.370 (−2.733, −1.994)* |
| HIV & Race Interaction | 0.549 (0.167, 0.943)* | 0.523 (−0.149, 1.119) | −0.026 (−0.744, 0.597) |
a positive value increases the odds of being closer to the Healthy Profile
a positive value increases the odds of being closer to the Healthy Profile
a positive value increases the odds of being closer to the Unhealthy Profile
In Table 3, we break down the sample into three groups, each representing those individuals who were closest to one of the three canonical profiles. That is, individuals listed under the “normal aging” profile were closest to the normal aging profile compared to either the premature aging or unhealthy profiles. The table then shows the percent of individuals within each of these broadly defined groups who had the condition represented by each covariate. That is, 34.9 percent of the individuals closest to the normal aging profile were HIV-infected. We did not test to determine whether there was a significant difference in the distribution of the covariates across the three canonical profiles because the variables used in Table 3 actually occur twice in the table – once in the MMTM (generating the column labels), and again in the table itself (as row variables).
Table 3.
Distribution of Risk Factors Across the Three Trajectories
| Normal Aging | Premature Aging | Unhealthy | |
|---|---|---|---|
| Number | 2028 | 704 | 630 |
| HIV Infected2 | 34.6 | 59.6 | 41.0 |
| AIDS | 3.8 | 39.7 | 11.9 |
| HCV Infected | 6.7 | 11.9 | 13.4 |
| Race (White) | 73.9 | 86.7 | 60.7 |
| Depressed | 72.1 | 71.4 | 81.2 |
| Medical Confounds | 31.6 | 25.4 | 39.8 |
| Cohort 1 | 65.7 | 95.3 | 53.8 |
Participants are classified based on profile with the largest membership weight gik
All numbers are percentages of the men closest to the trajectory who are positive for the risk factor (e.g., 34.6% of the men in closest to the “Normal Aging” trajectory are HIV-infected).
In order to demonstrate the effect of the individual covariates on closeness to the canonical profiles, we identified five study participants who had different combinations of the covariates. Figure 2 shows the trajectories for these individuals relative to the canonical trajectories generated by the MMTM. The top panel shows the trajectories for Mild impairment; the bottom panel for Severe impairment. With an increasing number of covariates the individual profiles move away from the normal aging profile, through the premature aging profile, towards the unhealthy profile.
Figure 2.
The effects of individual covariates on individual trajectories to mild impairment (A) and severe impairment (B). Case 1 has none of the covariates of interest, Case 2 has only HCV infection, Case 3 has HCV and HIV infection, Case 4 has HCV, HIV, and AIDS, and Case 5 has HCV, HIV, AIDS, and confounding medical conditions.
Discussion
We have identified, for the first time, three distinct, canonical profiles to cognitive impairment in HIV disease, and tested the effects of seven factors on the “closeness” of individuals to each of these profiles. The trajectories to cognitive impairment included: “normal aging”, “premature aging”, and an “unhealthy” profile. HIV, AIDS, HCV, cohort, confounding conditions, depression, and the interaction between cohort and race have a significant effect on the closeness of individuals to these profiles. These results provide new insights into the natural and treated history of HIV/AIDS by demonstrating that illness progression does not follow a single course. These results demonstrate that within the data there is evidence of three distinct pathways that individuals within this large group of men, both with and without HIV disease, can develop cognitive impairment. These results differ from traditional survival models, which assume the existence of only a single linear pathway, and from other trajectory models in that individuals’ trajectories are weighted combinations of the “canonical” trajectories. The analysis of the model covariates reveals how they affect individuals’ closeness to each of the three canonical profiles. Thus, these models provide a unique way of examining longitudinal data, and eventually may allow us to identify biologically meaningful subgroups of individuals based on the trajectories and the covariates, potentially affecting treatment and management.
Our finding of a “premature” aging trajectory is relevant to the ongoing discussions regarding the possibility that infection with HIV results in an accelerated or premature aging (see [30–32] for reviews). HIV disease moved individuals away from the healthy and unhealthy profiles, and towards the premature aging profile. A similar pattern was seen for AIDS, meaning that those HIV-infected individuals who ever met criteria for clinical AIDS also moved towards the premature aging profile. It is important to emphasize that while this profile is certainly consistent with what has been described as “premature aging” both in terms of HIV disease but also other conditions (e.g., [33]), we did not directly test “aging” biomarkers such as telomere length or immune senescence.
