Clustering Technique Applied to a Case-Control Study of the Association of Emergency Department Respiratory Admissions and School Outdoors and Indoors Exposures, New York State

• Activity Number: 554
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): JeanPierre Munsie* and Cris Pantea and Shao Lin
• Companies: CEH/New York State Department of Health and CEH/New York State Department of Health and CEH/New York State Department of Health
• Keywords: Clustering technique ; variables reduction ; respiratory admissions.

Abstract

Objective: Multiple highly correlated variables have been always a challenge in environmental health studies. To address this issue, we used clustering technique to eliminate and select variables from a long list of independent variables. Method: we collected data on the outdoor and indoor exposure of 68,083 Emergency Department visits for lower respiratory diseases of children 5-18 years old. Outdoor exposure include traffic counts, proximity to roads, industrial facilities and airports from homes and schools. School indoor exposure include more than 50 variables related to building and classroom conditions. The clustering procedure "proc varclus" from SAS/STAT was applied for variables reduction and selection. The rule of thumb is to choose the variable with the minimum 1-R2 ratio as the cluster representative. The 1-R2 is defined as follow: 1-R2 ratio = (1-R2 own cluster)/(1-R2 closest cluster). Results: the procedure returned seven clusters with correlated variables and the 1-R2 ratio. The variable with 1-R2 ratio close to zero was chosen from each cluster to create the final model. This method of variables reduction will benefits future environmental health studies.