![](images/null.gif) |
![](images/null.gif) |
![](images/null.gif) |
|
Four Distinct Biopsychosocial Phenotypes in HIV in the CHARTER Cohort-and Why They Matter
|
|
|
30th CROI, Conference on Retroviruses and Opportunistic Infections, February 19-22, 2023, Seattle
Mark Mascolini
One of four "distinct, clinically plausible, and complex" biopsychosocial phenotype may be assigned to people with HIV infection, according to a machine-learning analysis by researchers working with the US CHARTER Cohort [1]. The four phenotypes are not academic pigeonholes but categories that can inform the complex management of people with HIV.
The CHARTER team set out to define these biopsychosocial phenotypes in people with HIV to account for differences in neurocognitive performance and symptoms, depressed mood, and daily functioning. Neuropsychiatric problems like neurocognitive impairment and depression often affect people with HIV, and sometimes multiple problems affect the same person, the researchers noted. Differences between these conditions are reflected in different clinical presentations, disability, and treatment.
To better define these differences in cognitive and mood complications in people with HIV, CHARTER investigators turned to machine learning, a branch of artificial intelligence that "focuses on the use of data and algorithms to imitate the way that humans learn, gradually improving its accuracy" [2].
The analysis involved people in the US CHARTER cohort consisting of 1580 people with HIV at six university medical centers. Enrollment in CHARTER had few comorbidity exclusions. Cohort members were advised to start antiretroviral therapy when their CD4 count dropped below 350.
Domain deficit scores identified seven cognitive domains. Further testing included four subscales of the Beck Depression Inventory II, five components of self-reported symptoms according to the Patient's Assessment of Own Functioning Inventory, and one total score of dependence in instrumental activities of daily living (IADLs). The machine learning component of the analysis used a two-stage unsupervised machine learning clustering approach involving (1) self-organizing maps and (2) K-means clustering.
The machine learning exercise generated four distinct phenotypic groups:
1. Healthy
2. Very mild neurocognitive impairment plus moderate to severe depression plus mild symptoms/IADL impairment
3. Mild neurocognitive impairment with severe depression, cognitive symptoms, and IADL impairment
4. Moderate neurocognitive impairment with no depression, no cognitive symptoms, and no IADL impairment
When people entered CHARTER, the cohort had 608 people with phenotype 1, 270 with phenotype 2, 205 with phenotype 3, and 497 with phenotype 4. Age at cohort entry averaged about 43 in all four groups, years of education averaged about 12.5, and about 47% were black.
Men made up 82% with phenotype 1, 73% with phenotype 2, 67% with phenotype 3, and 76% with phenotype 4 (P < 0.001). Average months of HIV infection also differed across the four phenotypes: 50.4 with phenotype 1, 56.6 with phenotype 2, 58.3 with phenotype 3, and 61.0 with phenotype 4 (P < 0.001). Proportions with a viral load below 50 copies did not differ significantly across groups (40.1% to 44.4%), but proportions taking antiretrovirals did differ: 65.4% with phenotype 1, 72.1% with phenotype 2, 72.1% with phenotype 3, and 76.4% with phenotype 4 (P = 0.002).
Median lowest-ever CD4 count also differed significantly across groups: 187 with phenotype 1, 170 with phenotype 2, 171 with phenotype 3, and 133 with phenotype 4 (P = 0.006). But median current CD4 count did not differ across groups (405 to 465).
CHARTER investigators concluded that they identified "four distinct, clinically plausible, and complex biopsychosocial phenotypes among people with HIV, characterized by differences in neurocognitive performance, depressive mood symptoms, cognitive symptoms, and IADL dependence."
Why does it matter? Because care may differ by phenotype. For example, the CHARTER team suggested, "whereas a person with depression but without cognitive impairment might respond best to a combination of psychotherapy and a traditional daily antidepressant medication, someone who has both depression and cognitive impairment (ie, risk factors for poor adherence) might respond best to a long-acting antiretroviral regimen and a neuroprotective agent to address both components of their disability and to facilitate adherence to the interventions."
The investigators noted the low proportion of women in the analysis, but they observed that it approximated the proportion of females with HIV in the United States. They added that more study is needed to validate these results in other groups, to see how often people move from one phenotype to another, and to identify factors linked to such transitions.
References
1. Tang B, Ellis RJ, Vaida F, et al. Biopsychosocial phenotypes in people with HIV in the CHARTER Cohort. 30th CROI, Conference on Retroviruses and Opportunistic Infections, February 19-22, 2023, Seattle. Abstract 474.
2. IBM. What is machine learning? https://www.ibm.com/topics/machine-learning
![0228231](../images/023023/022923-6/0228231.gif)
![0228233](../images/023023/022923-6/0228233.gif)
![0228234](../images/023023/022923-6/0228234.gif)
![0228235](../images/023023/022923-6/0228235.gif)
|
|
|
![](images/null.gif) |
![](images/null.gif) |
|
|