National Institute
for Complex Data
Structures

Home | Links |Contact

National Program on Statistical Innovation for the

Analysis of Complex Data in Medical and Health Sciences

 

Brief Overview

 

 

Project Leaders:

 

·       Richard Cook, University of Waterloo

·       Paul Gustafson, Univ of British Columbia

·       Peter Song, University of Waterloo

·       Michal Abrahamowicz, McGill University

·       Wendy Lou, University of Toronto

·       Liqun Wang, University of Manitoba

 

 

Complex data structures arise from many subject-matter fields such as medical and health sciences, social sciences, economics and finance, and engineering. Such data are widely used to study causality, develop policy, and carry out program assessments. In medical and health sciences, statistical complexity often arises because of temporal and spatial structure in the collection of data, often with added complexity because data are incomplete, in terms of variables which may be missing for some subjects, or badly measured for all subjects. One important example is that of longitudinal data, where measurements of each subject are observed repeatedly over a number of time points, which essentially comprises a collection of time series. Along the considerable progress on the techniques of data storage and administration, it becomes increasingly common that multiple outcomes are measured for each subject through time, with possibly spatial structures. Statistical methods for the design and analysis of longitudinal data are critical for valid and efficient research in subject-matter areas. In these settings, numerous statistical challenges have arisen, which cannot be addressed by conventional regression techniques in that one observation is collected from one subject.

 

One important issue in the analysis of complex data structures is to deal with serial correlation among repeated measurements from one subject.  On the one hand, over-time trajectories of the response variable of interest provide rich information of individual records, so it is possible to model and understand the time-dependent behavior or relationship for the variables in a complex system under investigation. On the other hand, the autocorrelation of time series demands more complicated modeling strategies and computational effort.  In addition, complicating factors such as missing data and measurement errors will give rise to much tougher challenges that statisticians have ever confronted in the past.

 

Although the statistical community has been making considerable effort to attack some of these difficulties in the analysis of complex data structures arising in medical and health sciences, many key issues remain unsolved and call for further endeavor. This project is responding to this call and trying to solidify the strength of Canadian researchers to make a collective effort on solving some difficult problems in methodology and applications of complex data analysis. Through the funding of the NPCDS, we plan to coordinate a number of research activities at sites across Canada, with an emphasis in the training of highly qualified personnel. Such an organized research action is necessary and beneficial for Canadian methodological researchers, graduate students and practitioners. In addition, we will establish a network with industry and ensure a direct and efficient technology transfer. In the future we hope to build a wider and stronger community with the participation from researchers and practitioners from many subject-matter fields and industry.

 

Long term goals of this project include: (1) Develop methodology for the analysis of complex data structures arising in medical and health sciences. (2) Establish a network of Canadian statisticians, subject-area specialists and industrial partners, with timely communications regarding transferring innovative methods to product users and accessing new application-motivated problems. (3) Provide training of graduate students and post-doctoral fellows in both complex data research and applications. Through internships and summer schools, students will be prepared with hands-on experience in the analysis of data in medical and health studies. (4) Hold research meetings, in conjunction with different subject-area specialists and industry partners, such as meetings organized with the participation of medical and public health scientists and epidemiologists.

 

Back to top