|
Activity Number:
|
263
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, August 4, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Health Policy Statistics
|
| Abstract - #304174 |
|
Title:
|
Data-Mining Techniques for Longitudinal Naturalistic Data
|
|
Author(s):
|
Anthony Zagar*+ and Robert Obenchain
|
|
Companies:
|
Eli Lilly and Company and Risk Benefit Statistics LLC
|
|
Address:
|
Lilly Corporate Center, Indianapolis, IN, 46285,
|
|
Keywords:
|
latent classes ; prognostic scores ; longitudal data ; naturalistic data ; propensity scores ; unsupervised methods
|
|
Abstract:
|
We classify (cluster) patient Y-response trajectories (treatment progression patterns) that are associated with patient X-characteristics such as alternative treatment choices, comorbidities and baseline disease severity. "Unsupervised" approaches to analysis of observational data cluster patients in X-space (McClellan, McNeil & Newhouse 1994); the earlier "supervised" approaches form strata of patients matched on propensity score (estimated treatment fraction) predicted by patient X-characteristics (Rosenbaum & Rubin 1983, 1984.) Our classes of outcome Y-trajectories (latent growth profiles) expedite estimation of longitudinal prognostic scores (Hansen 2008). We focus on differences in Y- trajectories between treated and control patients with otherwise similar X-characteristics, rather than on shape differences among class profiles, and discuss an example where this can be useful.
|