One of the key strengths of the MMTM is that individuals are assumed to have weighted membership in each of the canonical profiles. It is possible for an individual to be an ideal member of one profile or to be a mixed member of all three profiles, allowing individuals to display characteristics of each profile to varying degrees. This can be seen not only in Figure 2, but also in Supplemental Figure 1 that presents 100 (total) random individuals’ trajectories to each cognition classification. This ability of individuals to have mixed memberships in the trajectories is important because there is variability in the prevalence and importance of risk factors, which directly influences membership as a function of age.
Under the MMTM, the covariates that we tested here should not be viewed in the traditional sense of affecting the risk of a certain cognitive classification, but rather as affecting the “closeness” of an individual’s profile to a canonical profile. For example, the presence of a confounding condition does not increase an individual’s risk for severe cognitive impairment, but rather increases the likelihood that that individual will identify more closely with the unhealthy profile than the healthy profile. This is a critical distinction for understanding the power of the MMTM as compared to other methods of analyzing epidemiologic data.
To decide how many trajectories to include in the final model, we explored both two and three trajectory models and were able to obtain estimates from both. We then compared the prediction accuracy of the two models using the method of posterior predictive checking (see Chapter 24 of [34]), replicating the original classifications to obtain 1,000 different simulated datasets. The models performed equally well with regard to overall accuracy (See Supplemental Materials), but the three trajectory model was easier to interpret, leading us to present those findings here. The model presented here should not be thought of as a “true” model, but rather the best solution for a three profile model. If we had more participants at the upper and lower ends of the age distribution, this could easily result in a change in the number and shape of the canonical profiles.
Individuals identifying more closely with the premature aging profile are also more likely to die sooner (See Supplemental Figure E-2), and individuals with HIV disease and those with AIDS are likely to be closer to this profile than the other two. This observation adds to our confidence in the interpretability and validity of these trajectories in that HIV-infected individuals with severe cognitive impairment have a higher risk of death [28, 35]. Additionally, the finding that cohort, HCV infection, confounding conditions, and depression, and AIDS affect the closeness of individuals to the normal aging profile (i.e., generally away from that profile) may be a direct reflection of the potential for each these conditions to negatively affect brain health.
The covariates in the model do not fully explain trajectory membership, suggesting that other factors may also influence individual trajectories. These may include ApoE genotype, recreational drug use, and access and adherence to combination antiretroviral therapies (cART). Unfortunately, genetic data were not available from a large enough number of individuals to include ApoeE genotype as a predictor in the model. On the other hand, information on the use of both recreational drugs and cART are time varying predictors, and we had decided to delay including such variables in this first use the MMTM framework in the context of HIV Disease. Nevertheless, these results emphasize the value of MMTMs as a novel, analytic tool, and demonstrate their ability to identify patterns within large data sets that might have otherwise gone unnoticed.
There are several limitations to our study; first, MMTMs cannot easily account for time-varying covariates, such as incident infection or disease (or CD4+ cell counts), an obvious limitation in the study of HIV disease. Thus, additional work needs to be done in order to develop computationally efficient methodologies for the incorporation of time-varying covariates into the models, and to allow for the visualization of the effects of these covariates. Second, our data are restricted to gay and bisexual men evaluated as part of the MACS; we did not evaluate trajectories among women or among younger substance users. For the MMTM to be most useful, longer observation periods are essential, and while the Women’s Intra-agency HIV Study (http://statepiaps.jhsph.edu/wihs/) has similar biological and medical comorbidity information, it has only limited neuropsychological evaluations. Therefore, at the present time, large, long-term follow-up studies that could be merged with this data set are not available.
In conclusion, the MMTM technique uses patterns within the data to identify distinct, canonical profiles; the results are not constrained by any a priori assumptions about the trajectories. These techniques are highly innovative in that they are able to: 1) account for “reversing” states – i.e., moving from HAD to mild impairment and even normal cognition, as might be expected following viral suppression; 2) express each individual’s pathway as a weighted combination of the canonical trajectories; and, 3) determine the extent to which risk factors for cognitive impairment affect the “closeness” of an individual to each of the canonical trajectories. This tool has the potential for meaningful application in a variety of domains related to HIV disease, and holds the promise to reveal important data-driven insights into the natural and treated history of HIV disease.
