Simultaneous Modeling of Multiple Outcomes over Time: A Hierarchical Modeling Approach

• Activity Number: 152
• Type: Topic Contributed
• Date/Time: Monday, August 9, 2004 : 2:00 PM to 3:50 PM
• Sponsor: Section on Government Statistics
• Author(s): Abhik Das*+ and Charlotte Gard and Henrietta Bada and W. Kenneth Poole
• Companies: RTI International and RTI International and University of Kentucky Chandler Medical Center and RTI International
• Keywords: longitudinal data ; hierarchical models ; covariance structure ; simultaneous inference ; syndrome ; multiple outcomes

Abstract

Multiple outcomes assessed longitudinally over time are common in public health research. This work, motivated by the Maternal Lifestyle Study, involves children prenatally exposed to cocaine prospectively followed up for the onset of abnormal behavior problems assessed by several standardized instruments measuring different domains of behavior. Thus, at each time point, each child contributed to multiple, possibly correlated outcomes that collectively constitute one syndrome, while specific outcomes affected by cocaine are also of scientific interest. Since it is neither substantively appropriate nor statistically efficient to fit separate longitudinal models for each outcome, we utilize the covariance structure in hierarchical models to simultaneously fit all the behavioral outcomes as a function of cocaine exposure and other covariates. The advantages of this approach are: (a) it recognizes that all outcomes together constitute one syndrome; (b) it boosts statistical efficiency by incorporating correlated outcomes and avoiding multiple comparisons; (c) it allows for outcome-specific exposure effects; so we can identify the specific behavioral domains affected by cocaine.