|
Activity Number:
|
477
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Health Policy Statistics
|
| Abstract - #310042 |
|
Title:
|
Classification Models for Identifying Chronic Disease Cases and Noncases in Administrative Claims Data
|
|
Author(s):
|
Lisa M. Lix*+ and William D. Leslie and Marina Yogendran and Richard Baumgartner and Christopher Bowman and Souradet Shaw
|
|
Companies:
|
University of Manitoba and University of Manitoba and Manitoba Centre for Health Policy and National Research Council Canada and National Research Council Canada and University of Manitoba
|
|
Address:
|
408 727 McDermot Avenue, Winnipeg, MB, R3E 3P5, Canada
|
|
Keywords:
|
prevalence ; classification tree ; artifical neural network ; osteoporosis ; administrative claims
|
|
Abstract:
|
Administrative claims data are a popular tool for chronic disease surveillance. Algorithms to identify disease cases are often based on a simple heuristic applied to individual databases. This research compares artificial neural networks and classification trees to distinguish disease cases from non-cases in multiple databases, using osteoporosis as an example. Training and test data were obtained from Manitoba, Canada. Data features included disease diagnosis codes, prescription records for disease treatment, diagnosis and prescription records for comorbid conditions, and demographics. Both classification models produced similar estimates of discriminative power and prediction error. Misclassification rates ranged from 0.31 to 0.22 depending on the data features included in the model. The model predicted probabilities were used to estimate to estimate osteoporosis prevalence.
|
