Abstract:
|
Electronic medical records (EMR) are increasingly used to support clinical research. A key aspect of EMR data is a patient's treatment history recorded using current procedural terminology (CPT) codes, which describe medical services and procedures. A natural research question is whether different subgroups of patients have different utilization patterns, and interest may lie in the entire spectrum of potential services. In this talk, we detail testing, estimation, and data visualization methods for quantifying the significance and magnitude of differences in the endorsement of every CPT code between two groups of subjects. There are two major issues in the data: sparsity in coding for rare procedures, and the potential for code substitution where similar procedures may be referenced using alternative codes. We address these issues by exploring likelihood ratio tests, conditional exact tests, and sequence kernel association tests for inference. We use penalized regression to display code-wise differences across subgroups. In addition, we discuss dynamic visualization tools that can help healthcare providers and researchers to explore variation in healthcare utilization data.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.