Medicaid data as a resource for epidemiologic studies: strengths and limitations.

Large claims databases from third-party insurance programs such as Medicaid have attracted the interest of epidemiologists because of their enormous size and apparent comprehensiveness. Over 20 million people are covered by the various state Medicaid programs and most states maintain detailed computerized records of all reimbursed health care encounters on a recipient-specific basis. For states covering medication costs, data on drug exposures are particularly complete and accurate. However, Medicaid claims data also have many limitations that can pose major methodological difficulties. Foremost among these is the uneven validity and completeness of the diagnoses appearing on claims. Likewise, the unique identification of specific program participants is not straightforward, although useful approaches can often be developed to track individuals over time. Consideration of the limitations as well as the possible strengths of claims-based data makes it possible to choose appropriate study hypotheses as well as to attempt solutions, where possible, to the biases of this methodology.