Related Experiment Videos
Using multiple data features improved the validity of osteoporosis case ascertainment from administrative databases.
Lisa M Lix1, Marina S Yogendran2, William D Leslie3
1Manitoba Centre for Health Policy, University of Manitoba, Canada; Department of Community Health Sciences, University of Manitoba, Canada.
Journal of Clinical Epidemiology
|July 16, 2008
Summary
Developing osteoporosis case-finding algorithms using administrative data with multiple features improved sensitivity. These methods can be applied to identify other chronic diseases in large populations.
Area of Science:
- Health Informatics
- Epidemiology
- Data Science
Background:
- Administrative databases are valuable for public health research.
- Osteoporosis case ascertainment from administrative data presents challenges.
Purpose of the Study:
- To construct and validate algorithms for identifying osteoporosis cases using administrative data.
- To estimate population prevalence of osteoporosis via developed algorithms.
Main Methods:
- Applied artificial neural networks, classification trees, and logistic regression to hospital, physician, and pharmacy data.
- Compared algorithm performance using diagnosis codes, prescription drugs, comorbidities, and demographics.
- Validated algorithms against a bone mineral density testing program.
Main Results:
- Algorithms using expanded features (including prescriptions and age) showed better performance than diagnosis codes alone.
- Neural networks and classification trees had similar validation measures, but neural networks yielded lower prevalence estimates.
- Improved sensitivity for osteoporosis case detection was observed with multiple data features.
Conclusions:
- Multiple administrative data features enhance osteoporosis case-detection algorithm sensitivity without compromising specificity.
- Prevalence estimates using expanded features were slightly lower than primary data collection studies.
- Developed classification methods are adaptable for identifying other chronic diseases within administrative datasets.