Menu
Home
Log in / Register
 
Home arrow Computer Science arrow Social Computing, Behavioral-Cultural Modeling and Prediction
< Prev   CONTENTS   Next >

System-Subsystem Dependency Network for Integrating Multicomponent Data and Application to Health Sciences

Abstract. Two features are commonly observed in large and complex systems. First, a system is made up of multiple subsystems. Second there exists fragmented data. A methodological challenge is to reconcile the potential parametric inconsistency across individually calibrated subsystems. This study aims to explore a novel approach, called systemsubsystem dependency network, which is capable of integrating subsystems that have been individually calibrated using separate data sets. In this paper we compare several techniques for solving the methodological challenge. Additionally, we use data from a large-scale epidemiologic study as well as a large clinical trial to illustrate the solution to inconsistency of overlapping subsystems and the integration of data sets.

Keywords: Generalized dependency network, Gibbs sampler, Systemsubsystem modeling.

1 Introduction

Diametrically opposed to reductionist thinking, system thinking seeks both local and global models that incorporate evidence about the delayed and distal effects of human intervention, policies, and situations [1]. Accordingly, system science requires a new set of tools to aid the understanding of complexity, feedback loop, stock-and-flow, and interactions between system components. Yet, for empirical researchers in system sciences, one important barrier is the lack of tools for integrating models built for different components, or subsystems–meaningful and scientifically self-contained smaller systems–of the entire system. Virtual world modelers often assemble submodels, which are independently calibrated using ad hoc and diverse data, into a system model that is amenable to simulation experiments. The Foresight Tackling Obesities: Future Choices project [2] provides an illustrative example of a system-subsystem. The Tackling Obesities project is a U.K. government initiative to find a policy response to the obesity epidemic over the next several decades. A centerpiece of the project is a collection of system maps, compiled by over 300 epidemiologists, nutritionists, geneticists, biologists, and social scientists for schematically capturing the different drivers of obesity. By not treating obesity only as a medical condition, the system approach redefines the nations health as a societal and economic narrative. Foresight researchers also created an overlay of subsystem contours. For example, physiology, individual physical activity, physical activity environment, individual psychology, and societal influences all exist as subsystems. Partly for interpretability and simplicity, the Foresight subsystem contours were drawn not to overlap with each other. However, in reality, most subsystems are neither closed nor independent of one another. More realistically, subsystems often share common drivers and (sub)components. Indeed, the interface between subsystems could be more interesting than the components; if a system can be compartmentalized into completely unrelated and independent subsystems, then the system approach would have almost become a moot exercise. From a model building perspective, it would be easier, perhaps as part of a divide-and-conquer strategy, to first develop models for subsystems and then “glue” the established subsystem models together to form a system model.

This paper attempts to explore methods for integrating subsystems, which may overlap with each other, into a coherent system through a generalized version of the dependency network (GDN) [3], a modeling tool developed for handling complex and reciprocal relationships. The subsystem method is primarily motivated by two applications the study of childhood obesity, which we briefly described above, and that of the aging and disablement process in older adults. Because of space limitation, in this paper we only focus on the latter. A contemporary view of the human disablement that emerged from the literature is that it is a complex dynamic system that consists of multiple subsystems interacting with each other [4]. Two particular challenges when one applied a system science approach to the aging and disablement model are: (1) inference in the presence of a large number of variables that can be approximately categorized into different subsystems but the subsystems also have overlapping variables, and (2) integration of data sets collected from different studies, which possibly have different emphases on the types of data they respectively collect. In this paper we compare several techniques for solving the two issues within the system-subsystem dependency network framework. We respectively use data from a large-scale epidemiologic study to illustrate the solution to the former issue, and additional data from large-clinical trial to illustrate the latter.

 
Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >
 
Subjects
Accounting
Business & Finance
Communication
Computer Science
Economics
Education
Engineering
Environment
Geography
Health
History
Language & Literature
Law
Management
Marketing
Philosophy
Political science
Psychology
Religion
Sociology
Travel