The SMART project, Statistical Modeling of Aging and Risk of Transitions, is a consortium of 11 longitudinal studies on cognition in the elderly accompanied by autopsy measurements. SMART has data on 11,541 elderly subjects, with 3,001 of these coming to autopsy. The mean age at death was 88.4 years, enabling the study of cognitive impairment occurring in the ninth decade of life, which is strongly associated with the presence of cerebrovascular disease, hippocampal sclerosis, and non-Alzheimer's disease pathologies. The purpose of the SMART project was to examine transitions from the cognitively intact state to impaired states including subjective memory complaints, Mild Cognitive Impairment, and eventually dementia. Risk factors often associated with the occurrence of these impaired states in younger elderly subjects do not necessarily apply to this cohort including smoking, diabetes, and hypertension. The main statistical tool, multi-state models, is affected by back transitions from impaired states to the normal state, interval censoring, and incomplete covariate data. We clarify the risk factors and pathologies associated with cognitive impairment in advanced old age.