Zombies, Immortals and the Which Hunt: NDI Record Linkage - It's More Than False Positives and False Negatives

• Activity Number: 251
• Type: Contributed
• Date/Time: Tuesday, August 13, 2002 : 2:00 PM to 3:50 PM
• Sponsor: Section on Government Statistics*
• Author(s): Bryan Sayer*+ and Christine Cox and Sandra Rothwell
• Affiliation(s): SAIC and National Center for Health Statistics and National Center for Health Statistics
• Keywords: Record Linkage ; Probabalistic Matching ; National Death Index

Abstract

The National Death Index (NDI) is the set of nearly all deaths registered in the U.S. since 1979, cataloged at the National Center for Health Statistics. Researchers submit study files to be matched against the NDI in order to determine both vital status and cause of death. The matching algorithm used is a two-step variation of the probabilistic approach developed by (Fellegi and Sunter, 1969) and (Rogot, Sorlie, and Johnson, 1986). For each submission record, the algorithm first selects a set of potential matches based on agreement on at least one of nine criteria. Then each possible match is scored on 11 items and summed (Horm, 1996). We test the accuracy of the matching algorithm using a sample of 12,700 subjects with a known vital status and a known death certificate number for decedents. We find that 94.4% of the decedents are correctly identified as deceased, and 99.75% of the non-decedents are correctly identified as alive. However, we find that a few of the decedents are not matched to the correct certificate, and we find that an additional 3% of the decedents have the correct certificate in their pool of possible matches, but the certificate does not score high enough.