Cognitive Training for the Remediation of Functional Brain Health in HIV

The typical patterns of cognitive deficits in HIV reflect degraded brain network functioning, likely due to a combination of brain health insults: some generic (aging), some HIV-specific (inflammation, diffuse demyelination and inherent vulnerability that varies across individuals). Consistent with this view, the cognitive domains most affected are those that rely on extended networks (e.g. attention and executive functions relying on fronto-parietal and fronto-striatal circuits), exquisite timing (psychomotor function), or both. These network-based cognitive functions are vulnerable, but they are also resilient: there is a high degree of learning-dependent plasticity in networks involving the frontal lobes. This argues that the cognitive deficits in HIV may be amenable to remediation through cognitive training, and suggests mechanisms by which this might occur. There are many forms of cognitive rehabilitation; approaches that take advantage of advances in our understanding of the mechanisms of neuroplasticity and the neural systems supporting human cognition are likely to be highest yield. In this study, we will make use of Plasticity-based Adaptive Cognitive Remediation (PACR), a powerful method for harnessing this plastic potential. Conceptually, PACR applies well-understood techniques derived from brain plasticity and implicit/procedural/perceptual learning to improve the speed and accuracy of information processing, with exercises that are designed to drive generalized improvements. Simultaneously, these exercises heavily engage neuromodulatory systems to re-establish their normal control over learning and memory. As an individual restores these degraded abilities through intensive procedural learning, the encoding of naturalistic information significantly improves, and all resulting declarative memory and cognitive functions based on the quality of that incoming information necessarily improve as well, leading to improvement that generalizes beyond the trained tasks. Multiple randomized controlled studies have now demonstrated that PACR improves cognitive and functional abilities in patient populations with cognitive dysfunction similar in type and magnitude to patients with cognitive deficits due to HIV.

PACR runs in a web browser on any Internet connected computer and is implemented in an engaging game-like format. The participant selects one of the cognitive exercises scheduled for the day, and performs that exercise for fifteen minutes. The exercise itself contains the core science stimuli and task built into a game-like experience. Participants perform tens to hundreds of trials over the course of the fifteen-minute session, with each trial providing auditory and visual feedback and rewards to indicate if the trial was performed correctly or incorrectly. After each trial, the difficulty of the next trial is updated to ensure that within a session, the participant gets ~85% of trials correct. Thus, training is individually tailored to maximize its effectiveness. Summary screens including game metrics (points, levels) and exercise metrics (usage, progress) are shown to the participant at the end of each session. The scheduling mechanism ensures that a patient progresses through the exercises in a defined order, generally moving from more simple (early sensory processing) exercises to more complex (multimodal, cognitive control) exercises over the course of the 8 weeks experience. At any point in time, the participant only has access to a subset (typically six) of these exercises, four of which are performed per day.

Each exercise has specific criteria for completion, and after those criteria are met the exercise is removed from the active set and the next exercise added. This mechanism ensures both ongoing novelty and engagement for the participant, and that the participant progresses smoothly through the complete set of exercises over the program use period.

Free access will be provided to the PACR program, and a tailored cognitive training program will be developed in both French and English, specifically targeting domains and mechanisms that are most affected in HIV. This will include selection of the most suitable training modules (12 are planned), and optimization of the Web-based presentation and feedback to ensure acceptability to this target group.

The treatment goal will be use of the assigned program in 30 minutes sessions, five sessions per week, for 8 weeks after randomization; program use will be any mix of at home (or community Internet resource) or in-clinic sessions.

The outcome will be responder status (defined as improvement of >0.5 logits) on the B-CAM. With the assumption that the outcome is drawn from a binomial distribution with an expected probability of response of 10% (n=3) with no intervention, 30 subjects in the immediate training group will allow detection of a positive response at P<0.05 if 7 or more persons respond. The observed responses in both groups will provide more accurate estimates to plan for a scale up of this work to a full trial. An exploratory analysis will evaluate response in only those who completed at least 60% of the training sessions, recognizing that power here will be reduced, but the information nonetheless important.

Participants from both groups will also be compared to all those eligible for randomization to this intervention in the platform as a whole. Generalized estimating equations (GEE) will be applied as a secondary, more general approach here that permits other time points to be modeled, and consideration of other outcomes. This accommodates either binary (responder status) or continuous (scores on cognitive tests) outcomes. This analysis uses a regression model, but clustering of outcomes within time is controlled. For binary outcomes, the effect of group (immediate or control) is expressed as an odds ratio; for continuous outcomes the parameter is an effect size equivalent to an adjusted paired-t-test. An interaction term tests whether the effect differed by group (i.e. was larger in the immediate training group as hypothesized).

Additional analyses will be used to explain changes in B-CAM score as a function of changes expected from the intervention. As the intervention cohort is small, we will use concordance parameters, rather than a regression model, to quantify the degree to which changes in hypothesized mechanisms by which the interventions operate are concordant (at the individual level) with changes in the outcomes (cognitive ability).