Abstract
OBJECTIVES: The aim of this study was to develop a refined method for harmonizing longitudinal cognitive data across several large-scale studies in people with HIV (PWH), in whom cognitive complications are common and heterogeneous in presentation. STUDY DESIGN AND SETTING: We developed a refined method for harmonizing longitudinal cognitive data across five large-scale studies in PWH that used different cognitive batteries with only some overlapping tests-Women's Interagency HIV Study (WIHS), Multicenter AIDS Cohort Study, CNS HIV Antiretroviral Therapy Effects Research (CHARTER), National NeuroAIDS Tissue Consortium, and the HIV Neurobehavioral Research Program. Traditional data harmonization methods using latent variable models focus on cross-sectional data and require the presence of common cognitive tests to serve as "linking" assessments. However, the absence of such common tests for certain cognitive domains can preclude the direct application of these traditional techniques. To address these challenges, we developed a harmonization method that leveraged a second-order factor model, which capitalized on the structural relationships among cognitive domains. RESULTS: Our approach yielded harmonized cognitive domain scores that are demographically consistent across different cohorts and exhibit strong correlations with the raw or log transformed (eg, timed outcomes) cognitive test scores. These harmonized scores accurately reflected variations according to age, educational status, and other demographic factors, while preserving participants' longitudinal cognitive trajectories. CONCLUSION: Our harmonization methods are essential for future analyses of large-scale, retrospective data to understand the heterogeneity in cognitive complications in PWH. These methods can be applied to harmonize new datasets with similar measures.