The Minimum Level of Sensitivity and Predictive Value Positive in the Evaluation of Health Outcomes with Administrative Data

• Activity Number: 477
• Type: Contributed
• Date/Time: Wednesday, August 1, 2007 : 2:00 PM to 3:50 PM
• Sponsor: Section on Health Policy Statistics
• Author(s): Yong-Fang Kuo*+ and Tracy Nguyen-Oghalai
• Companies: The University of Texas Medical Branch and The University of Texas Medical Branch
• Keywords: Sensitivity ; Predict Value Positive ; Measurement Error ; Claims

Abstract

Health administrative data is widely used to assess outcomes in population-based studies. To measure outcomes, claims based algorithms reaching sensitivity (Sen) of 90% and PVP of 70% were developed. To understand the impact of Sen and PVP on a bivariate association with dichotomous variables, a series of 100 simulations were performed for the outcome with combinations of variety levels of Sen and PVP, and different sample sizes and OR. Our results suggest that PVP of 70% can be useful in detecting OR of 1.5 with sample size of 1000. Increasing in OR or sample size increase bias and decrease coverage probability (CP), given the same PVP and Sen. PVP is more important than Sen in estimating association (by bias, mean square error, CP). The true association was always underestimated. Simulations can be used to evaluate outcome algorithms from claims, based on minimum levels of Sen and PVP.