Methodological Development

National Human Genome Research Institute

Investigators

Linked publications, trials & patents

Abstract

During the current reporting period, we developed and employed novel methods aimed at measuring, describing and modeling social networks. We continue to develop a framework for the psychometric evaluation of multiplex networks measuring a common relational construct. By multiplexity, we mean how different types of relationships overlap. This new research provides a general approach to modeling multivariate network systems addressing both questions about construct measurement and questions that consider the complexity of social systems that inherently involve multiple types of resource exchanges. We currently have a manuscript describing these new multi-layer models under review. Our work on network dynamics focuses on post-intervention changes in network composition on the one hand, and development of novel tools for the analysis of temporally unfolding micro-social processes on the other. We have been principal developers in a family of models broadly called Relational Events Models (REM) for social action. REMs can be employed to understand how a social behavior unfolds in time using an event history perspective. These novel methods have resulted in publicly available software through the R-CRAN and has been applied to animal models of interpersonal behavior. A recent publication used these models to understand how genomic information regarding child obesity effects parent feeding behavior. As well, an invited chapter describing these techniques is in press. In collaboration with Lise Getoor's lab, we have developed a computational approach for reconciling network data obtained through a multi-informant design. A paper describing and comparing various computational approaches was recently published. As well, we have demonstrated how the use of multi-informant approaches to family history assessment can potentially improve risk evaluations in the clinic. In addition, new research has examined how co-presence metrics derived from hospital administrative data can be used as an index test to predict nosocomial infection, identify inpatients who are subclinically infected, and to evaluate the impact of social influence on mortality in cancer patients receiving chemotherapy in an open setting. Novel metrics of co-presence have been developed, including a measure of consistent co-presence and co-presence thresholds that represent critical windows that increase the likelihood of infection during outbreaks of hospital born infections. During the reporting period, one manuscript from this work has been published, two are currently in review, and one manuscript is being finalized for submission.