文章基本信息

标题：Design and Analysis of Group-Randomized Trials: A Review of Recent Methodological Developments
本地全文：下载
作者：David M. Murray ; Sherri P. Varnell ; Jonathan L. Blitstein 等
期刊名称：American journal of public health
印刷版ISSN：0090-0036
出版年度：2004
卷号：94
期号：3
页码：423-432
语种：English
出版社：American Public Health Association
摘要：We review recent developments in the design and analysis of group-randomized trials (GRTs). Regarding design, we summarize developments in estimates of intraclass correlation, power analysis, matched designs, designs involving one group per condition, and designs in which individuals are randomized to receive treatments in groups. Regarding analysis, we summarize developments in marginal and conditional models, the sandwich estimator, model-based estimators, binary data, survival analysis, randomization tests, survey methods, latent variable methods and nonlinear mixed models, time series methods, global tests for multiple endpoints, mediation effects, missing data, trial reporting, and software. We encourage investigators who conduct GRTs to become familiar with these developments and to collaborate with methodologists who can strengthen the design and analysis of their trials. Group-randomized trials (GRTs) are comparative studies designed to evaluate interventions that operate at a group level, manipulate the physical or social environment, or cannot be delivered to individuals. 1 Examples include school-, worksite-, and community-based studies designed to improve the health of students, employees, and residents, respectively. Just as the randomized clinical trial (RCT) is the gold standard in public health and medicine when allocation of individual participants is possible, the GRT is the gold standard when allocation of identifiable groups is necessary. There are 4 characteristics that distinguish the GRT from the more familiar RCT. First, the unit of assignment is an identifiable group; such groups are formed not at random but rather through some physical, social, geographic, or other connection among their members. Second, different groups are assigned to each condition, creating a nested or hierarchical structure for the design and the data. Third, the units of observation are members of those groups nested within both their condition and their group. Fourth, usually only a limited number of groups are assigned to each condition. These characteristics create several problems in the design and analysis of GRTs. 1 The major design problem is that a limited number of often heterogeneous groups makes it difficult for randomization to distribute potential sources of confounding evenly in any single realization of the experiment. This increases the need to use design strategies that will limit confounding and analytic strategies to deal with confounding when it is detected. The major analytic problem is that there is an expectation for a positive intraclass correlation (ICC) among observations of members of the same group. 2 That ICC reflects an extra component of variance attributable to the group above and beyond the variance attributable to its members. This extra variation will increase the variance of any group-level statistic beyond what would be expected with random assignment of members to conditions. Moreover, with a limited number of groups, the degrees of freedom available to estimate group-level statistics are limited. Any test that ignores either the extra variation or the limited degrees of freedom will have a type I error rate that is inflated, and this effect will only worsen as the ICC increases. 3 Cornfield 4(p101–102) warned of this danger 25 years ago when he noted that ignoring these problems was “an exercise in selfdeception . . . and should be discouraged.” That warning was followed by a gradual increase in the number of methods papers in this area. The first comprehensive text on the design and analysis of GRTs appeared in 1998. 1 It detailed the design considerations for the development of GRTs, described the major approaches to their analysis both for Gaussian and binary data, and presented methods for power analysis applicable to most GRTs. We use that text as a point of departure for this review and assume that readers are familiar with its basic material. Over the past 5 years, many articles have discussed the methodological issues involved in GRTs generally or in design papers describing new trials. 5– 28 The second textbook on design and analysis of GRTs appeared in 2000. 29 That text provided a good history of GRTs and examined the role of informed consent and other ethical issues. It focused on extensions of classical methods, although it also included material on regression models for Gaussian, binary, count, and time-to-event data. Other textbooks on analysis methods germane to GRTs appeared during the same period, 30– 33 as well as a large number of articles on new methods relevant to the design and analysis of GRTs. In the sections that follow, we bring the reader up to date on many of these developments.