Supplementary Material
Acknowledgments
Funding: AI35042, AI35039, AI35040, AI35041, AI35043, TR000424, AG034852, MH098745
The work described in this report was submitted by S.A.M. in partial fulfilment of the requirements for a Bachelor of Science degree in the Dietrich College of Arts and Science of the University of Pittsburgh; she is now in the Epidemiology program at the School of Public Health, Harvard University. We are grateful to Mikhail Popov, M.S.P. for his help in implementing the analysis software. In addition, we thank Christopher Cox, Ph.D. for insightful and helpful comments on an earlier draft of the manuscript.
Data in this manuscript were collected by the Multicenter AIDS Cohort Study (MACS) with centers at Baltimore (U01-AI35042): The Johns Hopkins University Bloomberg School of Public Health: Joseph B. Margolick (PI), Barbara Crain, Adrian Dobs, Homayoon Farzadegan, Joel Gallant, Lisette Johnson-Hill, Cynthia Munro, Michael W. Plankey, Ned Sacktor, Ola Selnes, James Shepard, Chloe Thio; Chicago (U01-AI35039): Feinberg School of Medicine, Northwestern University, and Cook County Bureau of Health Services: Steven M. Wolinsky (PI), John P. Phair, Sheila Badri, Maurice O’Gorman, David Ostrow, Frank Palella, Ann Ragin; Los Angeles (U01-AI35040): University of California, UCLA Schools of Public Health and Medicine: Roger Detels (PI), Otoniel Martínez-Maza (Co-P I), Aaron Aronow, Robert Bolan, Elizabeth Breen, Anthony Butch, Beth Jamieson, Eric N. Miller, 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), Lawrence A. Kingsley (Co-PI), James T. Becker, Ross D. Cranston, Jeremy J. Martinson, John W. Mellors, Anthony J. Silvestre, Ronald D. Stall; and the Data Coordinating Center (UM1-AI35043): The Johns Hopkins University Bloomberg School of Public Health: Lisa P. Jacobson (PI), Alvaro Munoz (Co-PI), Alison Abraham, Keri Althoff, Christopher Cox, Jennifer Deal, Gypsyamber D’Souza, Priya Duggal, Janet Schollenberger, Eric C. Seaberg, Sol Su, Pamela Surkan. The MACS is funded primarily by the National Institute of Allergy and Infectious Diseases (NIAID), with additional co-funding from the National Cancer Institute (NCI). 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-TR000424 (JHU CTSA). Website located at http://www.statepi.jhsph.edu/macs/macs.html. The contents of this publication are solely the responsibility of the authors and do not represent the official views of the National Institutes of Health (NIH).
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).
Footnotes
Disclosure: Eric N. Miller is the author of the reaction time software used in this study (CalCAP) and has a financial interest in the software.
The members of the Neuropsychology Working Group include James T. Becker, Pim Brouwers, Christopher Cox, Jenna Fahey, Rebecca Godfrey, Karl Goodkin, Robin Huebner, Andrew J. Levine, Eileen M. Martin, Donna M. Martineck, Eric M. Miller, Ann Ragin, Sandra Reynolds, JoanaDarc Roe, Ned Sacktor, Janet Schollenberger, Eric Seaberg, and Matthew Wright.
