Experimental versus Observational Data in the Economic Evaluation

It is commonly suggested that there are two com-peting approaches to the economic evaluation of health technologies. In a trial-based study, economic data (e.g., use of resources) are collected concur-rently with the clinical study. In a modeling (or in-tegrative) study, data from a number of sources are synthesized. In the first case, the economic data can be viewed as experimental and are typically analyzed in the same way as the clinical data. In the second case, the data for the economic evaluation could be either experimental (e.g., efficacy data from clinical trials) or observational (e.g., resource-use data extracted from patient chart review or a claims database). In-deed, there are various potential sources of obser-vational data in economic evaluations, including claims databases, epidemiologic studies such as the Framingham Heart Study, and surveys of patients’ health-state-preference values. Even trial-based studies use some data obtained from outside the clinical trial, such as the prices (or unit costs) of health care resources.l Recently there has been considerable debate about the relative merits and demerits of trial-based and modeling approaches to economic evaluation. In an editorial in the New England Journal of Med-icine, Kassirer and Angel2 argued that “bias can compromise even original scientific studies, but we believe that the opportunities for introducing bias into economic studies are far greater, given the dis-cretionary nature of model building and data selec-tion in those analyses.” Furthermore, in its draft Principles for the Review of Pharmacoeconomic Promotion,3 the Food and Drug Administration (FDA) argued that research to substantiate pharmacoeconomic claims, including cost-effectiveness and quality-of-life claims, must meet traditional standards for adequate and well-controlled studies. It also suggested that models to provide estimates of pharmacoeconomic parame-ters be used only when it is impractical or impos-sible to gather data using adequate and well-Address correspondence and reprint requests to Professo