While the experimental Alzheimer's drugs recently developed by pharmaceutical companies failed to stop the progression of Alzheimer's disease, clinicians strive to seek clues on how the patients would be when they visit back next year, based upon the patients' current clinical and neuropathologic diagnosis results. This is related to how to precisely identify the transitional patterns of Alzheimer's disease. Due to the complexities of the diagnosis of Alzheimer's disease, the condition of the disease is usually characterized by multiple clinical and neuropathologic measurements, including Clinical Dementia Rating (CDRGLOB), Mini-Mental State Examination (MMSE), a score derived from the clinician judgement on neuropsychological tests (COGSTAT), and Functional Activities Questionnaire (FAQ). In this research article, we investigate a class of novel joint random-effects transition models that are used to simultaneously analyze the transitional patterns of multiple primary measurements of Alzheimer's disease and, at the same time, account for the association between the measurements. The proposed methodology can avoid the bias introduced by ignoring the correlation between primary measurements and can predict subject-specific transitional patterns.