This proposal is aimed at developing new statistical methodology for evaluating preventive intervention trials in the mental health field. The existing methods for analysis of such longitudinal and developmental data from preventive trials implicitly depend on simplified assumptions about the ways individuals are assigned to intervention, the manner in which interventions are implemented, the stochastic nature of the outcome data, and reasons for data being missing. Even the most carefully designed preventive studies depart from these and other related assumptions, and these departures if ignored may lead to substantially incorrect conclusions about the effects of a prevention program. The consequences may be quite severe -- we may detect an apparent effect when it does not exist or we may fail to detect a meaningful effect. This proposal aims at reducing the possibilities of either type of error by developing new methods which take into account possible departures from these usual assumptions.
The specific aims of this proposal focus on major unresolved and generic issues in prevention research: how to make adjustments for nonrandom assignment to interventions, how to analyze nonrandomly missing data, how to make valid inferences about intervention effects on longitudinal data when we know little about the stochastic characteristics of the process, and how to examine the effectiveness of screening programs which in and of themselves may affect the outcomes of preventive trials. These new statistical methods will be tested against the classical experimental design methods using a variety of current and practical problems in prevention research. Three longitudinal data sets -- two with data now available and the third involving trials now developing -- will be used to test our methods on preventive interventions aimed at promoting mental health of children through school interventions. We will also examine the application of these methods to problems in adult mental disorders, especially in schizophrenia research. Recommendations for analytical methods coming out of this study should have wide applicability to prevention research because of the genetic matire of the prevention trial problems which are examined in this proposal. This proposal will involve close collaboration between biostatisticians and discipline oriented researchers in the mental health field over fundamental problems in prevention research. In the first three years of this grant we will be focusing on essential prevention problems, framing these problems in statistical terms, developing a broad statistical foundation for making inferences about preventive trials, and testing these new methods on two prevention data sets. In the fourth and fifth years we will continue further development of new analytic methods and test them on two state of the art preventive trials now being implemented.
Showing the most recent 10 out of 120 publications
