Validation and Application of the Medicare Bayesian Improved Surname Geocoding (MBISG) Version 2.0

• Activity Number: 443 - Adjustment for Social Risk Factors in Health Care Provider Performance Assessment
• Type: Invited
• Date/Time: Wednesday, August 2, 2017 : 8:30 AM to 10:20 AM
• Sponsor: Health Policy Statistics Section
• Author(s): Marc N Elliott* and Amelia M Haviland and Ann Haas and John L Adams and Jacob W Dembosky and Shondelle Wilson-Frederick and Joshua Mallet and Sarah Gaillot and Samuel C Haffer
• Companies: RAND Corporation and Carnegie Mellon University and RAND Corporation and Kaiser Permanente and RAND Corporation and CMS Office of Minority Health and RAND Corporation and Centers for Medicare & Medicaid Services and Centers for Medicare & Medicaid Services

Abstract

The Medicare Bayesian Improved Surname Geocoding (MBISG) method estimates a vector of six race/ethnicity probabilities (White, Black, Hispanic, Asian, AI/AN, and multiracial) for individuals based on surname, address, and an imperfect administrative race/ethnicity variable using Bayes' rule. Recent work substantially improved the performance of the method by (a) allowing the association of SS data with race/ethnicity to vary by age, (b) newly addressing compound surnames, (c) incorporating additional data elements, and (d) embedding the Bayesian core within a multinomial regression framework, resulting in MBISG version 2.0. Here we present the results of validation work using self-reported race/ethnicity from US Census data as a gold standard, including applications to estimating racial/ethnic disparities in quality of care.