Impact of Patient Selection Criteria on Prevalence Estimates and Prevalence of Diagnosed Dementia in a Medicaid Population

OBJECTIVES: This study estimated the prevalence of diagnosed dementia among Indiana Medicaid beneficiaries in 2004. The dependence of prevalence estimates upon use of several patient selection criteria to identify patients with dementia also was evaluated. METHODS: Indiana Medicaid claims data were analyzed for the period July 1, 2002 to December 31, 2004. An expert panel survey was conducted to assess perceived specificity of ICD codes used in previous studies to define dementia. Prevalence estimates were calculated with varying levels of each selection criteria, that is, ICD code set, interval of data examined, and number of occurrences of dementia-related claims. To assess specificity and sensitivity of the dementia patient selection criteria, Minimum Data Set data for a subset of beneficiaries that resided in a nursing home any time in 2004 were examined. RESULTS: Depending on the patient selection criteria used, estimates of prevalence of diagnosed dementia for individuals 40 years old or older varied from 7.7% to 15.3%, whereas prevalence estimates for individuals 60 years old or older varied from 14.5% to 26.6%. When the following selection criteria were used: (1) occurrence of one or more dementia-related claims, (2) the expert panel ICD set, and (3) up to 30 months of data for defining dementia, the prevalence estimates in the Indiana Medicaid population were 10.9% for individuals 40 years old or older and 20.3% for individuals 60 years old or older. CONCLUSIONS: Careful selection of claims-based criteria for identifying patients with dementia is important because the criteria may affect estimates by 100%. Prevalence of diagnosed dementia among Indiana Medicaid beneficiaries was 3 to 4 times higher than the reported prevalence from a decade ago in Medicaid populations of other states, even when the same patient selection criteria were used. A number of factors beyond increased occurrence of the disease including increased screening, greater likelihood of recording dementia codes in claims, or other factors may be responsible. The combination of patient selection criteria used in this study had good sensitivity, specificity, and accuracy when compared with Minimum Data Set data.