|
Activity Number:
|
333
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 8, 2006 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Health Policy Statistics
|
| Abstract - #307120 |
|
Title:
|
Use of Risk-Adjustment Models To Predict Future High Medical Cost Cases: Is the Model Performance Sensitive to the Time Intervals in Claims Data?
|
|
Author(s):
|
Ya-Chen Tina Shih*+ and Lirong Zhao and Ying Xu
|
|
Companies:
|
M. D. Anderson Cancer Center and M. D. Anderson Cancer Center and M. D. Anderson Cancer Center
|
|
Address:
|
HSR Department of Biostatistics and Applied Math, Houston, TX, 77030,
|
|
Keywords:
|
risk-adjustment model ; prediction ; high cost cases ; claims data
|
|
Abstract:
|
Risk-adjustment models are used to identify persons that are likely to incur high medical costs in the future. They typically included last year's annual medical expenditures (expen) as a covariate; thus, their applications require at least having one full year of claims data while decision-makers often need feedback in a more timely fashion. We explored whether the performance of these models is sensitive to the time intervals of expen in claims. Using a 1% random sample from the Taiwanese National Health Insurance claims, we defined high cost cases as having expen at the top 1% of the distri. and employed logistic regressions to predict these cases. We compared the area under the curve from the ROC curves for various RA modes using claims at 3 time intervals -- 3, 6, 12 months, and concluded that models with shorter time intervals performed equally well as those with longer intervals.
|
