Evaluation of selection bias in an incident-based dementia autopsy case series

Neuropathological (np) relative frequency estimates of dementia may be biased if the autopsied subjects are not representative of all dementia subjects within a target population. We identified characteristics that differed between autopsied and non-autopsied subjects from an incident-based dementia case series and compared autopsy-based estimates of the relative frequency of np diagnoses before and after adjusting for potential selection bias. Clinically demented subjects who were autopsied (n = 206), had died but were not autopsied (n = 271), were still alive (n = 71), or had dropped out of the study (n = 82) were included. Compared with non-autopsied subjects, autopsied subjects were more likely to be Caucasian, educated beyond high school, and married. They also tended to have a lower baseline Mini-Mental State Examination score and were more likely to have a clinical diagnosis of Alzheimer disease (AD) than non-autopsied subjects. Neuropathological AD with Lewy bodies (LB) had the largest crude relative frequency estimate at 38% of the autopsy sample, followed by 25% for AD with vascular lesions, 13% for pure AD, 13% for LB (with or without vascular lesions), and 8% for pure vascular pathologies. Adjustment for potential sources of selection bias had little effect on relative frequency estimates, suggesting that np diagnoses in the autopsied subjects provide reasonable dementia relative frequency estimates among all clinically demented cases in this series.