Contributions of Authors
| Samantha A. Molsberry | conceptualization or design of the study, analysis and interpretation of data, drafting the manuscript |
| Fabrizio Lecci | analysis and interpretation of data, revising the manuscript |
| Lawrence Kingsley | acquisition of study data, analysis and interpretation of data, revising the manuscript |
| Brian Junker | conceptualization or design of the study, interpretation of data, revising the manuscript |
| Sandra Reynolds | revising the manuscript |
| Karl Goodkin | revising the manuscript |
| Andrew J. Levine | acquisition of study data, revising the manuscript |
| Eileen Martin | acquisition of study data, revising the manuscript |
| Eric N. Miller | acquisition of study data, revising the manuscript |
| Cynthia A. Munro | acquisition of study data, revising the manuscript |
| Ann Ragin | acquisition of study data, revising the manuscript |
| Ned Sacktor | acquisition of study data, revising the manuscript |
| James T. Becker | acquisition of study data, conceptualization or design of the study, analysis and interpretation of data, revising the manuscript |
References
- 1.Clifford DB, Ances BM. HIV-associated neurocognitive disorder. The Lancet infectious diseases. 2013;13:976–986. doi: 10.1016/S1473-3099(13)70269-X. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Letendre S. Central nervous system complications in HIV disease: HIV-associated neurocognitive disorder. Topics in antiviral medicine. 2011;19:137–142. [PMC free article] [PubMed] [Google Scholar]
- 3.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. doi: 10.1212/01.WNL.0000287431.88658.8b. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Liner KJ, 2nd, Ro MJ, Robertson KR. HIV, antiretroviral therapies, and the brain. Current HIV/AIDS reports. 2010;7:85–91. doi: 10.1007/s11904-010-0042-8. [DOI] [PubMed] [Google Scholar]
- 5.Al-Khindi T, Zakzanis KK, van Gorp WG. Does antiretroviral therapy improve HIV-associated cognitive impairment? A quantitative review of the literature. Journal of the International Neuropsychological Society : JINS. 2011;17:956–969. doi: 10.1017/S1355617711000968. [DOI] [PubMed] [Google Scholar]
- 6.Joska JA, Gouse H, Paul RH, Stein DJ, Flisher AJ. Does highly active antiretroviral therapy improve neurocognitive function? A systematic review. Journal of neurovirology. 2010;16:101–114. doi: 10.3109/13550281003682513. [DOI] [PubMed] [Google Scholar]
- 7.Spudich S. HIV and neurocognitive dysfunction. Current HIV/AIDS reports. 2013;10:235–243. doi: 10.1007/s11904-013-0171-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.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. doi: 10.1007/s13365-010-0006-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.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. Archives of neurology. 2002;59:812–818. doi: 10.1001/archneur.59.5.812. [DOI] [PubMed] [Google Scholar]
- 10.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. International journal of epidemiology. 2014 doi: 10.1093/ije/dyu092. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Vivithanaporn P, Heo G, Gamble J, Krentz HB, Hoke A, Gill MJ, et al. Neurologic disease burden in treated HIV/AIDS predicts survival: a population-based study. Neurology. 2010;75:1150–1158. doi: 10.1212/WNL.0b013e3181f4d5bb. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Vivithanaporn P, Nelles K, DeBlock L, Newman SC, Gill MJ, Power C. Hepatitis C virus co-infection increases neurocognitive impairment severity and risk of death in treated HIV/AIDS. Journal of the neurological sciences. 2012;312:45–51. doi: 10.1016/j.jns.2011.08.025. [DOI] [PubMed] [Google Scholar]
- 13.McCutchan JA, Marquie-Beck JA, Fitzsimons CA, Letendre SL, Ellis RJ, Heaton RK, et al. Role of obesity, metabolic variables, and diabetes in HIV-associated neurocognitive disorder. Neurology. 2012;78:485–492. doi: 10.1212/WNL.0b013e3182478d64. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Manrique-Vallier D, Fienberg S. PhD Dissertation. Carnegie-Mellon University Department of Statistics; Pittsburgh PA: 2010. Longitudinal Mixed Membership Models With Applications. [Google Scholar]
- 15.Manrique-Vallier D, Fienberg S. Longitudinal Mixed Membership Models for Disability Survey Data. Proceedings of the XXVth International Symposium on Methodology Issues, Statistics Canada; 2010. [Google Scholar]
