Pacific AB
Handling silently missing data in Medicare Advantage encounter data (306600)
*Laura Anne Hatfield, Harvard Medical SchoolKeywords: missing data, weighting, claims
All studies must deal with missing data. In addition to obviously missing data, for which a flag indicates that a variable should have a value but that it is not present, administrative data are subject to “silent” missingness in both variables and records. In medical billing claims, variables are silently missing when diagnosis codes are never recorded; records are silently missing when a claim for a service does not appear in the analysis data set. A large literature exists on methods for imputing or otherwise adjusting for obviously missing variables, but less attention has been paid to silently missing data. This talk will discuss methods for assessing and addressing these forms of missingness in administrative data, including silently missing data. We apply our methods to the newly available Medicare Advantage (MA) encounter data that private Medicare plans have begun submitting to Centers for Medicare and Medicaid Services for casemix adjustment. We develop weighting methods to adjust for missing codes and records using external sources of information, including beneficiary summary files and utilization submitted as part of HEDIS quality measures.