P4-036: Identifying community-dwelling individuals with Alzheimer's disease using observational data

Background: Using diagnosis codes in administrative data to estimate the prevalence of Alzheimer's disease (AD) has been shown to introduce errors of omission and commission. The objectives of this study were to identify the most sensitive and specific alternative disease definition for case ascertainment and to provide insight on causes of widely-varied cost estimates in the cost-of-illness literature. Methods: Retrospective cross-sectional analyses were performed on a nationally representative sample of elderly, community-dwelling Medicare beneficiaries using the 1999-2004 waves of the Medicare Current Beneficiary Survey (MCBS) (n=57,669). Individuals with AD were identified based on six definitions according to (1) survey report only, (2) diagnosis only, or (3) use of Alzheimer's prescription medicine only, or some combination of the three. Agreement between different case definitions was assessed using a Venn diagram, kappa coefficient, sensitivity and specificity. We also explored the extent to which estimates in overall health care and prescription drug expenditures vary by how Alzheimer's cases are defined. Results: Using any of the three case definitions, 3,198 individuals (5.55%) were identified as having AD. Using survey report alone yielded more cases (n=1,994 or 3.46%) than diagnosis codes alone (n=1,589 or 2.76%) or Alzheimer's medication use alone (n=1,160 or 2.01%). Kappa coefficients were low, ranging from 0.37 to 0.40. Using diagnosis in medical claims as a definitive measure for AD (i.e., “gold standard”), survey report was specific (97.7%) but had low sensitivity (44.2%) and Alzheimer's medication use was slightly more specific (98.9%) but less sensitive (34.8%). Per capita health expenditures ranged from $16,547 to $24,937, and drug expenditures ranged from $2,303 to $3,519, depending on how AD was defined. Conclusions: Different case definitions identify different subsets of individuals with AD. Self-report appears to be a highly specific approach to identifying Alzheimer's cases without the loss of sensitivity found in Alzheimer's medication use. As a consequence of different disease definitions employed, the costs of AD varied widely. One should exercise caution in interpreting current cost-of-illness studies and in applying these estimates to policy initiatives.