Gold-standard clinical dementia adjudication is resource intensive and infeasible in large, population-representative studies which are critical for public health research. Algorithmic dementia classification uses models to predict cognitive impairment and was developed to circumvent challenges of the gold-standard adjudication process. Several algorithms have been developed to classify dementia in the Health and Retirement Study (HRS) and rely on information in the Aging, Demographics, and Memory Study (ADAMS), a substudy of HRS initiated in 2001. Existing algorithms cannot incorporate neuropsychological measures as they are unavailable in HRS, and models cannot be adapted to include more comprehensive measures available in newer studies.I propose a novel Bayesian latent class mixture modeling framework for algorithmic dementia classification that incorporates information from neuropsychological measures and can be adapted to include more comprehensive measures available in updated studies. The model uses latent class mixture models to create synthetic versions of datasets, incorporating information on relationships between sociodemographic, health, and cognitive measures and cognitive impairment classes through prior distributions based on studies with gold-standard adjudicated cases. This work involves three studies on aging: The Health and Retirement Study (HRS), The Harmonized Cognitive Assessment Protocol (HCAP, HRS substudy), and the Aging and Demographics Study (ADAMS, HRS substudy). Simulation studies were conducted to evaluate the role of study sample size and priors specified based on different data sources and sampling frames and their impact on algorithmic dementia classification results and inferences on racial/ethnic differences in dementia. Analyses using priors from ADAMS accurately captured cognitive impairment classes preserved racial/ethnic differences in dementia for Black vs. White participants. Priors better calibrated to the analytic sample however improved estimates for Black and Hispanic participants and preserved racial/ethnic differences in dementia for Black vs. White and Hispanic vs. White participants. Applying the model to HCAP 2016 yielded reasonable estimates of cognitive impairment classes with proportions of impaired participants in line with findings published by HCAP investigators. This dissertation lays important groundwork for strengthening algorithmic dementia classification in population-representative studies. Outcomes from this work are directly applicable to existing studies on AD/ADRD that are harmonizable with HRS/HCAP.