|Grant Number:||7R03CA135671-03 Interpret this number|
|Primary Investigator:||Haneuse, Sebastien|
|Organization:||Harvard School Of Public Health|
|Project Title:||Design Considerations for Two-Phase Studies|
In identifying associations between exposures and diseases researchers have a wide variety of study designs at their disposal. Although two-phase studies have been shown to provide substantial efficiency gains over the traditional case-control design, they have not been widely adopted. A recent survey of 4,792 studies in five top-line epidemiological/medical journals, published since 2002, found just a single study having employed the two-phase design. This may be due, in part, to a lack of published guidance on how to design and plan a two-phase study and the practical difficulties of not having general-purpose software. To address these issues and broad the use of two-phase studies, this research proposes to develop a flexible framework for investigating design considerations in the context of planning a two-phase study. This will involve two steps; (i) development of a conceptual framework which encompasses the decisions required at the planning stage and (ii) developing an algorithmic simulation-based framework which facilitates the investigation of potential choices in a variety of settings. Building on this, we propose to conduct an extensive and comprehensive simulation study to explore the impact of various choices associated with the design of a two-phase study. This proposal is motivated by ongoing and future research conducted by the Breast Cancer Surveillance Consortium (BCSC). As new opportunities for scientific research within the BCSC arise it will be important to make the best possible use of the existing database/infrastructure. The two-phase design provides a framework to achieve this, and improved guidance on the planning and design of such studies will be crucial for future cost-effective and yet statistically powerful studies. A key component of this work, therefore, will be the dissemination of methods and results, and the delivery of software; algorithms/code, together with documentation, will be developed in commonly used statistical packages and made publicly available.