Accurate prediction of longitudinal changes in cognitive function in treated HIV-disease would potentially allow targeted intervention in those at greatest risk of cognitive decline. High resolution neuroimaging and machine learning techniques have shown promise in other disease areas but there are little data in treated HIV-disease. We sought to build a multivariate model using volumetric neuroimaging data only to accurately predict cognitive function.
T1-weighted neuroimaging data from virally suppressed people living with HIV (PLWH) from the CHARTER cohort (n=139) were segmented into grey and white matter and spatially normalised before entered into machine learning models. Cognitive function was tested across seven cognitive domains with longitudinal decline determined using regression based change scores as previously described. Predictive ability to assess cognitive function was determined using leave-one-out cross validation. Additionally, a multivariate model of brain ageing (training set n=2001, age range 18-90, model R²=0.88) was used to measure the deviation of apparent brain age from chronological age and assess its relationship with cognitive function.
Cognitive impairment, defined using the global deficit score (GDS), was present in 37.4%. However, it was generally mild, with only 4.3% meeting the criteria for HIV-associated dementia, and occurred more commonly in PLWH with confounding comorbidities (p<0.001). Longitudinal cognitive decline was present in 14.5%. Although multivariate prediction of cognitive impairment as a dichotomous variable at baseline was poor (table), prediction of the global T-score was better than a comparable linear model (adjusted R2=0.08, p<0.01 vs adjusted R2=0.01, p=0.14). Accurate prediction of longitudinal changes in cognitive function was not possible (p=0.82). Brain-predicted age exceeded chronological age by 1.17 (-0.14-2.53) years, but was greatest in those with confounding comorbidities (5.87 [1.74-9.99] years) and prior AIDS (3.03 [0.00-6.06] years). The relationship between this deviation and cognitive function was attenuated when covarying for comorbidity status.
Although cognitive impairment was present in about a third of virally suppressed patients it was confounded by comorbid conditions. Accurate prediction of cognitive function using multivariate models of T1-weighted MRI data was not achievable, which may reflect the small sample size, heterogeneity of the data or that impairment was usually mild.