Sleep health and cognitive function among people with and without HIV: The use of different machine learning approaches

Study Objectives: We investigated associations between actigraphy-assessed sleep measures and cognitive function in people with and without HIV using different analytical approaches to better understand these associations and highlight differences in results obtained by these approaches. Methods: Cognitive and 7-day/night actigraphy data were collected from people with HIV (PWH) and lifestyle-similar HIV-negative individuals from HIV and sexual health clinics in the United Kingdom/Ireland. A global cognitive T-score was obtained averaging the standardized individual cognitive test scores accounting for sociodemographics. Average and SD of 11 sleep measures over 7 days/nights were obtained. Rank regression, partial least-squares (PLS) regression, random forest, sleep dimension construct, and latent class analysis (LCA) were applied to evaluate associations between global T-scores and sleep measures. Results: In 344 PWH (median age 57 years, 86% males), average sleep duration, efficiency, and wake after sleep onset were not associated with global T-scores according to rank regression (p = 0.51, p = 0.09, p = 0.16, respectively). In contrast, global T-scores were associated with average and SD of length of nocturnal awakenings, SD of maintenance efficiency, and average out-of-bed time when analyzed by PLS regression and random forest. No associations were found when using sleep dimensions or LCA. Overall, findings observed in PWH were similar to those seen in HIV-negative individuals (median age 61 years, 67% males). Conclusions: Using multivariable analytical approaches, measures of sleep continuity, timing, and regularity were associated with cognitive performance in PWH, supporting the utility of newer methods of incorporating multiple standard and novel measures of sleep-wake patterns in the assessment of health and functioning.