Evaluating Missing Data Methods for Health Disparities Study Using HCUP State Inpatient Databases

• Activity Number: 377
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 11:15 AM
• Sponsor: Health Policy Statistics Section
• Author(s): Wei Zhang* and Andrew Gelman and Stephen Lyman and Yan Ma
• Companies: The George Washington University and Columbia University and Hospital for Special Surgery and The George Washington University
• Keywords: Missing data ; Multiple imputation ; Racial Disparities ; Total knee arthroplasty

Abstract

Racial disparities (RD) in healthcare outcomes in the U.S. have been identified in recent decades for total knee arthroplasty (TKA). We sought to study RD in TKA using the HCUP State Inpatients Databases (SID). As with any large scale data collection effort, the HCUP SID have a moderate amount of missing data (MD) in several variables. In particular,"patient race", a key indicator for health disparities research (HDR), has a high proportion of missingness. As a result, researchers often conduct inappropriate analysis leading to invalid inferences. The goal of this study was to compare five imputation methods (mean imputation, random draw, hot deck, joint multiple imputation[MI], conditional MI) for MD in the SID. For each method, we compared the distributions of imputed values to those of observed values. We also performed three regression analyses to assess racial disparities in hospital length of stay, in-hospital complications and utilization of high-volume hospitals among TKA patients. The results showed that conditional MI prediction was uniformly equivalent or superior to the best performing alternatives. More details of our research findings will be given in the talk.