Analysis Methods for Volume-Outcome Studies

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Numerous studies have appeared in the cancer literature in the past few years linking hospital and surgeon procedure volume with patient outcomes. Results from these studies have direct policy implications, since regionalization is a considered strategy to improve the quality and efficiency of many different types of health care. Evaluation of an association between hospital or surgeon procedure volume and patient outcomes involves complex statistical issues that arise from the fact that the unit of observation is the patient, but these studies include multiple patients per hospital or surgeon as well as multiple hospitals or surgeons. Hence, patient outcomes tend to be correlated within hospitals or within surgeons, i.e., patients treated at the same hospital or by the same surgeon, may be more likely to experience similar outcomes than patients treated by a hospital or surgeon with the same volume. This phenomenon is referred to as "clustering" of outcomes, in the presence of clustering, standard statistical methods that assume patient outcomes are independent, are invalid. The general goal of this proposal is to critically examine the validity of widely-available statistical techniques that have been used in the context of volume-outcome studies such as generalized estimating equations and random effects models. The volume-outcome setting is unique in that "volume" reflects both the primary factor under study and also the cluster size, a fact that may well invalidate assumptions inherent in the use of available methods that correct for clustering. Simultaneous evaluation of the effects of hospital volume and surgeon volume is also hampered by the fact that the data are cress-classified, i.e., individual surgeons will perform surgeries at several hospitals. Through a detailed simulation study, the statistical validity of available statistical techniques in this context will be critically evaluated. Our methodological research will heighten awareness of clustering in health policy studies. Upon completion of our research plan, we will make recommendations about various analytic strategies for clustered binary data.