Identifying neuropsychiatric disorders in the Medicare Current Beneficiary Survey: the benefits of combining health survey and claims data

Background: To address the impact of using multiple sources of data in the United States Medicare Current Beneficiary Survey (MCBS) compared to using only one source of data to identify those with neuropsychiatric diagnoses. Methods: Our data source was the 2010 MCBS with associated Medicare claims files (N = 14, 672 beneficiaries). The MCBS uses a stratified multistage probability sample design to select a nationally representative sample of Medicare beneficiaries. We excluded those participants in Medicare Health Maintenance Organizations (n = 3894) and performed a cross-sectional analysis. We classified neuropsychiatric conditions according to four broad categories: intellectual/developmental disorders, neurological conditions affecting the central nervous system (Neuro-CNS), dementia, and psychiatric conditions. To account for different baseline prevalence differences of the categories we calculated the relative increase in prevalence that occurred from adding information from claims in addition to the absolute increase to allow comparison among categories. Results: The estimated proportion of the sample with neuropsychiatric disorders increased to 50.0 (both sources) compared to 38.9 (health survey only) and 33.2 (claims only) with an overlap between sources of only 44.1 %. Augmenting health survey data with claims led to an increase in estimated percentage of intellectual/developmental disorders, psychiatric disorders, Neuro-CNS disorders and dementia of 1.3, 5.9, 11.5 and 3.8 respectively. In the community sample, the largest relative increases were seen for dementia (147.6 %) and Neuro-CNS disorders (87.4 %). With the exception of dementia, larger relative increases were seen in the facility sample with the greatest being for intellectual/developmental disorders (121.5 %) and Neuro-CNS disorders (93.8 %). Conclusions: The magnitude of potentially underestimated sample proportions using health survey only data varied strikingly according to the category of diagnosis and setting. Augmentation of survey data with claims appears essential particularly when attempting to estimate proportion of the sample affected by conditions that cause cognitive impairment which may affect ability to self-report. Augmenting proxy survey data with claims data also appears to be essential when ascertaining proportion of the facility-dwelling sample affected by neuropsychiatric disorders.