- 16.Lecci F, Junker B, Kuller LH, Lopez OL, Becker JT. Empirically Derived Trajectories to Dementia Over 15 Years of Follow-Up Identified Using Mixed Membership Models. doi: 10.1093/aje/kwv051. Under Editorial Review. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Lecci F. An Analysis of Development of Dementia through the Extended Trajectory Grade of Membership Model. In: Airoldi EM, Blei DE, et al., editors. Handbook of Mixed Membership Models. Fienberg SE: Chapman & Hall; In Press. [Google Scholar]
- 18.Connor JT. Multivariate Mixture Models to Describe Longitudinal Patterns of Frailty in American Seniors. Pittsburgh, PA: Carnegie Mellon University, ProQuest, UMI Dissertations Publishing; 2006. p. 3275170. [Google Scholar]
- 19.Erosheva E, Fienberg S, Joutard C. Describing disability through individual-level mixture models for multivariate binary data. Annals of Applied Statistics. 2007;1:502–537. doi: 10.1214/07-aoas126. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Kaslow RA, Ostrow DG, Detels R, Phair JP, Polk BF, Rinaldo CR. The Multicenter AIDS Cohort Study (MACS): Rationale, organization, and selected characteristics of the participants. American Journal of Epidemiology. 1987;126:310–318. doi: 10.1093/aje/126.2.310. [DOI] [PubMed] [Google Scholar]
- 21.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. doi: 10.1016/s0140-6736(87)91725-9. [DOI] [PubMed] [Google Scholar]
- 22.Miller EN, Selnes OA, McArthur MB. Neuropsychological test performance in HIV1-infected homosexual men: The Multicenter AIDS Cohort Study (MACS) Neurology. 1990;40:197–203. doi: 10.1212/wnl.40.2.197. [DOI] [PubMed] [Google Scholar]
- 23.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. doi: 10.1080/13803390490509565. [DOI] [PubMed] [Google Scholar]
- 24.Miller EN, Satz P, Visscher B. Computerized and conventional neuropsychological assessment of HIV1-infected homosexual men. Neurology. 1991;41:1608–1616. doi: 10.1212/wnl.41.10.1608. [DOI] [PubMed] [Google Scholar]
- 25.Selnes OA, Miller EN. Development of a screening battery for HIV-related cognitive impairment: The MACS experience. In: Grant I, Martin A, editors. Neuropsychology of HIV Infection. New York: Oxford University Press; 1994. pp. 176–190. [Google Scholar]
- 26.Gelman A, Hill J. Data Analysis Using Regression And Multilevel/Hierarchical Models. Cambridge University Press; 2007. [Google Scholar]
- 27.Sevigny JJ, Albert SM, McDermott MP, Schifitto G, McArthur JC, Sacktor N, et al. Am evaluation of neurocognitive status and markers of immune activation as predictors of time to death in advanced HIV infection. Arch Neurol. 2007;64:97–102. doi: 10.1001/archneur.64.1.97. [DOI] [PubMed] [Google Scholar]
- 28.McArthur JC, Hoover DR, Bacellar H, Miller EN, Cohen BA, Becker JT, et al. Dementia in AIDS patients: incidence and risk factors. Neurology. 1993;43:2245–2253. doi: 10.1212/wnl.43.11.2245. [DOI] [PubMed] [Google Scholar]
- 29.Goodkin K, Miller E, Cox C, Reynolds S, Becker J, Martin E, et al. Acceleration of aging effects on neurocognitive function in HIV infection: Evidence from the Multi-Center AIDS Cohort Study. doi: 10.1016/S2352-3018(17)30098-X. Under editorial review. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Watkins CC, Treisman GJ. Neuropsychiatric complications of aging with HIV. Journal of neurovirology. 2012;18:277–290. doi: 10.1007/s13365-012-0108-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Holt JL, Kraft-Terry SD, Chang L. Neuroimaging studies of the aging HIV-1-infected brain. J Neurovirol. 2012;18:291–302. doi: 10.1007/s13365-012-0114-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Zapata HJ, Shaw AC. Aging of the human innate immune system in HIV infection. Current opinion in immunology. 2014;29C:127–136. doi: 10.1016/j.coi.2014.06.007. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Becker JT, Butters N, Hermann A, D’Angelo N. A comparison of the effects of long-term alcohol abuse and aging on the performance of verbal and nonverbal divided attention tasks. Alcoholism: Clin Exp Res. 1983;7:213–219. doi: 10.1111/j.1530-0277.1983.tb05444.x. [DOI] [PubMed] [Google Scholar]
- 34.Lopez OL, Kuller LH, Fitzpatrick A, Ives D, Becker JT, Beauchamp N. Evaluations of dementia in the cardiovascular health cognition study. Neuroepidemiology. 2003;22:1–12. doi: 10.1159/000067110. [DOI] [PubMed] [Google Scholar]
- 35.McArthur J. Neurologic manifestations of AIDS. Medicine. 1987;66:407–437. doi: 10.1097/00005792-198711000-00001. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.