Abstract Details

Many PLWH have cognitive impairment, despite the success of antiretroviral therapy. The in-depth information on MCI subtypes is lacking. Unsupervised machine learning can reveal hidden subgroups within heterogeneous data that might be difficult to recognize.

The HIV Netherlands Australia Thailand Research Collaboration (HIV-NAT) 207 study enrolled Thai PLWH aged ≥50 years. Cognitive performance was evaluated by the Thai-validated Montreal Cognitive Assessment (MoCA). Participants who scored between 23 and 27 (i.e. cut-off point ± 2) were included in the analysis. The total score of each of the seven cognitive domains served as cluster variables for the K-means cluster algorithm. Clustering validation was done using the elbow method, Silhouette coefficient, Calinski-Harabasz Index, and Davies-Bouldin Index.

Among 340 PLWH who completed the MoCA, 177 (52.1%) scored between 23 and 27 and were included in the analysis. The median (IQR) age was 54 (51-58) years, 118 (66.7%) were male, 149 (84.2%) had >6 years of education, median CD4 at entry was 620 (489-795) cells/mL, and 170 (96.1%) were virally suppressed. K-means cluster demonstrated five independent clusters; all had language impairment: (1) predominant memory with mild language impairment (25.4%), (2) congruous mild visuospatial/executive function-language impairment (22.0%), (3) predominant abstraction with mild visuospatial/executive function-language-memory impairment (19.2%), (4) predominant language with mild memory impairment (18.6%), and (5) congruous language-abstraction impairment (14.7%). There were significant differences in the proportion of participants with education of 6 years or less between clusters (p=0.012). There were no significant differences in HIV-related factors between clusters.

Five clusters were demonstrated among aging Thai PLWH with borderline MCI; all were multiple domains involvement and had language impairment. A longitudinal study is warranted to identify differences in clinical significance and prognosis between each cluster.